2019 Book WearableRoboticsChallengesAndT PDF

Biosystems & Biorobotics
Maria Chiara Carrozza
Silvestro Micera
José L. Pons Editors
Wearable Robotics:
Challenges and
Trends
Proceedings of the 4th International
Symposium on Wearable Robotics,
WeRob2018, October 16–20, 2018,
Pisa, Italy
Volume 22
Series editor
Eugenio Guglielmelli, Campus Bio-Medico University of Rome, Rome, Italy
e-mail: e.guglielmelli@unicampus.it
Editorial Board
Dino Accoto, Campus Bio-Medico University of Rome, Rome, Italy
Sunil Agrawal, University of Delaware, Newark, DE, USA
Fabio Babiloni, Sapienza University of Rome, Rome, Italy
Jose M. Carmena, University of California, Berkeley, CA, USA
Maria Chiara Carrozza, Scuola Superiore Sant’Anna, Pisa, Italy
Paolo Dario, Scuola Superiore Sant’Anna, Pisa, Italy
Arturo Forner-Cordero, University of Sao Paolo, São Paulo, Brazil
Masakatsu G. Fujie, Waseda University, Tokyo, Japan
Nicolas Garcia, Miguel Hernández University of Elche, Elche, Spain
Neville Hogan, Massachusetts Institute of Technology, Cambridge, MA, USA
Hermano Igo Krebs, Massachusetts Institute of Technology, Cambridge, MA, USA
Dirk Lefeber, Universiteit Brussel, Brussels, Belgium
Rui Loureiro, Middlesex University, London, UK
Marko Munih, University of Ljubljana, Ljubljana, Slovenia
Paolo M. Rossini, University Cattolica del Sacro Cuore, Rome, Italy
Atsuo Takanishi, Waseda University, Tokyo, Japan
Russell H. Taylor, The Johns Hopkins University, Baltimore, MA, USA
David A. Weitz, Harvard University, Cambridge, MA, USA
Loredana Zollo, Campus Bio-Medico University of Rome, Rome, Italy
Aims & Scope
Biosystems & Biorobotics publishes the latest research developments in three main areas:
1) understanding biological systems from a bioengineering point of view, i.e. the study of
biosystems by exploiting engineering methods and tools to unveil their functioning principles
and unrivalled performance; 2) design and development of biologically inspired machines
and systems to be used for different purposes and in a variety of application contexts. The
series welcomes contributions on novel design approaches, methods and tools as well as case
studies on specific bioinspired systems; 3) design and developments of nano-, micro-,
macrodevices and systems for biomedical applications, i.e. technologies that can improve
modern healthcare and welfare by enabling novel solutions for prevention, diagnosis,
surgery, prosthetics, rehabilitation and independent living.
On one side, the series focuses on recent methods and technologies which allow multiscale,
multi-physics, high-resolution analysis and modeling of biological systems. A special
emphasis on this side is given to the use of mechatronic and robotic systems as a tool for basic
research in biology. On the other side, the series authoritatively reports on current theoretical
and experimental challenges and developments related to the “biomechatronic” design of novel
biorobotic machines. A special emphasis on this side is given to human-machine interaction
and interfacing, and also to the ethical and social implications of this emerging research area, as
key challenges for the acceptability and sustainability of biorobotics technology.
The main target of the series are engineers interested in biology and medicine, and
specifically bioengineers and bioroboticists. Volume published in the series comprise
monographs, edited volumes, lecture notes, as well as selected conference proceedings and
PhD theses. The series also publishes books purposely devoted to support education in
bioengineering, biomedical engineering, biomechatronics and biorobotics at graduate and
post-graduate levels.
About the Cover
The cover of the book series Biosystems & Biorobotics features a robotic hand prosthesis.
This looks like a natural hand and is ready to be implanted on a human amputee to help them
recover their physical capabilities. This picture was chosen to represent a variety of concepts
and disciplines: from the understanding of biological systems to biomechatronics,
bioinspiration and biomimetics; and from the concept of human-robot and human-machine
interaction to the use of robots and, more generally, of engineering techniques for biological
research and in healthcare. The picture also points to the social impact of bioengineering
research and to its potential for improving human health and the quality of life of all
individuals, including those with special needs. The picture was taken during the
LIFEHAND experimental trials run at Università Campus Bio-Medico of Rome (Italy) in
2008. The LIFEHAND project tested the ability of an amputee patient to control the
Cyberhand, a robotic prosthesis developed at Scuola Superiore Sant’Anna in Pisa (Italy),
using the tf-LIFE electrodes developed at the Fraunhofer Institute for Biomedical
Engineering (IBMT, Germany), which were implanted in the patient’s arm. The implanted
tf-LIFE electrodes were shown to enable bidirectional communication (from brain to hand
and vice versa) between the brain and the Cyberhand. As a result, the patient was able to
control complex movements of the prosthesis, while receiving sensory feedback in the form
of direct neurostimulation. For more information please visit http://www.biorobotics.it or
contact the Series Editor.
More information about this series at http://www.springer.com/series/10421

Maria Chiara Carrozza Silvestro Micera
•
José L. Pons
Editors
Wearable Robotics:
Challenges and Trends
Proceedings of the 4th International
Symposium on Wearable Robotics, WeRob2018,
October 16–20, 2018, Pisa, Italy
123
Editors
Maria Chiara Carrozza José L. Pons
Scuola Superiore Sant’Anna Spanish National Research Council
The BioRobotics Institute Cajal Institute
Pisa, Italy Madrid, Spain
Silvestro Micera
Scuola Superiore Sant’Anna, Translational
Neural Engineering Area
The BioRobotics Institute
Pisa, Italy
ISSN 2195-3562 ISSN 2195-3570 (electronic)

ISBN 978-3-030-01886-3 ISBN 978-3-030-01887-0 (eBook)
https://doi.org/10.1007/978-3-030-01887-0
Library of Congress Control Number: 2018956723
© Springer Nature Switzerland AG 2019

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Contents
Wearable Sensors for Robotic Exoskeletons

Position Sensing and Control with FMG Sensors for Exoskeleton
Physical Assistance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Muhammad R. U. Islam, Kun Xu, and Shaoping Bai
Force Localization Estimation Using a Designed Soft Tactile Sensor . . . 8
Merve Acer and Adnan Furkan Yıldız
EIT-Based Tactile Sensing Patches for Rehabilitation and Human
Machine Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
Stefania Russo, Nicola Carbonaro, and Alessandro Tognetti
Synthesis and Optimization Considerations for a Knee Orthosis Based
on a Watt’s Six-Bar Linkage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
Evagoras Xydas, Banu Abdikadirova, and Kostas Konstantinos
Wearable Sensory Apparatus Performance While Using Inertial
Measurement Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Grega Logar, Zoran Ivanic, and Marko Munih
WeFiTS: Wearable Fingertip Tactile Sensor . . . . . . . . . . . . . . . . . . . . . 28
Elif Hocaoglu
Soft Wearable Robots

Characterisation of Pressure Distribution at the Interface of a Soft
Exosuit: Towards a More Comfortable Wear . . . . . . . . . . . . . . . . . . . . 35
Michele Xiloyannis, Domenico Chiaradia, Antonio Frisoli,
and Lorenzo Masia
Realizing Soft High Torque Actuators for Complete Assistance
Wearable Robots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Allan J. Veale, Kyrian Staman, and Herman van der Kooij
v
vi Contents
Application of a User-Centered Design Approach to the Development

of XoSoft – A Lower Body Soft Exoskeleton . . . . . . . . . . . . . . . . . . . . . 44
Valerie Power, Adam de Eyto, Bernard Hartigan, Jesús Ortiz,
and Leonard W. O’Sullivan
Preliminary Experimental Study on Variable Stiffness Structures
Based on Textile Jamming for Wearable Robotics . . . . . . . . . . . . . . . . . 49
Ali Sadeghi, Alessio Mondini, and Barbara Mazzolai
Towards Embroidered Sensing Technologies for a Lower Limb
Soft Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
M. Totaro, E. Bottenberg, R. Groeneveld, L. Erkens, A. Mondini,
G. J. Brinks, and L. Beccai
Recent Results from Evaluation of Soft Wearable Robots
in Clinical Populations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
Conor Walsh
Subject-Centered Based Approaches for Controlling Wearable Robots

Toward an Affordable Multi-Modal Motion Capture System
Framework for Human Kinematics and Kinetics Assessment . . . . . . . . 65
Randa Mallat, Vincent Bonnet, Mohamad Khalil, and Samer Mohammed
High Power Series Elastic Actuator Development
for Torque-Controlled Exoskeletons . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
Mehmet C. Yildirim, Ahmet Talha Kansizoglu, Polat Sendur,
and Barkan Ugurlu
Investigation on Variable Impedance Control
for Modulating Assistance in Walking Strategies
with the AUTONOMYO Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
A. Ortlieb, P. Lichard, F. Dzeladini, R. Baud, H. Bleuler, A. Ijspeert,
and M. Bouri
Improving Usability of Rehabilitation Robots: Hand Module
Evaluation of the ARMin Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . 80
Fabian Just, Daniel Gunz, Jaime Duarte, Davide Simonetti, Robert Riener,
and Georg Rauter
Lower Limb Exoskeletons, from Specifications to Design . . . . . . . . . . . . 85
M. Bouri
Contents vii
Robotic and Neuroprosthetic Balance Management Approaches

for Walking Assistance
Novel Perturbation-Based Approaches Using Pelvis Exoskeleton
Robot in Gait and Balance Training After Stroke . . . . . . . . . . . . . . . . . 91
Zlatko Matjačić, Matjaž Zadravec, Nataša Bizovičar, Nika Goljar,
and Andrej Olenšek
Balance During Bodyweight Supported
and Robot-Assisted Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
Eva Swinnen, Jean-Pierre Baeyens, Nina Lefeber,
Emma De Keersmaecker, Stieven Henderix, Marc Michielsen,
and Eric Kerckhofs
Maintaining Gait Balance After Perturbations to the Leg: Kinematic
and Electromyographic Patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
Eleonora Croci, Roger Gassert, and Camila Shirota
A New Sensory Feedback System for Lower-Limb Amputees:
Assessment of Discrete Vibrotactile Stimuli Perception
During Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
Mariangela Filosa, Ilaria Cesini, Elena Martini, Giacomo Spigler,
Nicola Vitiello, Calogero Oddo, and Simona Crea
Fast Online Decoding of Motor Tasks with Single sEMG Electrode
in Lower Limb Amputees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Federica Barberi, Federica Aprigliano, Emanuele Gruppioni,
Angelo Davalli, Rinaldo Sacchetti, Alberto Mazzoni,
and Silvestro Micera
A Wearable Haptic Feedback System for Assisting Lower-Limb
Amputees in Multiple Locomotion Tasks . . . . . . . . . . . . . . . . . . . . . . . . 115
Ilaria Cesini, Giacomo Spigler, Sahana Prasanna, Domitilla Taxis,
Filippo Dell’Agnello, Elena Martini, Simona Crea, Nicola Vitiello,
Alberto Mazzoni, and Calogero Maria Oddo
Benchmarking Wearable Robots

COVR – Towards Simplified Evaluation and Validation
of Collaborative Robotics Applications Across a Wide Range
of Domains Based on Robot Safety Skills . . . . . . . . . . . . . . . . . . . . . . . . 123
Jule Bessler, Leendert Schaake, Catherine Bidard, Jaap H. Buurke,
Aske E. B. Lassen, Kurt Nielsen, José Saenz, and Federico Vicentini
Monitoring Upper Limbs During Exoskeleton-Assisted
Gait Outdoors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Matteo Lancini, Simone Pasinetti, Valeria Montini,
and Giovanna Sansoni
viii Contents
What Do People Expect from Benchmarking of Bipedal Robots?

Preliminary Results of the EUROBENCH Survey . . . . . . . . . . . . . . . . . 132
R. Conti, F. Giovacchini, L. Saccares, N. Vitiello, J. L. Pons,
and D. Torricelli
Modeling Human-Exoskeleton Interaction: Preliminary Results . . . . . . 137
M. C. Sánchez-Villamañán, D. Torricelli, and J. L. Pons
Human-in-the-Loop Bayesian Optimization of a Tethered Soft Exosuit
for Assisting Hip Extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
Myunghee Kim, Ye Ding, Charles Liu, Jinsoo Kim, Sangjun Lee,
Nikolaos Karavas, Conor Walsh, and Scott Kuindersma
A Review of Performance Metrics for Lower Limb Wearable Robots:
Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
D. Torricelli, D. Pinto-Fernandez, R. Conti, N. Vitiello, and J. L. Pons
Flexible and Transparent Technologies for Innovative

Wearable Robotics
Development of Polymer Optical Fiber Sensors for Lower Limb
Exoskeletons Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
Arnaldo G. Leal-Junior, Anselmo Frizera, Carlos Marques,
and Maria José Pontes
T-FLEX: Variable Stiffness Ankle-Foot Orthosis for Gait Assistance . . . 160
Miguel Manchola, Daya Serrano, Daniel Gómez, Felipe Ballen,
Diego Casas, Marcela Munera, and Carlos A. Cifuentes
A Series Elastic Dual-Motor Actuator Concept
for Wearable Robotics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Tom Verstraten, Raphaël Furnémont, Pablo López-García, Stein Crispel,
Bram Vanderborght, and Dirk Lefeber
Towards Design Guidelines for Physical Interfaces on Industrial
Exoskeletons: Overview on Evaluation Metrics . . . . . . . . . . . . . . . . . . . 170
M. Sposito, S. Toxiri, D. G. Caldwell, J. Ortiz, and E. De Momi
Design and Control of a Transparent Lower Limb Exoskeleton . . . . . . 175
Wilian M. dos Santos and Adriano A. G. Siqueira
Development and Testing of Full-Body Exoskeleton AXO-SUIT
for Physical Assistance of the Elderly . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
S. Bai, S. Christensen, M. Islam, S. Rafique, N. Masud, P. Mattsson,
L. O’Sullivan, and V. Power
Contents ix
Wearable Robotics for Rehabilitation and Assistance

in Latin America
Artificial Vision Algorithm for Object Manipulation with a Robotic
Arm in a Semi-Autonomous Brain-Computer Interface . . . . . . . . . . . . . 187
M. A. Ramírez-Moreno, S. M. Orozco-Soto, J. M. Ibarra-Zannatha,
and D. Gutiérrez
Design Specifications and Usability Issues Considered
in the User Centered Design of a Wearable Exoskeleton
for Upper Limb of Children with Spastic Cerebral Palsy . . . . . . . . . . . 192
Alberto I. Perez-Sanpablo, Catherine Disselhorst-Klug,
Juan M. Ibarra Zannatha, Josefina Gutierrez-Martínez,
Alicia Meneses Peñaloza, Elisa Romero-Avila, and Santos M. Orozco-Soto
Stance Control with the Active Knee Orthosis ALLOR for Post-Stroke
Patients During Walking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
A. C. Villa-Parra, J. Lima, D. Delisle-Rodriguez, A. Frizera-Neto,
and T. Bastos
Gait Phase Detection for Lower Limb Prosthetic Devices . . . . . . . . . . . 201
Pablo E. Caicedo, Carlos F. Rengifo, Luís E. Rodríguez,
and Wilson A. Sierra
Lower Limb Exoskeletons in Latin-America . . . . . . . . . . . . . . . . . . . . . 206
Antonio J. del-Ama, Jose M. Azorín, José L. Pons, Anselmo Frizera,
Thomaz Rodrigues, Ángel Gil-Agudo, Javier O. Roa, and Juan C. Moreno
Development of a Visual-Inertial Motion Tracking System
for Muscular-Effort/Angular Joint-Position Relation to Obtain
a Quantifiable Variable of Spasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
S. M. Orozco-Soto, A. I. Pérez-Sanpablo, P. Vera-Bustamante,
and J. M. Ibarra-Zannatha
Wearable Robotic Solutions for Factories of the Future

Towards Standard Specifications for Back-Support Exoskeletons . . . . . 219
Stefano Toxiri, Matteo Sposito, Maria Lazzaroni, Lorenza Mancini,
Massimo Di Pardo, Darwin G. Caldwell, and Jesús Ortiz
Lift Movement Detection with a QDA Classifier
for an Active Hip Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Baojun Chen, Lorenzo Grazi, Francesco Lanotte, Nicola Vitiello,
and Simona Crea
The Effect of a Passive Trunk Exoskeleton on Functional Performance
and Metabolic Costs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
S. J. Baltrusch, J. H. van Dieën, S. M. Bruijn, A. S. Koopman,
C. A. M. van Bennekom, and H. Houdijk
x Contents
Industrial Wearable Exoskeletons and Exosuits Assessment Process . . . 234

Jawad Masood, Angel Dacal-Nieto, Víctor Alonso-Ramos,
M. Isabel Fontano, Anthony Voilqué, and Julia Bou
Trunk Range of Motion in the Sagittal Plane with and Without
a Flexible Back Support Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
Matthias B. Näf, Axel S. Koopman, Carlos Rodriguez-Guerrero,
Real-Time Control of Quasi-Active Hip Exoskeleton
Based on Gaussian Mixture Model Approach . . . . . . . . . . . . . . . . . . . . 244
Mišel Cevzar, Tadej Petrič, Marko Jamšek, and Jan Babič
Optimizing Design Characteristics of Passive and Active Spinal
Exoskeletons for Challenging Work Tasks . . . . . . . . . . . . . . . . . . . . . . . 249
Monika Harant, Manish Sreenivasa, Matthew Millard, Nejc Šarabon,
and Katja Mombaur
Human Modeling and Simulation for Neurorehabilitation Engineering

Calibration and Validation of a Skeletal Multibody Model
for Leg-Orthosis Contact Force Estimation . . . . . . . . . . . . . . . . . . . . . . 257
Francisco Mouzo, Urbano Lugris, Javier Cuadrado,
Josep M. Font-Llagunes, and Francisco J. Alonso
A Continuous and Differentiable Mechanical Model of Muscle
Force and Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
Matthew Millard, David Franklin, and Walter Herzog
SimCP: A Simulation Platform to Predict Gait Performance
Following Orthopedic Intervention in Children with Cerebral Palsy . . . 267
Friedl De Groote, Lorenzo Pitto, Hans Kainz, Antoine Falisse,
Eirini Papageorgiou, Mariska Wesseling, Sam Van Rossom,
Kaat Desloovere, and Ilse Jonkers
Bio-inspired Walking: From Humanoids to Assistive Devices . . . . . . . . 271
Renaud Ronsse
Design of a Hand Exoskeleton for Use with Upper
Limb Exoskeletons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
Peter Walker Ferguson, Brando Dimapasoc, Yang Shen, and Jacob Rosen
A Real-time Graphic Interface for the Monitoring of the Human
Joint Overloadings with Application to Assistive Exoskeletons . . . . . . . . 281
Marta Lorenzini, Wansoo Kim, Elena De Momi, and Arash Ajoudani
Contents xi
Smart Human-Machine Systems for Lower-Limb Assistance

and Rehabilitation After Paralysis
Study of Algorithms Classifiers for an Offline BMI Based
on Motor Imagery of Pedaling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Mario Ortiz, Marisol Rodríguez-Ugarte, Eduardo Iáñez,
and José M. Azorín
Exoskeleton Control for Post-stoke Gait Training of a Paretic Limb
Based on Extraction of the Contralateral Gait Phase . . . . . . . . . . . . . . . 294
Gabriel Aguirre-Ollinger, Ashwin Narayan, Hsiao-Ju Cheng,
and Haoyong Yu
Design of a Passive Exoskeleton to Support Sit-to-Stand Movement:
A 2D Model for the Dynamic Analysis of Motion . . . . . . . . . . . . . . . . . 299
Luís P. Quinto, Sérgio B. Gonçalves, and Miguel T. Silva
Walking Assistance of Subjects with Spinal Cord Injury with an Ankle
Exoskeleton and Neuromuscular Controller . . . . . . . . . . . . . . . . . . . . . . 304
M. Arquilla, I. Pisotta, F. Tamburella, N. L. Tagliamonte, M. Masciullo,
A. R. Wu, C. Meijneke, H. van der Kooij, A. J. Ijspeert, and M. Molinari
Center of Mass and Postural Adaptations During Robotic
Exoskeleton-Assisted Walking for Individuals
with Spinal Cord Injury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Arvind Ramanujam, Kamyar Momeni, Syed R. Husain,
Jonathan Augustine, Erica Garbarini, Peter Barrance, Ann M. Spungen,
Pierre K. Asselin, Steven Knezevic, and Gail F. Forrest
Exoskeleton Controller and Design Considerations: Effect on Training
Response for Persons with SCI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
Gail F. Forrest, Arvind Ramanujam, Ann M. Spungen,
Christopher Cirnigliaro, Kamyar Momeni, Syed R. Husain,
Jonathan Augustine, Erica Garbarini, Pierre K. Asselin,
and Steven Knezevic
Biorobotics Approaches to Understand and Restore Human Balance

Integrating Posture Control in Assistive Robotic Devices
to Support Standing Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
T. Mergner and V. Lippi
A Computational Framework for Muscle-Level Control
of Bi-lateral Robotic Ankle Exoskeletons . . . . . . . . . . . . . . . . . . . . . . . . 325
Guillaume Durandau, Herman van der Kooij, and Massimo Sartori
A Conductive Fabric Based Smart Insole to Measure the Foot
Pressure Distribution with High Resolution . . . . . . . . . . . . . . . . . . . . . . 329
Xinyao Hu, Chuang Luo, Dongsheng Peng, and Xingda Qu
xii Contents
Training Balance Recovery in People with Incomplete SCI Wearing

a Wearable Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
E. H. F. van Asseldonk, A. Emmens, T. J. H. Brug, I. Pisotta, M. Arquilla,
F. Tamburella, M. Masciullo, N. L. Tagliamonte, R. Valette, M. Molinari,
and H. van der Kooij
Modular Composition of Human Gaits Through Locomotor
Subfunctions and Sensor-Motor-Maps . . . . . . . . . . . . . . . . . . . . . . . . . . 339
Andre Seyfarth, Maziar A. Sharbafi, Guoping Zhao,
and Christian Schumacher
Model-Based Posture Control for a Torque-Controlled
Humanoid Robot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
Maximo A. Roa, Bernd Henze, and Christian Ott
Exoskeleton Research in Europe

XoSoft - Iterative Design of a Modular Soft Lower
Limb Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
Jesús Ortiz, Christian Di Natali, and Darwin G. Caldwell
Preliminary Usability and Efficacy Tests in Neurological Patients
of an Exoskeleton for Upper-Limb Weight Support . . . . . . . . . . . . . . . . 356
M. Caimmi, I. Carpinella, R. Di Giovanni, D. Ellena, L. Molinari Tosatti,
D. Cattaneo, M. Ferrarin, and C. Solaro
Symbitron: Symbiotic Man-Machine Interactions in Wearable
Exoskeletons to Enhance Mobility for Paraplegics . . . . . . . . . . . . . . . . . 361
Herman van der Kooij, Edwin van Asseldonk, Gijs van Oort,
Victor Sluiter, Amber Emmens, Heide Witteveen,
Nevio Luigi Tagliamonte, Federica Tamburella, Iolanda Pisotta,
Marcella Masciullo, Matteo Arquilla, Marco Molinari, Amy Wu,
Auke Ijspeert, Florin Florin Dzeladini, Freygardur Thorsteinsson,
Arash Arami, Etienne Burdet, Hsien-Yung Huang, Wouter Gregoor,
and Cor Meijneke
Beyond Robo-Mate: Towards the Next Generation of Industrial
Exoskeletons in Europe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Jesús Ortiz, Stefano Toxiri, and Darwin G. Caldwell
The SoftPro Project: Synergy-Based Open-Source Technologies
for Prosthetics and Rehabilitation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
Cristina Piazza, Manuel G. Catalano, Matteo Bianchi, Emiliano Ricciardi,
Domenico Prattichizzo, Sami Haddadin, Andreas R. Luft,
Olivier Lambercy, Roger Gassert, Eike Jakubowitz,
Herman Van Der Kooij, Frederick Tonis, Fabio Bonomo,
Benjamin de Jonge, Tomas Ward, Kristin D. Zhao, Marco Santello,
and Antonio Bicchi
Contents xiii
EUROBENCH: Preparing Robots for the Real World . . . . . . . . . . . . . . 375

D. Torricelli and J. L. Pons
Poster Session
Actuation Requirements for Assistive Exoskeletons: Exploiting
Knowledge of Task Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
Stefano Toxiri, Andrea Calanca, Tommaso Poliero, Darwin G. Caldwell,
and Jesús Ortiz
Grasping Detection with Force Sensor Embedded
in a Hand Exoskeleton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 386
Jorge A. Díez, José M. Catalán, Andrea Blanco, Juan Barios,
Santiago Ezquerro, Arturo Bertomeu-Motos, and Nicolás García-Aracil
XoSoft Connected Monitor (XCM) Unsupervised Monitoring
and Feedback in Soft Exoskeletons of 3D Kinematics, Kinetics,
Behavioral Context and Control System Status . . . . . . . . . . . . . . . . . . . 391
Chris T. M. Baten, Wiebe de Vries, Leendert Schaake, Juryt Witteveen,
Daniel Scherly, Konrad Stadler, Andres Hidalgo Sanchez, Eduardo Rocon,
Danny Plass-Oude Bos, and Jeroen Linssen
Tactile and Proximity Servoing by a Multi-modal Sensory
Soft Hand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
John Nassour and Fred H. Hamker
Improved Fabrication of Soft Robotic Pad for Wearable
Assistive Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
Yi Sun, Aaron Jing Yuan Goh, Miao Li, Hui Feng, Jin Huat Low,
Marcelo H. Ang Jr., and Raye Chen Hua Yeow
The Exosleeve: A Soft Robotic Exoskeleton for Assisting
in Activities of Daily Living . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
Rainier F. Natividad, Sin Wai Hong, Tiana M. Miller-Jackson,
and Chen-Hua Yeow
Exoskeleton with Soft Actuation and Haptic Interface . . . . . . . . . . . . . . 410
Ivanka Veneva, Dimitar Chakarov, Michail Tsveov, and Pavel Venev
Comparison of a Soft Exosuit and a Rigid Exoskeleton
in an Assistive Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
Domenico Chiaradia, Michele Xiloyannis, Massimiliano Solazzi,
Lorenzo Masia, and Antonio Frisoli
Design of Soft Exosuit for Elbow Assistance Using Butyl Rubber
Tubes and Textile . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 420
John Nassour, Sidhdharthkumar Vaghani, and Fred H. Hamker
xiv Contents
Optimizing Body Thickness of Watchband-Type Soft Pneumatic

Actuator for Feedback of Prosthesis Grasping Force . . . . . . . . . . . . . . . 425
Masashi Sekine, Kazuya Kawamura, and Wenwei Yu
The Effect of Negative Damping at the Hip Joint During Level
Walking: A Preliminary Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 430
Jongwon Lee, Juwhan Bae, Chilyong Kwon, Seokjin Hwang,
and Gyoosuk Kim
Overview and Challenges for Controlling
Back-Support Exoskeletons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435
Maria Lazzaroni, Stefano Toxiri, Darwin G. Caldwell, Elena De Momi,
and Jesús Ortiz
Assessment of a Hand Exoskeleton Control Strategy Based on User’s
Intentions Classification Starting from Surface EMG Signals . . . . . . . . 440
Nicola Secciani, Matteo Bianchi, Alessandro Ridolfi, Federica Vannetti,
and Benedetto Allotta
Contribution of a Knee Orthosis to Walking . . . . . . . . . . . . . . . . . . . . . 445
O. Bordron, C. Huneau, É. Le Carpentier, and Y. Aoustin
Human Trunk Stabilization with Hip Exoskeleton
for Enhanced Postural Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 450
Marko Jamšek and Jan Babič
Development of a Wearable Haptic Feedback System
for Limb Movement Symmetry Training . . . . . . . . . . . . . . . . . . . . . . . . 455
Amre Eizad, Muhammad Raheel Afzal, Hosu Lee, Sung-Ki Lyu,
and Jungwon Yoon
Failure Mode and Effect Analysis (FMEA)-Driven Design
of a Planetary Gearbox for Active Wearable Robotics . . . . . . . . . . . . . . 460
Pablo López García, Stein Crispel, Tom Verstraten, Elias Saerens,
Bryan Convens, Bram Vanderborght, and Dirk Lefeber
Introducing Series Elastic Links . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
Andrea Calanca, Luca Bettinelli, Eldison Dimo, Rudy Vicario,
Mauro Serpelloni, and Paolo Fiorini
Polymer Optical Fiber Sensors Approaches
for Insole Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 470
Arnaldo G. Leal-Junior, Antreas Theodosiou, Anselmo Frizera,
Maria F. Domingues, Cátia Leitão, Kyriacos Kalli, Paulo André,
Paulo Antunes, Maria José Pontes, and Carlos Marques
Contents xv
Pushing the Limits: A Novel Tape Spring Pushing Mechanism

to be Used in a Hand Orthosis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
Claudia J. W. Haarman, Edsko E. G. Hekman, Hans S. Rietman,
and Herman van der Kooij
Design and Preliminary Validation of a Smart Personal
Flotation Device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
Julian Fraize, Mirjam Furth, and Damiano Zanotto
Introducing Compound Planetary Gears (C-PGTs): A Compact
Way to Achieve High Gear Ratios for Wearable Robots . . . . . . . . . . . . 485
Stein Crispel, Pablo López García, Tom Verstraten, Bryan Convens,
Elias Saerens, Bram Vanderborght, and Dirk Lefeber
Model-Based Approach in Developing a Hand Exoskeleton
for Children: A Preliminary Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
Matteo Bianchi, Nicola Secciani, Alessandro Ridolfi, Federica Vannetti,
and Guido Pasquini
Design of Bio-joint Shaped Knee Exoskeleton Assisting
for Walking and Sit-to-Stance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
Mehmet F. Kapci and Ramazan Unal
ANT-M: Design of Passive Lower-Limb Exoskeleton
for Weight-Bearing Assistance in Industry . . . . . . . . . . . . . . . . . . . . . . . 500
Berkay Guncan and Ramazan Unal
Effects of an Inclination-Controlled Active Spinal Exoskeleton
on Spinal Compression Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 505
A. S. Koopman, S. Toxiri, M. P. de Looze, I. Kingma, and J. H. van Dieën
Novel Mechanism of Upper Limb Exoskeleton for Weight Support . . . . 510
Daegeun Park, Jesus Ortiz, and Darwin G. Caldwell
Human-Centered Design of an Upper-Limb Exoskeleton
for Tedious Maintenance Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
Andrea Blanco, Jorge A. Díez, David López, José V. García,
José M. Catalán, and Nicolás García-Aracil
A Supernumerary Soft Robotic Hand-Arm System
for Improving Worker Ergonomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520
Andrea S. Ciullo, Manuel G. Catalano, Antonio Bicchi,
and Arash Ajoudani
An Optimization Approach to Design Control Strategies
for Soft Wearable Passive Exoskeletons . . . . . . . . . . . . . . . . . . . . . . . . . 525
Andres F. Hidalgo Romero, Eveline Graf, and Eduardo Rocon
xvi Contents
Actuator Optimization for a Back-Support Exoskeleton:

The Influence of the Objective Function . . . . . . . . . . . . . . . . . . . . . . . . . 530
Tommaso Poliero, Stefano Toxiri, Darwin G. Caldwell, and Jesús Ortiz
Design of MobIle Digit Assistive System (MIDAS): A Passive Hand
Extension Exoskeleton for Post Stroke Rehabilitation . . . . . . . . . . . . . . 535
Titus S. Hansen, Chris K. Bitikofer, Bahram E. Sobbi, and Joel C. Perry
Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 541
Wearable Sensors for Robotic
Exoskeletons
Position Sensing and Control with FMG
Sensors for Exoskeleton Physical Assistance
Muhammad R. U. Islam1(&), Kun Xu2, and Shaoping Bai1

1
Department of Materials and Production,
Aalborg University, Aalborg, Denmark
{mraza,shb}@mp.aau.dk
2
Robotics Institute, Beihang University, Beijing, China
xk007@buaa.edu.cn
Abstract. Human intention decoding is a primary requirement to control an

exoskeleton. In this work, a new method of decoding human intention by
Forcemyography (FMG) is explored to estimate elbow joint angle during arm
motion. The method utilizes an FSR-based sensor band to read muscle con-
traction and relaxation. The readings of the sensor band are mapped to the
desired joint angle by using coarse Gaussian support vector machine
(SVM) regression algorithm. The estimated joint angle is further used to control
an elbow joint exoskeleton. Results show that the new method is able to estimate
reliably the joint angle for controlling the exoskeleton.
1 Introduction
Robotic exoskeletons have strong potential of use as assistive devices to assist the
elderly and workers [1] and therefore effective methods of estimating human intention
are needed for the assistance control. In the context of cognitive human robot inter-
action several solutions have been proposed using EEG [2], EMG [3] and FMG [4].
Among these methods, FMG has gained more interest due to its non-invasive nature
and simple mechanical and electronic interface, and a good performance as well [5].
Intention detection by FMG is mostly implemented by FSR sensors. The sensors
can be embedded inside a strap to detect muscle contraction and relaxation and
interpret different movements. In the reported works, FMG has been used to classify
forearm [4, 6] and ankle muscle activities [7], but not the upper arm muscle activities.
In this work, FMG is used to estimate the elbow joint rotation angle, which is
achieved by detecting muscle activity of upper arm muscles and using SVM regression
algorithm to interpret the readings. The developed method is able to detect small
changes in joint angle, hence, increasing the reachable space. Moreover, the FMG
sensor can also measure simultaneously the muscle strength/effort, which is not pos-
sible in other sensors like accelerometers and rotational sensors. In our previous work,
the determination of motion type has been reported in [6]. In this paper, we have
focused on the joint angle estimation and its application in the control of the
exoskeleton motion.

M. C. Carrozza et al. (Eds.): WeRob 2018, BIOSYSROB 22, pp. 3–7, 2019.
https://doi.org/10.1007/978-3-030-01887-0_1
4 M. R. U. Islam et al.
2 FMG Sensor
It is known that the elbow joint motion is primarily governed by biceps and triceps. The
contraction and relaxation of muscles cause the muscle shape and hardness to change.
It is observed through experiments that due to muscle shape change, the perimeter of
upper arm at the middle point increases during flexion and decreases during extension.
Moreover, the change is associated to joint angle.
The change in muscle volume can be detected through an FSR based sensor band
(S-Band). The S-Band is designed and constructed using an array of three FSRs placed
inside a flexible strap and worn on the upper arm as shown in Fig. 1a. As the muscle
contracts, the shape and hardness change of muscle will cause an outward normal force
on S-Band. This change in normal force can be registered in relation to the arm bending
angle. Therefore, with further post-processing, the force read by S-Band can be inferred
in terms of joint angle.
Fig. 1. (a) S-Band design and placement (b) placement of E-EXO on human arm
3 Elbow Exoskeleton Design
An elbow joint exoskeleton (E-EXO) is developed as shown in Fig. 1b. The E-EXO
has a range of motion of 0o–130o. The motor has its built-in Hall sensors and an
incremental encoder. An absolute encoder is integrated separately to get the actual joint
angle of E-EXO.
4 SVM Implementation
The forces read by S-Band are interpreted as joint angle using SVM regression algo-
rithm with 5-fold cross-validation framework. The details on hardware setup to collect
data for training session, protocol followed for collecting data and testing are given in
forthcoming sections.
4.1 Hardware Setup

The hardware is comprised of S-Band and an accelerometer ADXL-335. This
accelerometer is able to measure accelerations (±3g) in three axes. By wearing the
accelerometer on the wrist, it is calibrated to provide the elbow joint angle.
Position Sensing and Control with FMG Sensors for Exoskeleton Physical Assistance 5
All the data including FSR’s force readings and acceleration read through Arduino
Due is transmitted to the MATLAB based GUI through serial port.
During experiments, subjects were instructed to keep shoulder and wrist in neutral
position. Training data was collected by keeping static pose of forearm at five different
joint angles for once and each position was maintained for 5s. Subject started from a
joint angle near 0° and ended up near 110°, while the actual value of joint angle was
computed through the accelerometer.
4.2 Real-Time Estimation

In the real-time joint angle estimation, subject performed two tasks i.e. keeping static
pose at two random elbow joint angles and flexion of arm from neutral position.
5 Experiments and Results
5.1 Joint Angle Estimation

Four subjects were recruited for the experiments of real-time estimation of joint angle
with the developed method. The results are provided in Table 1.
Table 1. Error calculated between actual and estimated joint angle

Participants/Tasks Holding position Flexion
Position 1 Position 2
Subject 1 9.8o 7.12o 19.53o
o o
Subject 2 2.59 6.52 3.84o
o o
Subject 3 4.32 6.81 6.51o
o o
Subject 4 7.23 6.76 10.2o
A maximum of 9.8o mean error was recorded for holding position, which is not too
significant, as the absolute precision is not required. In the task of flexion, it can be seen
that measurements on three out of four subjects have shown error within 10o. The
measurement with one subject showed a high error of 19.8o. This can be resulted
because the model was trained for constant positions, not for dynamic movements. In
addition, the muscle contraction profiles are different for static pose and dynamic
movements. Figure 2 shows the results for the joint angle estimation tasks performed
by Subject 4.
5.2 E-EXO Control

A bilateral rehabilitation exercise was performed to control the E-EXO in real-time.
S-Band was worn on right arm and E-EXO was worn by the other person on same arm.
The desired joint angle estimated in MATLAB was transmitted back to Arduino Due
control unit, where the control system generates the corresponding desired joint
6 M. R. U. Islam et al.
Fig. 2. Joint angle estimation
velocity signal, which is controlled through ESCON’s (motor driver) built-in PI

velocity controller. The control loop, shown in Fig. 3, is run on 50 Hz frequency. The
position tracking result is shown in Fig. 4. It can be seen in Fig. 4 that the exoskeleton
is able to track the desired trajectory and is also not sensitive to chattering/noise, which
can cause discomfort to the subject.
Fig. 3. E-EXO control structure
Fig. 4. E-EXO position control results using estimated joint angle
6 Conclusion
This work presents FMG based position sensing and control method. The developed
sensor, S-Band, is able to read the muscle volume change with acceptable performance
and therefore has proven to be effective to estimate the elbow joint angle. The proposed
Position Sensing and Control with FMG Sensors for Exoskeleton Physical Assistance 7
method finds it application in classifying and estimating forearm, wrist and lower limb
motions for rehabilitation and assistance purposes. In future, the work will be focused
on improving the design and integrating the sensor within the exoskeleton, to use the
same control strategy for assistance in daily routine tasks.
Acknowledgment. The reported work is partially supported by EU-AAL Joint Programme

through project AXO-SUIT and Innovation Fund Denmark through project EXO-AIDER.
References
1. Christensen, S., Bai, S.: Kinematic analysis and design of a novel shoulder exoskeleton using
a double parallelogram linkage. J. Mech. Robot. 10(4), 041008 (2018)
2. Bi, L., et al.: EEG-based brain-controlled mobile robots: a survey. IEEE Trans. Hum. Mach.
Syst. 43, 161–176 (2013)
3. Gopura, R., et al.: Recent trends in EMG-based control methods for assistive robots. In:
Electrodiagnosis New Frontiers of Clinical Research, pp. 237–268 (2013)
4. Kadkhodayan, A., Jiang, X., Menon, C.: Continuous prediction of finger movements using
force myography. J. Med. Biol. Eng. 36, 594–604 (2016)
5. Jiang, X., et al.: Exploration of force myography and surface electromyography in hand
gesture classification. Med. Eng. Phys. 41, 63–73 (2017)
6. Islam, M., Bai, S.: Intention detection for dexterous human arm motion with FSR sensor
bands. In: Proceedings of Companion 2017 ACM/IEEE International Conference on Human-
Robot Interaction, pp. 139–140
7. Jiang, X., et al.: Ankle positions classification using force myography: an exploratory
investigation. In: 2016 IEEE Healthcare Innovation Point-of-Care Technology Conference,
pp. 29–32 (2016)
Force Localization Estimation
Using a Designed Soft Tactile Sensor
Merve Acer(&) and Adnan Furkan Yıldız
Mechanical Engineering Department, İstanbul Technical University,

İstanbul, Turkey
{acerm,yildizadn}@itu.edu.tr
Abstract. Wearable tactile sensors are significant in biomedical robotic

applications where force feedback is important. In this work, a soft tactile sensor
is proposed for force localization. The tactile sensor was manufactured by using
layer-by-layer technique that enables flexibility. The sensor has 9 lead zirconate
titanate (PZT) elements placed in 3 3 matrix form which are 4 4 mm2 and
the spatial resolution is 3 mm. The voltage values gathered from the sensor were
conditioned by a charge amplifier circuit. A human inspired machine learning
procedure called Neural Networks was used for force localization. The success
rates with respect to different network structures were presented and the maxi-
mum success was realized as 80.71%.
1 Introduction
Nowadays the robotic research focuses on building systems that can interact with the
environment effectively and safely. In the future, the robotic systems will not only be
tasked in the known, safe spaces but also collaborate with humans and even worn by
the humans to perform more complicated tasks in unknown spaces. However, the
physical and functional properties of the robots are limited by their actuator, sensor
components and by their physical architecture. We need to find new methodologies for
building soft, embeddable/wearable sensors that enable more functions. Besides, we
also need to find algorithms to use the developed sensors effectively and make the
system smarter.
Tactile sense is an important and developed sense which can provide more infor-
mation about the unstructured environment than vision especially in terms of force
feedback [1]. If tactile sensors can be developed as smart sensitive skins with the
requirements of biomedical applications [1, 2] then the technology will also be
developed and provide new solutions. This paper presents a wearable, soft 3 3 PZT
based tactile sensor with a sensitivity of 0.578–0.821 V/N for force localization using
Neural Networks (Fig. 1). The sensor description and data acquisition, the test setup,
and the neural network structures for the force localization from the provided signals
has been explained and finally the results have been investigated.
This project is supported by the Scientific and Technological Research Council of Turkey
(TUBITAK) with the project number: 215E139.

https://doi.org/10.1007/978-3-030-01887-0_2
Force Localization Estimation Using a Designed Soft Tactile Sensor 9
Fig. 1. Wearable tactile sensor for force localization using Neural Network structures for
machine learning algorithm.
2 Sensor Description and Test SetuP
The tactile sensor is composed of 9 PZT (PSI-5H4E) elements positioned as 3 3

matrix form and 5 layers as shown in Fig. 2. The gap between PZT taxels is 3 mm and
each PZT ceramic has ab area of 4 4 mm2. Two copper-kapton (Pyralux) electrodes
which were cut using laser cutter have been used for the top and bottom electrodes of
the PZT elements. In an attempt to have a human skin-like tissue on the sensor and to
protect the PZT elements from the force impacts, upper and lower parts of the sensor
were covered with a thin silicone (Ecoflex30). The detailed manufacturing process for
3 1 sensor array were presented in [3].
Fig. 2. (a) Layer by layer illustration of 3 3 matrix formed PZT based tactile sensor. (b) The
manufactured tactile sensor.
High input impedance characteristic of piezoelectric material requires an opera-

tional amplifier based signal conditioning unit. A charge amplifier circuit [4] has been
selected as the signal conditioner for our system due to the fact that the effect of cable
impedance can significantly be reduced in measurements. The resistance and the
capacitance values were selected with regard to the level of gain and the cut-off
frequency of the amplifier. The force inputs to the tactile sensor were provided by a
10 M. Acer and A. F. Yıldız
direct drive linear actuator mounted on the linear stage of z axis (Fig. 3a). National
Instruments 9264 DAQ was used for signal generation of the actuation which is a
10 Hz sine wave and the PZT taxels voltage output values were collected. The peaks of
the voltage values were measured via a peak detector block in LabVIEW. These
voltage peak values were then used for the training and validation processes of machine
learning algorithm. The applied forces on the sensor were also measured by HBM U9C
50N load cell.
Fig. 3. (a) The experimental setup (b) The coordinates of the sensor area (c) Continuous force
localization test of the tactile sensor for one row
3 Force Localization Estimation

3.1 Data Gathering
The force localization is possible continuously on the sensor although we have discrete
elements (Fig. 3c). For the 1st row of the sensor, starting from the PZT 1 to PZT 3, 1N
of force was applied on every 1 mm and the peak-to-peak output voltages from the PZT
elements are collected. Data acquisition process has been done for 14 14 mm2 with
1 mm sensitivity starting from (0, 0) to (14, 14) and therefore there are totally 225 data
points on the sensor used for training (Fig. 3b). Moreover, the validation data includes
196 data points which are shifted from training points by 0.5 mm starting from (0, 0.5)
till (13, 13.5) with again 1 mm distance between each points are set as validation
points.
3.2 Learning Algorithm

The position estimation was made from 0 to 14 mm for two dimensions. Since the
position is a continuous data for this range, regression based machine learning algo-
rithms should be used for the predictions. Multi-layer perceptron algorithm was used
Force Localization Estimation Using a Designed Soft Tactile Sensor 11
for learning procedure. The network comprises several neurons that were connected
with each other by weights and it was basically divided into three groups: the input
neurons having 9 neurons because of 9 PZT taxels voltage outputs, the output neurons
having 2 neurons for the estimation in 2D space and the hidden neurons The number of
the hidden neurons determines the smoothness of the decision boundary. Values of the
weights were set by a calculation method called back propagation [5, 6]. There are a
number of activation functions that can be used in network architecture [7]. In this
study, 3 of these activation functions were implemented to hidden layers one by one
without affecting input and output layers to observe the effects of the activation
functions on the learning process.
3.3 Estimation Results

In training process, different network combinations were used in order to find the
optimal network architecture and the estimations were made using two different group
of test data. All the network structures had 2 hidden layers. Moreover, the average
accuracy value was composed by the average value of two different test data accuracy
values. The network parameters and the estimation results have been presented in
Table 1.
Table 1. Estimation results

Hidden neuron number per layer Used activation functions Average accuracy [%]
3 Log-Sigmoid 78.94
10 Log-Sigmoid 75.71
20 Log-Sigmoid 72.73
3 Tan-Sigmoid 76.29
10 Tan-Sigmoid 67.20
20 Tan-Sigmoid 54.12
3 Rectified linear unit 44.01
4 Conclusion
In this paper, an application of force localization estimation using a designed soft PZT
based tactile sensor have been illustrated. Force localization estimation was made for
14 14 mm2 area on the tactile sensor using Artificial Neural Networks with 3 dif-
ferent activation functions. The best accuracy was obtained by using rectified linear
unit activation function with 10 neurons for each hidden layer and the average esti-
mation accuracy is 80.71%.
12 M. Acer and A. F. Yıldız
References
1. Tiwana, M.I., Redmond, S.J., Lovell, N.H.: A review of tactile sensing technologies with
applications in biomedical engineering. Sens. Actuators A Phys. 179, 17–31 (2012)
2. Kenry, J.C.Y., Lim, C.T.: Emerging flexible and wearable physical sensing platforms for
healthcare and biomedical applications. Microsyst. Nanoeng. 2, 16043 (2016)
3. Acer, M., Yıldız, A.F., Bazzaz, F.H.: Development of a soft PZT based tactile sensor array for
force localization. In: 2017 XXVI International Conference on Information, Communication
and Automation Technologies (ICAT), Sarajevo, pp. 1–6 (2017)
4. Karki, J.: Signal conditioning piezoelectric sensors. Application Report on Mixed Signal
Products, Texas Instruments Incorporated
5. Werbos, P.J.: Beyond regression: new tools for prediction and analysis in the behavioral
sciences. Ph.D. thesis, Harvard University, Cambridge, MA (1974)
6. Rumelhart, D.E., Hinton, G.E., Williams, R.J.: Learning representations by back-propagating
errors. Nature 323, 533–536 (1986)
7. Hagan, M.T., Demuth, H.B., Beale, M.H.: Neural Network Design. PWS Publishing, Boston
(1996)
EIT-Based Tactile Sensing Patches
for Rehabilitation and Human Machine
Interaction
Stefania Russo2 , Nicola Carbonaro1 , and Alessandro Tognetti1(B)

1
Information Engineering Department and the Research Center “E. Piaggio”,
University of Pisa, Pisa, Italy
alessandro.tognetti@unipi.it
2
Eawag, Swiss Federal Institute of Aquatic Science and Technology,
Department Process Engineering, Überlandstrasse 133,
8600 Dübendorf, Switzerland
Abstract. We present the development of an innovative stretchable tac-

tile sensor based on electrical impedance tomography (EIT) for appli-
cations in wearable robotics and rehabilitation. To extract the tactile
information we exploit the electrical impedance tomography technique
to reconstruct the local conductivity changes of a piezoresistive fabric.
The EIT method poses several new challenges in the reconstruction,
counterbalanced by the overcoming of many of the drawbacks of the cur-
rent tactile sensors. Results obtained are preliminary but encouraging
and we believe that the combination of the EIT method with advanced
machine learning techniques will enable reliable wearable tactile sensing.
1 Introduction
In the fields of neuro- and physical- rehabilitation and human machine inter-
action (HMI), there is a strong need for unobtrusive and conformable sensing
devices that do not interfere with the subject’s body and/or the robot’s mechan-
ics. Flexibility and stretchability of the sensing elements are the key factors to
achieve effective and reliable wearable human-robot interfaces able to detect user
movements, recognise movement intention and sense tactile interaction. We have
previously developed flexible and stretchable sensing interfaces able to detect
human posture and movement based on textile strain sensors and goniometers.
Textile goniometers are double layer piezoresistive device that can reliably sense
the joint angle of rotational joints [1]. They have been combined and fused with
inertial microsensors to reconstruct the movement of upper [2] and lower limbs
[3] with applications in neuro- [4] and physical- [5] rehabilitation and HMI [6].
To obtain fully functional wearable robotic interfaces, the capability to detect
kinematic information has to be augmented with the tactile sensing modality
(i.e. detection of normal/shear force and stimulus location), again in a flexible,
The research leading to these results has received funding from the People
Programme (Marie Curie Actions) of the European Union’s Seventh Framework
Programme FP7/2007–2013/ under REA grant agreement number 608022.
c Springer Nature Switzerland AG 2019
https://doi.org/10.1007/978-3-030-01887-0_3
14 S. Russo et al.
stretchable and conformable fashion. To detect the force interaction in multiple

points between the subject and the robotic interface has many applications, span-
ning from prosthesis control [7], movement intention recognition in exoskeletons
[8], to assessment of rehabilitation recovery [9].
In the present work we present the preliminary development of an innovative
tactile sensor based on electrical impedance tomography (EIT). Our sensor is
simply made of a piezoresistive fabric, where the local conductivity changes due
to the applied mechanical stimuli. The change in conductivity is reconstructed
thanks to the solution of the EIT inverse problem where current injection and
voltage reading electrodes are placed only at the boundary of the sensing area.
This approach overcomes several drawbacks intrinsically linked to distributed
tactile sensing (no wires on the sensing area, conformability, possibility to adapt
to irregular surfaces), even if it poses several new challenges in the reconstruction
of the tactile stimuli.
2 Materials and Methods
The presence of different materials or wires embedded in a tactile sensor is usu-

ally one of the main cause in the reduction of their flexibility and/or stretchabil-
ity. An approach that has been recently used to compensate for this drawback,
which is the main topic of this work, is EIT [10]. This technique allows to place
the electrodes only on the boundary of the active sensing area of a tactile sensor.
As a consequence, no wiring is present inside the sensor. Therefore, EIT-based
sensors can be placed over different surfaces even with irregular shapes as typ-
ically occur in the human robot interaction scenarios we are considering. EIT
techniques are ill-posed non linear inverse problems, where the aim is to recon-
struct the conductivity distribution of the body under study from measurements
taken at electrodes placed the boundary. The reconstructed conductivity is then
showed in an image by applying an inverse reconstruction algorithm [11].
A typical EIT system consists of a current source, a switching mechanism
for generating current injection patterns between the boundary electrodes and a
data acquisition unit for potential measurements [12]. In order to address some
of the necessary requirements for rehabilitation and human robot interaction
applications, we have developed an EIT system which presents low power con-
sumption, precision in the measurements and high temporal resolution. A block
diagram of our EIT platform is shown in Fig. 1 and consists of 3 main elements
which are described below.
The transducer element is represented by block 3. We have used a thin,
stretchable, piezoresistive fabric material provided by Eeonyx. The material has
a surface resistance of 30 KΩ, it is low-cost, light weight, very flexible, bendable
and conformable to different surfaces. For validation purposes we have used a 3D-
printed circular frame made out of two disc layers to house the conductive sheet.
The frame presents 16 extrusions where conductive copper stripes are placed to
create the electrodes. A custom printed circuit board (PCB) illustrated by block
2 along with its simple schematic in Fig. 1, is used for performing a DC constant
EIT-Based Tactile Sensing Patches 15
Fig. 1. Block diagram of our EIT experimental platform.
current injection and voltages measurements. Finally, block 1 illustrates the data
acquisition and multiplexing elements of our sensor system. The data acquisition
(DAQ) card along with MATLAB data acquisition toolbox are used for both
multiplexer control, and voltage data collection. Then, the image reconstruction
is performed using EIDORS [13]. This is an open source benchmark computation
approach that is commonly used for EIT imaging.
3 Results
Figure 2 shows the reconstructed images when a touch input is applied in dif-
ferent positions over the sensor. Red colour indicates a positive changes in the
conductivity, while a blue colour represents the ringing artifacts which is typical
in EIT systems. It also appears that the location of the touch input influences
the reconstructed conductivity changes in that area and its proximity. This is
due to the ill-posed and non linear nature of EIT [12]. The sensing rate of an
EIT system is the rate at which the number of samples used for the image
reconstruction are collected. In our case this corresponds to 78 Hz. However, the
image reconstruction frame rate depends on both the sensing rate and the time
to compute the inverse solution. In this work, the image reconstruction frame
rate is calculated to be 30 Hz.
Fig. 2. Reconstructed EIT images for a touch input applied in different locations over
the sensor.
16 S. Russo et al.
4 Conclusions
EIT overcomes several of the drawbacks found in common flexible tactile sen-
sors. Nonetheless, EIT presents a major drawback, it is mathematically severely
ill-posed, non-linear, and is very sensitive to small changes in potential at the
boundary measurements. Therefore, the image reconstruction of the internal con-
ductivity is apt to errors and suffers from low spatial resolution. Also, the touch
input detection speed in negatively affected by the time required to compute the
inverse solution.
The main finding of this work is that EIT flexible sensor systems present a
very promising technology, and can used for applications in wearable robotics
and rehabilitation. However, issues such as reduced detection speed and low
spatial resolution can jeopardize their effective implementation.
We are already working on using machine learning strategies in order to
overcome the main drawbacks of EIT sensors. Machine learning can in fact pro-
mote the application of these sensors in real world scenarios to their ability of
empirical learning and extracting meaningful tactile information.
References
1. Tognetti, A., Lorussi, F., Dalle Mura, G., Carbonaro, N., Pacelli, M., Paradiso, R.,
De Rossi, D.: New generation of wearable goniometers for motion capture systems.
J. Neuroeng. Rehabil. 11(1), 56 (2014)
2. Lorussi, F., Carbonaro, N., De Rossi, D., Tognetti, A.: A bi-articular model for
scapular-humeral rhythm reconstruction through data from wearable sensors. J.
Neuroeng. Rehabil. 13(1), 40 (2016)
3. Tognetti, A., Lorussi, F., Carbonaro, N., De Rossi, D.: Wearable goniometer and
accelerometer sensory fusion for knee joint angle measurement in daily life. Sen-
sors 15(11), 28 435–28 455 (2015)
4. Lorussi, F., Carbonaro, N., De Rossi, D., Paradiso, R., Veltink, P., Tognetti, A.:
Wearable textile platform for assessing stroke patient treatment in daily life con-
ditions. Front. Bioeng. Biotechnol. 4, 28 (2016)
5. Lucchesi, I., Lorussi, F., Bellizzi, M., Carbonaro, N., Trotta, L., Casarosa,
S., Tognetti, A.: Daily life self-management and self-treatment of musculoskeletal
disorders through shoulphy. In: Proceedings of the 7th EAI International Confer-
ence on Wireless Mobile Communication and Healthcare, 14–15 Novembre 2017,
p. 8 (2017)
6. Bianchi, M., Haschke, R., Büscher, G., Ciotti, S., Carbonaro, N., Tognetti, A.: A
multi-modal sensing glove for human manual-interaction studies. Electronics 5(3),
42 (2016)
7. Carbonaro, N., Anania, G., Bacchereti, M., Donati, G., Ferretti, L., Pellicci,
G., Parrini, G., Vitetta, N., De Rossi, D., Tognetti, A.: An innovative multisen-
sor controlled prosthetic hand. In: XIII Mediterranean Conference on Medical and
Biological Engineering and Computing 2013, pp. 93–96. Springer (2014)
8. Donati, M., Vitiello, N., De Rossi, S.M.M., Lenzi, T., Crea, S., Persichetti, A.,
Giovacchini, F., Koopman, B., Podobnik, J., Munih, M., Carrozza, M.C.: A flexible
sensor technology for the distributed measurement of interaction pressure. Sensors.
13(1), 1021–1045 (2013). http://www.mdpi.com/1424-8220/13/1/1021
EIT-Based Tactile Sensing Patches 17
9. Sadarangani, G.P., Jiang, X., Simpson, L.A., Eng, J.J., Menon, C.: Force myog-
raphy for monitoring grasping in individuals with stroke with mild to moderate
upper-extremity impairments: a preliminary investigation in a controlled environ-
ment. Front. Bioeng. Biotechnol. 5, 42 (2017)
10. Silvera-Tawil, D., Rye, D., Soleimani, M., Velonaki, M.: Electrical impedance
tomography for artificial sensitive robotic skin: a review. IEEE Sens. J.15(4), 2001–
2016 (2015)
11. Adler, A., Arnold, J.H., Bayford, R., Borsic, A., Brown, B., Dixon, P., Faes, T.J.,
Frerichs, I., Gagnon, H., Gärber, Y.: GREIT: a unified approach to 2d linear eit
reconstruction of lung images. Physiol. Meas. 30(6), S35 (2009)
12. Holder, D.S.: Electrical Impedance Tomography: Methods, History and Applica-
tions. CRC Press, Boca Raton (2004)
13. Adler, A., Lionheart, W.R.: Uses and abuses of EIDORS: an extensible software
base for EIT. Physiol. Meas. 27(5), S25 (2006)
Synthesis and Optimization Considerations
for a Knee Orthosis Based on a Watt’s Six-Bar
Linkage
Evagoras Xydas(&), Banu Abdikadirova, and Kostas Konstantinos
Mechanical and Aerospace Engineering Department, School of Engineering,

Nazarbayev University, Astana, Kazakhstan
evagoras.xydas@nu.edu.kz
Abstract. Employing six-bar linkages for knee orthoses is an advantageous

choice in many ways. In relation to four-bar linkages the six-bar linkage tracks
more accurately the natural ankle trajectory and provides more flexibility in the
design for stability. Additionally, the 1-DOF mobility of linkages alongside the
available moment leverage offers advantages in cost-effectiveness over open-
kinematic-chain designs.
This work presents the analysis as well as the design and optimization con-
siderations of a knee orthosis that is based on a Watt’s six-bar linkage. The main
aim is to lay the foundations for analytical design and development of a knee
orthosis that is cost-effective, self-cancelling and optimized for passive-active
energy exchange as well as energy harvesting and return.
1 Introduction
Recently, it has been shown that metabolic cost during walking can be reduced by
using both passive or active means in the form of exoskeletons or orthoses. In this
context, passive means without external energy supplied to the device whereas active
devices are actuated by external sources. Despite the abundance of available passive,
active and hybrid exoskeletons, essentially only two research groups have managed to
establish reduction of metabolic cost in exoskeleton-assisted normal gait with statistical
significance. More specifically, Mooney et al. [1] employed a powered ankle
exoskeleton that does positive work by supporting foot extension during the push-off
stage, while disengaging during the rest of the gait stages. This work has been a turning
point regarding our understanding of human-exoskeleton interaction. Nevertheless, the
real challenge would be to achieve the reduction of metabolic cost without supplying
external power to the user. One way to achieve this is the improvement of physical
performance by harvesting energy during less demanding stages of walking (in which
probably inertia and gravity take over) and delivering it back during more demanding
stages like in push-off. Such a solution was developed by Collins et al. [2] and has
The authors wish to acknowledge Alfonso Hernández, CompMech, Department of Mechanical

Engineering, UPVEHU for the permission to use the GIM® software. (www.ehu.es/compmech).

https://doi.org/10.1007/978-3-030-01887-0_4
Synthesis and Optimization Considerations for a Knee Orthosis 19
shown a 7% reduction of metabolic cost in normal gait without any supply of power
externally. The device employs a clutched spring that stretches during toe flexion and
returns the energy during extension at the push-off stage, then disengages for the swing
phase. If one considers that our body has been evolving for millions of years this is a
truly remarkable invention that above all proves that the performance of the human
body can be improved by passive means. Follow-up work by Mooney et al. [1] has
established that a significant portion of the power exchange between the foot and the
ground mitigates to the knee and hip joints. This discovery suggests that energy-wise
performance can be further improved by using a knee exoskeleton or orthosis. The
current work presents the preliminary considerations for drafting a roadmap regarding
the analytical design and development of such a device. A Watt’s six-bar is initially
synthesized using established geometry for knee prostheses. Next, the desired opti-
mization and design variables and concepts are presented, followed by a set of
conclusions.
2 Synthesis of the Basic Linkage
Different authors have investigated different linkage arrangements for either orthotic or
prosthetic knee devices, by employing either four-bar or six-bar solutions. Among four-
bar and different inversions of Watt’s and Stephenson’s six-bars, it appears that Watt’s
six-bar can achieve the most accurate tracking of the ankle trajectory. Using the initial
dimensions that are based on athropometric date and Motion Capture (MOCAP)
analysis we performed synthesis and positional analysis for a Watt’s six-bar which is
shown in Fig. 1 (as generated by the analysis). Figure 2 presents the linkage as
designed in GIM software [3] for validation).
Fig. 1. Watt’s Six-bar (from calculations)

20 E. Xydas et al.
Fig. 2. Watt’s six-bar (GIM software [4])
2.1 Grashof Condition and Geometry

A basic concept in mechanism synthesis Grashof condition states that if,
LþSPþQ
then at least one link on a four-bar linkage will perform a full rotation, where L and S
are the longest and shortest links respectively and P, Q the remaining links.
3 Design and Optimization Considerations
It is essential to have at least one link performing a full rotation so that a unidirectional
motor can be used. This will result the lowest possible cost alongside simple control
and electronics.
In Fig. 3 it is seen that by changing the length of link B-D to a certain extent it
performs a full rotation over a full path of the mechanisms further point. In this stage,
geometric optimization can be carried out that ensures fit of the mechanism to a
standard ankle trajectory while achieving a Grashof condition.
3.1 Kinematic and Dynamic Optimization

The main goal in the mechanism’s design optimization is to generate a good
approximation of the required knee-torque profile with a series of appropriately
selected & positioned springs. This approach has been already employed in [10]
for the case of a four-bar linkage used in the design of a passive-active controlled
therapeutic mechanism. In that work, positions, and attachment configurations of
two appropriately selected springs were optimized aiming in the replication of a
required torque profile by minimizing the l2 norm over the full operating range.
The employed springs were not simultaneously attached at the mechanism’s links
and therefore, their forces contributed to the approximation of separate parts of the
Synthesis and Optimization Considerations for a Knee Orthosis 21
Fig. 3. Grashof case
sought-for torque profile. This approach led to a non-smooth – tangent discon-

tinuous – torque curve, which we would like to rectify in this work. Hence, in
addition to replicating the required knee-torque profile, in this worked it is desired
to achieve a smooth, tangent-continuous curve by imposing appropriate constraints
at curve’s segments’ endpoints. To this end, two different approaches for springs’
placement & properties selection in our optimization procedures:
(1) Optimize the length, position, and attachment of two (or more) strings bound at
fixed points that attach/detach to a single link for limited time during the mech-
anism’s operation cycle.
(2) Optimize the length, position, and attachment of one (or more) string(s), con-
tinuously & simultaneously attached to two links of the mechanism.
MATLAB’s gradient-based and evolutionary optimization algorithms are
employed in the problems above where, in both cases, the objective function comprises
the l2 norm calculated between the required and achieved torque’s pointsets over the
mechanism’s operating range. Finally, appropriate side constraints are used for limiting
the values of the design variables (springs’ lengths, attachment position coordinates,
etc.) while inclusion of the corresponding number of non-linear constraints imposes the
required continuity at torque-curve segments’ end-points.
4 Conclusion and Future Work
Overall, given the synthesis and optimization considerations described, four different
design/design optimization criteria are addressed: 1. Grashof condition, 2. Inertial
optimization by position of links centers of mass, 3. Passive-active optimization, 4.
Energy harvesting and return. The 1st consideration enables geometric fit and allows
the use of unidirectional servo-motor without the need for back-drivability and with
simple controls. The 2nd and 3rd conditions cancel out the mechanism inertia and
reduce the size of the motor, and the 4th essentially harvests the body energy during less
demanding stages of the stride and returns that energy in highly demanding positions.
22 E. Xydas et al.
References
1. Mooney, L.M., Rouse, E.J., Herr, H.M.: Autonomous exoskeleton reduces metabolic cost of
human walking during load carriage. J. NeuroEng. Rehabil. 13, 4 (2014)
2. Collins, S.H., Wiggin, M.B., Sawicki, G.S.: Reducing the Energy Cost of Human Walking
Using an Unpowered Exoskeleton (2015). https://doi.org/10.1038/nature14288
3. Petuya, V., Macho, E., Altuzarra, O., Pinto, C., Hernández, A.: Educational software tools for
the kinematic analysis of mechanisms. Comput. Appl. Eng. Educ. 22, 72–86 (2014)
4. Xydas, G.E., Mueller, A., Louca, L.S.: Minimally actuated fourbar linkages for upper limb
rehabilitation. Mech. Mach. Sci. 48, 131–146 (2018)
Wearable Sensory Apparatus
Performance While Using Inertial
Measurement Units
Grega Logar(B) , Zoran Ivanic, and Marko Munih
Faculty of Electrical Engineering, University of Ljubljana, Ljubljana, Slovenia

{grega.logar,zoran.ivanic,marko}@robo.fe.uni-lj.si
Abstract. With the objective of satisfying the technical and functional

requirements set for intention detection (ID) algorithm, we investigated
suitable strategies to reduce the number of remote units such as inertial
measurement units (IMU). The optimization should not affect the ID
algorithm, which needs to detect continuous transitions between different
locomotion modes such as ground-level walking, walking up and down
slopes, climbing/descending stairs, standing up, sitting down, turning
and other scenarios of real life. In the developed solution, the number of
remote units was reduced from 9 to 4. In order to achieve the same level
of ID algorithm performance, the WSA units need to perform almost per-
fectly. The main innovation is nearly perfect data transfer from remote
units to the master unit. This way a package loss below 0.05% of trans-
ferred packages is achieved.
1 Introduction
Several robotic-driven lower-limb prostheses have been developed as research
prototypes. Different configurations employ fully active, or semi-active only knee-
, ankle-foot- or combined knee-ankle- driven joints [1–4]. Active or semi-active
prostheses need to be triggered. They are triggered by the controller, that needs
feedback information typically provided by integrated encoders, sensors mea-
suring the interaction forces, EMG (electromyographic) electrodes or inertial
sensors. Common to known approaches is that the robotic driven joints are
operating on the basis of locally assessed information about the current state
in the gait cycle. Whole body awareness promises more precise assessment of
the current motion status, thus enabling richer feedback information and better
closed-loop control.
The objective of this work is to present how the performance of IMUs signals
can affect the algorithms of whole body awareness controller for various appli-
cations. Our version of remote units was developed in a way that can be easily
This work was supported by the CYBERLEGs Plus Plus project under grant 731931
of the call H2020-ICT-24-2015, and by the Slovenian Research Agency. We would
like to thank Filippo Dell’Agnello, Elena Martini, and Andrea Parri from Scuola
Superiore Sant’Anna, Pisa, Italy for their contributions.
https://doi.org/10.1007/978-3-030-01887-0_5
24 G. Logar et al.
implemented in any controller unit such as the one of the CYBERLEGs (CLs)
prosthesis preliminarily presented [5] and is further developed.
2.1 Wearable Sensory Apparatus
The WSA consists of a pair of pressure-sensitive insoles and a set of 9-

DOF inertial measurement units (IMUs). The units can be easily attached to
body segments with dedicated elastic straps with silicon to prevent them from
slipping.
Signals from a pair of instrumented shoes equipped with novel pressure-
sensitive insoles and a set of IMUs are equipped with 32-bit Microprocessor
Cortex-M4 (with floating point unit for onboard calculation). The design and all
know-how bonded to insoles is property of Scuola Superiore Sant’Anna (SSSA).
The Insole Control Units and IMUs are also equipped with RF module based on
MAC layer of IEEE 802.15.4-2011 (Ultra-wide Band) standard interface which
offer robust real-time transfer of sensors data. The electronic boards are pow-
ered by LiPo batteries. The instrumented shoes have the electronic board that
online pre-processes the data from the 16 pressure-sensitive elements. The Instru-
mented shoes and the set of IMUs consist of a 9-DOF single chip device (3D
accelerometer, 3D gyroscope, 3D magnetometer). The signals from sensors are
wirelessly transferred to the Central Cognitive Control Unit (CCU). The size of
the electronics and an example of mounting the units on the user can be seen on
Fig. 1.
Fig. 1. (Left) IMU electronic board with RF module, 9-DOF sensor and 32-bit micro-
processor. (Right) Active IMU board with LiPo battery in 3D-printed enclosure.
2.2 Transfer Protocol of Sensors Signals
The Master Unit (MU) is the board devoted to acquire wireless data coming
from the set of IMUs and insoles and communicates via SPI or USB with the
CCU. The software running on the CCU can be further split into two layers:
the algorithm handles data receiving on the lower level and processes the raw
signals on the higher level. The communication between MU and remote units
Wearable Sensory Apparatus Performance 25
are done via RF modules (DWM1000). The devices communicates using master-
slave architecture. MU communicate with remote units in 10 ms intervals. In the
beginning of interval the MU sends command “Start sampling” with broadcast
message to all remote units. Each remote then unit samples the data and starts
a delay timer for response message. When the delay timer expires, the data is
transmitted back to the MU. The delay timer differs based on remote address.
The synchronization between units is the “Start sampling” message. Insole data
are 58 Bytes long, IMU data are 22 Bytes long. Full package sent from the MU
to CCU is 230 Bytes long.
2.3 Background of Intention Detection Algorithms
The ID algorithm developed in the CYBERLEGs project was based on seven

IMUs (three IMUs on the shoe, the shank and the thigh of each side, and one
IMU on the trunk) and two pressure-sensitive insoles [6]. This configuration was
used to classify the locomotion mode in all the system configurations.
The ID algorithm used for the CLs++ developed by the SSSA team aimed
at addressing the following improvements. First, the number of IMUs used for
intention detection is reduced from 7 to 2, i.e. only the IMUs on both thighs.
The algorithm still relies on the use of the two shoes instrumented with pressure
sensors [7]. Second, the SSSA developed the algorithm that has a large chance
of predicting the upcoming locomotion transitions before foot-strike happens on
the new terrain. With this ability, we expected that walking with the robotic
prosthesis can be smoother and more natural. The detailed description of the
algorithms is too extensive and will be published in its own paper.
3 Results
The algorithms described relay on signals acquired from sensors. In order for
the algorithm to give accurate results the sensors need to perform perfectly.
Despite working in a controlled environment disturbances are present, such as
misalignment of sensor axes, the gyroscopes have bias, the magnetometers can
be pre magnetized. The biggest problem in ensuring proper algorithm results
is the package loss during data transferring. In previous WSA version we used
two different technologies for transferring data, the IEEE 802.15.4-2008 (Narrow
Band) and Bluetooth 2.0. The package loss was high, almost impossible for use
outside of controlled environment. The newest version uses only IEEE 802.15.4-
2011 (Ultrawide Band) technology, with which we achieve package loss below
0.05% of transferred packets (in average we lose 3 packets per minute at 100 Hz
sampling rate). The full comparison between WSA sets is shown in Table 1. One
master unit with current firmware can handle 9 remote units at 100 Hz sampling
rate (18 at 50 Hz, 4 at 200 Hz, and 1 at 800 Hz).
26 G. Logar et al.
Table 1. Comparison of CYBERLEG WSA technical specifications
CLs CLs Plus Plus

Num. of IMUs 7 (802.15.4-2008) 2 [5] (802.15.4-2011)
Num. of Insoles 2 (Bluetooth 2.0) 2 (802.15.4-2011)
Num. of remote units 9 + 3 4 + 1 [7 + 1]
Sensors 3 × 3DOF 1 × 9DOF
MCU 8-bit (16 Mhz) 32-bit (80 Mhz)
Data transfer rate 250 kbps 6.8 Mbps
RF bands 2.4 GHz 3.5–6.5 GHz
ID algorithm type Data fusion Rule-based
Battery managment No LiPo charger
Battery capacity 270 mAh 240 mAh
Idle time 5h 7h
Operating time 5h 4h
4 Discussion and Conclusion

For advanced sensory fusion processing (for example: the Complementary,
Kalman, Mahony, and Madgwick Filter) almost full number of packages is needed
for intended movement recognition. In order to achieve this we switched from
two different technologies for transferring data which are on the same very busy
frequency band 2.4 GHz to one technology which is on a different frequency
band between 3.5–6.5 GHz. This frequency band is less populated, therefore
fewer disturbances are present. The possibility for package loss is significantly
lower.
In previously developed algorithm we used sensory fusion for estimation of
segment orientation. We also needed to perform sensor calibration for aligning
the axes of sensors because they were in three different packages (integrated
circuits). The new algorithm no longer relays on sensory fusion for orientation
calculation, but they uses only raw signals. In addition we no longer use magne-
tometers for reliable detection of locomotion modes therefore we do not need to
perform magnetometer calibration and demagnetization.
The developed rule-based gait phase detection algorithm with a two-step pro-
cess based on signals from the WSA system demonstrates that the ID algorithm
and optimized packet transferring can perform with average success rate across
all subjects higher than 90% for all phases in real-time. The proposed system is
computationally simple.
Wearable Sensory Apparatus Performance 27
References
1. Lambrecht, B.G., Kazerooni, H.: Design of a semi-active knee prosthesis. In: Pro-
ceedings of the IEEE International Conference on Robotics and Automation, Kobe,
Japan, pp. 639–645 (2009)
2. Martinez-Villalpando, E.C., Herr, H.: Agonist-antagonist active knee prosthesis: a
preliminary study in level-ground walking. J. Rehabil. Res. Dev. 46(3), 361 (2009)
3. Au, S.K., Weber, J., Herr, H.: Powered ankle-foot prosthesis improves walking
metabolic economy. IEEE Trans. Robot. 25(1), 51–66 (2009)
4. Pillai, M.V., Kazerooni, H., Hurwich, A.: Design of a semi-active knee-ankle pros-
thesis. In: Proceedings of the 2011 IEEE International Conference on Robotics and
Automation, Shanghai, China, pp. 5293–5300 (2011)
5. Ambrozic, L.: CYBERLEGs: a user-oriented robotic transfemoral prosthesis with
whole-body awareness control. IEEE Robot. Autom. Mag. 21(4), 82–93 (2014)
6. Goršič, M.: Online phase detection using wearable sensors for walking with a robotic
prosthesis. Sensors 14(2), 2776–2794 (2014)
7. Parri, A.: Whole body awareness for controlling a robotic transfemoral prosthesis.
Front. Neurorobotics 11, 25 (2017)
WeFiTS: Wearable Fingertip Tactile
Sensor
Elif Hocaoglu1,2(B)
1
Department of Bioengineering, Imperial College of Science,
Technology and Medicine, London, UK
ehocaoglu@medipol.edu.tr
2
Electrical and Electronics Engineering, Istanbul Medipol University,
Istanbul, Turkey
Abstract. Thimble/glove-based wearable systems are opportunistically

placed on fingertips/hands and enable haptic devices, robotic/prosthetic
hands to gather valuable information about physical interaction with
the environment in an easy way. In particular, incipient slip detection
and force acquisition are two important phenomena for human/robotic
fingertips to successfully manipulate the real or virtual objects. In this
study, a wearable fingertip tactile sensor (WeFiTS) capable of sensing
incipient slippage and force variation in various directions to enhance
interaction and manipulation while performing tasks is proposed. The
fundamental design criteria of WeFiTS are also evaluated with FEA fea-
sibility analysis and experimental evaluation.
1 Introduction
Human hand characteristics have been a perfect model for robotic studies which
have been devoted to augment the interaction/manipulation success with the
environment. In particular, wearable devices for fingers have been developed
inspired by the properties of human fingers to be employed by haptic robotics,
robotic/prosthetic hands for real/virtual object manipulations and also human
gesture analyses. Recent studies [1,2] have designed to acquire information about
force/torque measurement by means of force/torque sensors and also contact
points with the aid of contact centroid algorithm.
Considering stable manipulation with both robotic fingers (grippers) and
human fingertips, slip perception is thought to be another important design
criteria. Human is capable of detecting slippage, particularly incipient slip, by
means of various types of receptors placed in fingertips. In literature, some groups
have focused on human-like slip sensible fingertip design to realize robust manip-
ulation [3]. However, such devices [4,5] are not easily adaptable to some stud-
ies, such as human integrated research, and specifically designed for respective
robotic hands.
This work has been partially supported by TUBITAK-BIDEB 2219 Grant, Turkey,
the EU-FP7 Grants CONTEST (ITN-317488) and the internal funding of Istanbul
Medipol University, Turkey.
https://doi.org/10.1007/978-3-030-01887-0_6
WeFiTS: Wearable Fingertip Tactile Sensor 29
In this study, WeFiTS is presented as a wearable fingertip tactile sensor to be

worn by human or robotic fingertips, adaptable to many types of manipulation
based studies, and provides essential information about exerted force upon object
and incipient slip detection at fingertip, which are achieved by means of force
sensing resistors and non-contact rotary encoders, respectively without requiring
complex algorithm and prior knowledge regarding object or environment.
2 Problem Definition
The sophisticated problem of fingertip design endowed with slip detection and
force sensing properties can be addressed with an easier and practical solution.
WeFiTS is proposed as a simple but functional wearable tactile sensor, that
allows adequate nest for distal phalanges and can be worn by gloves or fixed by
straps as illustrated in Fig. 1.
Fig. 1. Illustrative representation of WeFiTS (a) worn by a robotic hand (b) grasping
an object
WeFiTS includes two different sensor types, position and force sensors, which
are responsible for slip detection and force measurement in different direc-
tions, respectively. The simplified views of WeFiTS and its working principle
are depicted in Fig. 2. In the top view (a), force sensors represented with two
dots are placed on inner surface of WeFiTS in order to be directly connected
with fingertips. In the front (b) and left (c) views, two slip sensors are placed
in perpendicular to each other to sense the possible object slippage occurred in
two directions (d, e).
WeFiTS addresses to the requirement of simultaneous and independent slip and

force sensations for robotic/human fingers by means of a wearable easy-to-use
design.
30 E. Hocaoglu
(a)
Ø 9mm
(b) (c)
70°
(d) (e) (f)
Fig. 2. This figure shows top view and cross-section of right and front views of WeFiTS
together with the working principle of the sensors. In (a), two force sensors represented
with two dots are placed on the surface of the WeFiTS. In (b) and (c), two slip sensors
assembled to the bottom part of WeFiTS. In (d) and (e), the motion of objects in
different directions are schematically represented. In (f), the red dashed curve shows
the ratio of the interacted section of the circle with an object.
3.1 Design Criteria

Some important design criteria needs to be tackled to satisfy the requirements.
Lightness is a quite important design criterion to minimize the effort for lifting
WeFiTS. Size and shape are also two coupled primary design criteria requiring
meticulous attention to well suit to the fingertip’s geometry and customizable
to any-size of user. Ergonomy is imperative requirement to be considered in
order not to prevent the motion of fingers in a space. Finally, sensibility is the
fundamental design criteria such that WeFiTS senses the incipient motion of an
pinched object and the force variation exerted by fingertips. This criteria should
also cover ease in detection of the force and slippage without requiring complex
algorithms.
3.2 Electromechanical Design of WeFiTS

The mechanical design of WeFiTs is mainly composed of two layers with inte-
grated sensors presented in Fig. 3 as exploded view. Light-weight design criterion
is met with the selection of PLA material, whose density is 1.25 g/cm3 , by avoid-
ing more than required size for fingertips (18 × 27 × 11 mm) and designing the
overall platform durable but thinner enough. In order to satisfy the size and
shape criteria, two main bodies of WeFiTS are produced by means of 3d printer
manufacturing technique, which avails in terms of custom-based production for
any size of human or robotic fingertip. Ergonomy is addressed by designing the
overall system to interact with the environment most efficiently and safely. As
for the sensation criteria, force sensing is realized by placing force sensing resis-
tors (FlexiForce A101 by Tekscan Inc.), which are sensitive up to 44 N, into the
nest area of WeFiTs to measure the force in all axes as in Fig. 3. Feasibility
of the aforementioned design criteria is tested with FEA analyses under over-
load (30 N). Figure 4 presents the response of body under the static nodal stress
and displacement under static load. Negligible deformations with less than 1 mm
ensures that proposed design is feasible to satisfy the design criteria. The slip
WeFiTS: Wearable Fingertip Tactile Sensor 31
sensation is realized by detecting the motion of rotary body placed in between

two main bodies of WeFiTs as seen in Fig. 3 by means of 12 bit non-contact
rotary encoders (RS08, RLS), which is sensitive enough to detect quasi-static
motion (1 count for 0.09◦ revolution).
Fig. 3. This figure shows the exploded CAD view of WeFiTS with its descriptions.
Fig. 4. (a) Von Mises result under the static nodal stress for upper body (b) Result of
static displacement for upper body (c) Von Mises result under the static nodal stress
for lower body (d) Result of static displacement for lower body
The experimental evaluation is presented in Fig. 5 under various conditions.

In particular, when less than required force is exerted on an object, cylindrical
rotary body moves in the direction of slippage. Here, the negative and positive
signs on the y axis represent the direction of the rotation of cylindrical rotary
body. In order to explicitly show incipient slippage, rotation of encoder is rep-
resented in terms of count rather than degree. In addition, as emphasized with
32 E. Hocaoglu
the red and green arrows, the sufficient enough force stops the rotation of the
cylindrical bodies and encoder sends the last number to the microprosessor at
those moments.
10 Incipient slippage is detected

0 due to the insufficient amount of force.
Encoder [count]
-20
No force is acting on the object,
-40 slippage of the object is detected.
No slippage occurs
-60
-80 Applied force is proportionally

increased with the loads of
-100 the object to prevent slippage.
0 5 10 15 20 25
Force [N]
Fig. 5. The working principle of WeFiTs is presented in different cases.
4 Conclusion
This study presented a novel wearable fingertip tactile sensor sensible to slip-
page and force variation. The FEA results showed the feasibility analysis of the
mechanical design and the experimental evaluation of WeFiTs also represents
the proposed sensation criteria by showing the incipient slippage and applied
force level to the object.
References
1. Ferre, M., Galiana, I., Aracil, R.: Design of a lightweight, cost effective thimble-like
sensor for haptic applications based on contact force sensors. Sensors (Basel) 11(12),
495–509 (2011)
2. Battaglia, E., Grioli, G., Catalano, M.G., Santello, M., Bicchi, A.: ThimbleSense:
an individual-digit wearable tactile sensor for experimental grasp studies. In: IEEE
International Conference on Robotics and Automation (ICRA), pp. 2728–2735
(2014)
3. Ho, V., Hirai, S.: Understanding slip perception of soft fingertips by modeling and
simulating stick-slip phenomenon. In: Proceedings of Robotics: Science and Systems,
pp. 2728–2735 (2011)
4. Roberts, L., Singhal, G., Kaliki, R.: Slip detection and grip adjustment using optical
tracking in prosthetic hands. In: IEEE EMBC International Conference on Engi-
neering in Medicine and Biology Society, pp. 2929–2932 (2011)
5. Melchiorri, C.: Slip detection and control using tactile and force sensors.
IEEE/ASME Trans. Mechatron. 5(3), 235–243 (2000)
Soft Wearable Robots
Characterisation of Pressure Distribution
at the Interface of a Soft Exosuit:
Towards a More Comfortable Wear
Michele Xiloyannis1(B) , Domenico Chiaradia2 , Antonio Frisoli2 ,

and Lorenzo Masia3
1
School of Mechanical and Aerospace Engineering, Nanyang Technological
University, Singapore, Singapore
michele001@ntu.edu.sg
2
Scuola Superiore Sant’Anna, TeCIP Institute, PERCRO Laboratory, Pisa, Italy
{domenico.chiaradia,antonio.frisoli}@santannapisa.it
3
Department of Biomechanical Engineering, University of Twente,
Enschede, Netherlands
lormasia@gmail.com
Abstract. The rapid growth of wearable robots in the last decade

requires tackling practical issues that arise from their daily use, among
which comfort is of great importance. In this work we quantify the level
of comfort of a soft exosuit for the elbow by measuring the distribu-
tion of pressures at its interface with the human body. We do so with
five different cushioning materials, commonly used in sport equipment
and orthoses, and identify the ones exhibiting lower peaks of pressure.
Polyethylene sponge and neoprene result in the best padding.
1 Introduction
“It is extraordinary to me that the most mature and old technology in the human
timeline, the shoe, still gives us blisters. How can this be? We have no idea
how to attach things to our bodies.” [1]. These words clearly point out a critical
yet underrated issue in wearable robots: designing a comfortable mechanical
interface between the device and its wearer.
As the field of wearable robots rapidly grows and assistive devices promise
to soon become part of our daily lives, it is key to face the challenge of finding
a comfortable mechanical means to attach robots to the human body. This is
all the more important for devices intended for users with sensory impairment,
where the lack of feedback could lead to pressure ulcers or local blood flow
obstruction.
Extensive research exists on the assessment and optimisation of the design
of prosthetic sockets [2]; comfort is strongly correlated with the magnitude of
pressure peaks and shear forces at the interface between the socket and the skin,
although various other factors, such as temperature and perspiration, can play
a role.
https://doi.org/10.1007/978-3-030-01887-0_7
36 M. Xiloyannis et al.
Significantly less work exists for exoskeletons. De Rossi et al. proposed an

apparatus to monitor the distribution of pressure at the human-robot interface
of a lower-limb exoskeleton during gait training [3]. Using a similar approach,
Levesque et al. identified areas exhibiting peaks of pressure on the attachment
points of an active orthosis for gait assistance [4]; their findings suggest that the
distribution of pressure is affected more by the stiffness of the padding material
than by its thickness.
This data-driven approach to guide design choices is a promising path: Quin-
livan et al. used the same paradigm to optimise the topology and material com-
position of the attachment points of a soft exosuit for the lower limbs [5]. This
work highlighted the importance of the geometry of the interface (larger areas
lead to higher comfort) as well as advising the use of fabric materials that better
conform to the human body.
Being for exoskeletons or exosuits, materials such as neoprene or polyethilene
(PE) sponge are common choices for cushioning the human-robot interface, yet
there is no data-supported knowledge to justify the choice of one over others.
In this work, we evaluate the pressure distribution between the skin and the
anchor points of a soft exosuit for different cushioning materials. We test foams
and rubbers commonly used for padding sports equipment and orthoses and
quantify their comfort by measuring peaks of pressure at the interface with the
skin.
Fig. 1. The soft exosuit for assistance of the elbow joint. Pressure distribution measures
were taken at the interface between the distal strap and the skin. Photography by

c Stefano Mazzoni.
Characterisation of Pressure Distribution at the Interface of a Soft Exosuit 37
Fig. 2. Evaluation of pressure distribution at the human-suit interface of the distal

anchor point (shown in Fig. 1) in the worst-case scenario (forearm perpendicular to
gravity). (a) Examples of pressure mapping on the forearm using nitrile foam and
polyethylene (PE) sponge, respectively being the worst and best scenarios among the
tested materials. (b) Peak pressure at the human-suit interface. Opaque dots show the
mean over repetitions. Neoprene and PE sponge result in the lowest peaks.
2 Methods
The exosuit for assistance of the elbow joint consists of three wearable compo-
nents, highlighted in Fig. 1, and an electric motor (Maxon EC-i 40, 70W, 55:1
reduction), driving a pair of tendons to assist in both flexion and extension of
the joint. The two tendons, routed through a Bowden sheath to the elbow, are
attached on both side of the joint to mimic the biceps brachii and the triceps
muscles. The suit is equipped with a load cell (Futek LCM300) in series with
the flexing tendon and an absolute encoder to monitor the elbow’s position.
When the tendons are tensioned to apply a torque on the elbow, the proximal
and distal strap tend to migrate towards the center of the joint. To reduce the
amount of movement, the proximal strap is connected via inextensible fabric to
a harness that redistributes forces on the torso; the large contact area with the
body ensures a comfortable wear. The distal strap is where the highest pressures
are applied, hence where cushioning properties of different materials will be more
evident.
The distribution of pressure around the forearm was acquired using a pres-
sure mat (NexGen Ergonomics, BT5010, 28 cm × 28 cm) consisting of an array
of 256 pressure sensors, sampled at 100 Hz. The pressure mat was wrapped
around the surface of the forearm and secured using double-sided tape to pre-
vent displacement. One subject was asked to performed three sessions of three
flexion/extension movements between 0◦ and 90◦ (0◦ being full extension) with
the arm fixed on the side of the trunk, doffing and donning the exosuit between
sessions. This was done for each of the five materials listed in Fig. 2b: each
38 M. Xiloyannis et al.
material was cut in a 3mm-thick layer and used to line the internal surface
of the distal strap. The exosuit provided an assistance equal and opposite to
gravity, running the gravity-compensation algorithm described in [6].
3 Results and Conclusions

Unsurprisingly, the highest peaks in pressure were found when the elbow was
flexed at 90◦ , i.e. when the component of gravity acting on the arm reaches
its maximum value, on the posterior side of the forearm. Figure 2a shows the
distribution of pressure in this configuration when the distal strap was lined with
nitrile foam (top) and with polyethylene sponge (bottom), the former showing
higher peaks and a less homogeneous distribution. Figure 2b shows the peak
values of pressure for each of the nine repetition (transparent dots) and the mean
over repetitions (opaque dots). PE sponge is the material exhibiting the best
behaviour while rubber-based materials showed higher local peaks of pressure.
It is worth noting that obtained pressure profiles were due to the subject’s
weight only; it would be interesting to examine how they vary as the wearer
lifts extra load. Moreover, a more thorough study to measure comfort would
include an estimate of shear forces on the skin and examine how the force profiles
change across subjects with different muscle tone. This data, correlated with
physiological measures such as subcutaneous blood flow, could be used to drive
the design of optimal interfaces for wearable robots and would provide interesting
insights on their usability in daily life scenarios.
References
1. Herr, H.: The new bionics that let us run, climb and dance. TED, Lect. (2014)
2. Mak, F., Zhang, M., Boone, D.: State-of-the-art research in lower-limb prosthetic
biomechanics-socket interface: a review. J. Rehabil. Res. Dev. 38(2), 161–174 (2001)
3. de Rossi, S.M.M., Vitiello, N., Lenzi, T., Ronsse, R., Koopman, B., Persichetti,
A., Vecchi, F., Ijspeert, A.J., van der Kooij, H., Carrozza, M.C.: Sensing pressure
distribution on a lower-limb exoskeleton physical human-machine interface. Sensors
11(1), 207–227 (2011)
4. Levesque, L., Pardoel, S., Lovrenovic, Z., Doumit, M.: Experimental comfort assess-
ment of an active exoskeleton interface. In: IEEE 5th International Symposium on
Robotic Intelligence Sensors, Ottawa, October 2017
5. Quinlivan, B., Asbeck, A.T., Wagner, D., Ranzani, T., Russo, S., Walsh, C.J.: Force
transfer characterization of a soft exosuit for gait assistance. In: A 39th Mechanical
Robotic Conference, vol. 5, pp. 1327–1334 (2015)
6. Chiaradia, D., Xiloyannis, M., Antuvan, C.W., Frisoli, A., Masia, L.: Design and
embedded control of a soft elbow exosuit. In: Proceedings of IEEE International
Conference on Soft Robotic, Livorno, Italy (2018, in press)
Realizing Soft High Torque Actuators
for Complete Assistance Wearable Robots
Allan J. Veale(B) , Kyrian Staman, and Herman van der Kooij

7522 NB Enschede, The Netherlands
a.j.veale@utwente.nl
Abstract. Wearable robots enhance the ability of their wearers to phys-

ically interact with the world, and can benefit rehabilitation efficiency,
assistive devices’ effectiveness, and ergonomic support of workers. Wear-
able robots’ ergonomics and safety can be promoted by using actuators
made of soft materials, but soft actuators in the literature are unable to
produce the high torques required for lower limb activities of daily living
(ADLs), for example, extension of the knee for sit-to-stand.
This paper presents and validates a method for realizing a soft high
torque actuator, the pleated pneumatic interference actuator, for knee
extension. It was shown to produce a torque of 150 Nm at 2 bar and an
angle of 70°. Future work will see the development of a portable pressure
source to inflate the wearable actuator.
1 Introduction
Wearable robots provide a physical interface between their human wearers and
the wearers’ environment, enhancing the wearers’ interaction with the world.
This is beneficial in devices that assist with rehabilitation of sickness and injury,
improvement of independence to people with disabilities, and ergonomic support
of industrial workers. The elements exchanging forces between wearable robots’
wearers and their world are actuators. These actuators’ inherent physical softness
is desirable for an ergonomic and safe interaction.
Of actuators with a soft structure, often made from rubbers or fabrics, those
with the highest reported maximum torques include the Pneumatic Interference
Actuator (PIA) (7.3 Nm) [1], Yap et al.’s [2] 3D printed bending bellows for
gripping (10 Nm), and Otherlab’s elbow extension bellows (30 Nm estimated)
[3]. These torque levels are insufficient for complete assistance of ADLs in the
lower limbs where up to 180 Nm [4] is required for knee extension during sit-to-
stand. In the literature, the maximum torque provided by a soft actuator for knee
extension is 4.4 Nm [5]. This shows current soft actuators are unable to provide
high torques (enough for complete assistance of sit-to-stand at the knee, for
example), and to apply those torques to a wearable device for complete human
This work is supported by the Netherlands Organization for Scientific Research
(NWO), project number 14429.
https://doi.org/10.1007/978-3-030-01887-0_8
40 A. J. Veale et al.
movement assistance. Underlying factors include low operating pressures, limited

material strength, and the difficulty of effectively transmitting high torques to
the body with a wholly soft actuator.
This paper introduces the pleated PIA (PPIA), a soft high torque actuator,
and presents a method for integrating it in a lightweight wearable device such
that it provides complete assistance of knee extension during sit-to-stand.
2 Pleated Pneumatic Interference Actuator for Knee

Extension
The actuation mechanism proposed in this work is the pleated version of Nesler
et al.’s [1] PIA, a soft inflatable tube that generates torque due to buckling as
it interferes with itself on bending.
Figure 1a shows a tubular PIA. A flexible tube, made for example of a textile
reinforced rubber, has a minimal bending stiffness when deflated, but increases
in bending stiffness as its internal pressure is increased. If constrained against a
hinged mechanism such as the knee joint shown in Fig. 1a, it buckles locally when
bent to an angle θ. This buckling corresponds to a volume change of the tube
as it interferes with itself, generating a torque τ proportional to the inflation
pressure P and volume change of the buckled region.
Fig. 1. The PIA (a) and PPIA (b) are made of a flexible tube that produces a torque
τ when inflated to a pressure P at an angle θ
The torque of the tubular PIA for a given pressure and bending angle can
be increased by introducing pleats into its straight tube, as shown in Fig. 1b.
This pleat changes its unloaded shape from a straight tube to a tube doubled
Realizing Soft High Torque Actuators for Complete Assistance 41
back over itself (U shaped). If the P and θ of a PPIA and PIA are the same, the
PPIA has a greater volume displacement compared to its unloaded shape and
therefore increased torque generation compared to a standard PIA.
2.1 Pleated Pneumatic Interference Actuator Orthosis
The realization of the PPIA for the complete assistance of knee extension is
shown in Fig. 2. The wearable device consists of the PPIA, trousers, and an
attachment system. The PPIA’s tube is made of a 65 mm diameter marine fire-
hose (Seapeace marine Equipment Co. Ltd., China) with a polyester reinforce-
ment and natural rubber liner. Each side of the PPIA tube is laced to each side
of one of the trouser legs. The lace system consists of nylon loops and dyneema
cord. Velcro behind the tube helps to keep it aligned with the trouser leg during
inflation, and padding foam on the inside of the trouser leg prevents the trousers
from twisting about the wearer’s leg. The PPIA tube’s pleat is constrained by a
dyneema cord looped about the pleat’s folds (Fig. 1b), a network of straps, and
a textile shell. These ensure the pleat stably buckles in the desired mode. If not
used, the pleat can, for example, twist to one side of the trouser leg, significantly
decreasing the torque it applies to the wearer.
Fig. 2. The PPIA orthosis, shown in extended (a) and flexed (b) states, uses a PPIA
laced to trousers to apply a knee extension torque to the leg (here an instrumented
test leg)
42 A. J. Veale et al.
3 Results and Discussion

The PPIA knee extension orthosis was characterized with the aid of an anatomi-
cally realistic sized and weighted articulated leg (Fig. 2) instrumented to measure
the PPIA’s angle and estimate its torque. The PPIA, when inflated to 2 bar, pro-
duced a torque of 150 Nm at 70° flexion (the ROM of the test leg knee joint was
70°), and had a maximum profile (perpendicular distance from the leg surface)
of 120 mm. It had a weight of 0.7 kg.
Figure 3 shows the maximum torque-angle operating point of the PPIA (more
data will be presented at WeRob2018) and the knee extension torque-angle data
from three sit-to-stand datasets [4,6,7]. It shows the PPIA is able to provide
the complete torque requirement at an angle of 70°. Compared to the maximum
profile (45 mm) and weight (0.16 kg) of Sridar et al.’s [5] soft actuator for knee
extension it is 267 % thicker and 438 % heavier, but is able to provide complete
assistance (3400 % more torque).
Fig. 3. Sit-to-stand (STS) knee extension torque-angle data, its envelope, and the
maximum torque produced by the PPIA orthosis
The PPIA orthosis could be improved by increasing its pressure (it is rated
to 10 bar), allowing for a lighter, smaller diameter PPIA tube to deliver the
same required torque with a weight and size approaching that of Sridar et al.’s
device.
4 Conclusions and Future Work

This paper presents a method for realizing a soft actuator, the PPIA, for com-
plete assistance of knee extension torques during sit-to-stand. Thus, to the our
knowledge, the PPIA was validated to generate a higher torque than other pre-
viously documented completely soft actuators.
Realizing Soft High Torque Actuators for Complete Assistance 43
One significant limitation of both the PPIA and Sridar et al.’s knee orthoses
is that they are tethered and require a portable pressure source, such as a com-
pressor or compressed gas cannister to allow for untethered operation. Hence,
future work will focus on developing an untethered pressure source for the PPIA
and optimizing its weight and size for wearability, which will be assessed on
artificial and human legs.
References
1. Nesler, C.R., Swift, T.A., Rouse, E.J.: Initial design and experimental evaluation of
a pneumatic interference actuator. Soft Robot. 5(2), 138–48 (2018)
2. Yap, H.K., Ng, H.Y., Yeow, C.-H.: High-force soft printable pneumatics for soft
robotic applications. Soft Robot. 3(3), 144–58 (2016)
3. Swift, T.: Otherlab orthotics: a fundamental jump in exoskeleton technology (2015).
https://www.youtube.com/watch?v=Ku-TavNuwwc
4. Pai, Y.-C., Rogers, M.W.: Speed variation and resultant joint torques during sit-to-
stand. Arch. Phys. Med. Rehabil. 72(11), 881–5 (1991)
5. Sridar, S., Nguyen, P.H., Zhu, M., Lam, Q.P., Polygerinos, P.: Development of a soft-
inflatable exosuit for knee rehabilitation. In: IEEE/RSJ International Conference on
Intelligence Robotic System, pp. 3722–3727, 24–28 September 2017
6. Mak, M.K.Y., Levin, O., Mizrahi, J., Hui-Chan, C.W.Y.: Joint torques during sit-
to-stand in healthy subjects and people with parkinsons disease. Clin. Biomech.
18(3), 197–206 (2003)
7. Roebroeck, M.E., Doorenbosch, C.A.M., Harlaar, J., Jacobs, R., Lankhorst, G.J.:
Biomechanics and muscular activity during sit-to-stand transfer. Clin. Biomech.
9(4), 235–44 (1994)
Application of a User-Centered
Design Approach to the Development
of XoSoft – A Lower Body Soft Exoskeleton
Valerie Power1(&), Adam de Eyto1, Bernard Hartigan1, Jesús Ortiz2,

and Leonard W. O’Sullivan1
1
Design Factors Research Group, School of Design,
Faculty of Science and Engineering, University of Limerick, Limerick, Ireland
{Valerie.Power,Adam.DeEyto,Bernard.Hartigan,
Leonard.OSullivan}@ul.ie
2
Department of Advanced Robotics, Istituto Italiano di Tecnologia,
Via Morego, 30, 16163 Genoa, Italy
Jesus.Ortiz@iit.it
Abstract. The objective of this research was to apply a user-centered design

approach to the development of a soft exoskeleton for lower limb assistance.
There has been a clear shift from hard to soft robotic exoskeletons in recent
years. Soft exoskeleton technologies typically comprise sensors and actuators
embedded in fabric/technical textiles. This approach to physical assistance offers
benefits in usability for wearers, but also presents challenges e.g. how the
concepts are put on/off and worn for long durations considering the personal
needs of the wearer. Presented is a structured three-cycle development approach
which considers user-centered design principles, but also a participatory user-
driven design-test-redesign methodology. Target users for the concept (older
adults, individuals post-stroke or incomplete spinal cord injury) were involved
in concurrent design evaluation and development throughout the design process.
1 Introduction
User-centered design (UCD) – or human-centered design – is defined in ISO 9241-210

as “an approach to interactive systems development that aims to make systems usable
and useful by focusing on the users, their needs and requirements, and by applying
human factors/ergonomics, and usability knowledge and techniques” [1]. For wearable
systems such as assistive exoskeletons, the rationale for adopting a UCD approach is
clear: to ensure that the system is effective and efficient, to promote successful uptake
and continued use, to improve user satisfaction and acceptance, to enhance user well-
being, and to mitigate safety risks and other adverse effects.
This research was completed as part of the XoSoft project, which has received funding from the
European Union’s Horizon 2020 framework programme for research and innovation under grant
agreement number 688175.

https://doi.org/10.1007/978-3-030-01887-0_9
Application of a UCD Approach to the Development of XoSoft 45
XoSoft (www.xosoft.eu) is a soft modular biomimetic lower body exoskeleton

which aims to assist people with mobility impairments (e.g. older adults, people with
stroke, incomplete spinal cord injury). UCD is a core feature of XoSoft, as clearly-
defined user groups have played key roles throughout the project to inform the design
of the system.
The current paper describes the application of UCD principles and methods to the
development of XoSoft. Findings in relation to the practical benefits and challenges of
applying a UCD approach are also discussed.
2 User-Centered Design of XoSoft
2.1 User-Centered Design Principles

A set of core principles underpin UCD [1, 2]:
1. The design is based upon a clear understanding of the requirements of the intended
users, tasks and environments;
2. Users are actively involved throughout design and development;
3. The design is driven and refined by user-centered evaluation, facilitated by early
prototyping;
4. The design process is iterative, with cyclical design, evaluation and redesign pro-
cesses repeated as often as required;
5. The design addresses the whole user experience, including parallel development of
training and support services, organizational structures etc.;
6. The design team is multidisciplinary, offering a variety of skills, expertise and
insights.
These principles are of distinct value to the design of assistive exoskeletons, as
novel, complex systems for which achieving usability and user acceptance are vital [3].
2.2 XoSoft Design Process and Methods

The principles of UCD were adopted to guide the structure and content of the XoSoft
design process. This ensured that users and their needs were the driving forces for
technology development. A variety of appropriate methods were used to achieve the
desired aims of each stage of the process: (1) planning the UCD process, (2) specifying
the context of use, (3) specifying user/organizational requirements, (4) producing
designs and prototypes, and (5) carrying out user-based evaluations [4].
Figure 1 presents an overview of the XoSoft UCD process. A three-cycle iterative
design approach was taken. A defined timeline for completion of specific prototypes (a,
b, c) and associated testing/user evaluation campaigns acted as milestones in the
project.
46 V. Power et al.
Fig. 1. Overview of the XoSoft UCD process. The prototypes developed (a, b, c) are
summarized in the left column. The nature of testing scheduled for each prototype is listed in the
right column. The timeline is presented according to project months e.g. Month 4 = M4.
At the outset of the project (Months 1–4), a cross-sectional mixed-methods study

was carried out to define prioritized user requirements for three target primary user
groups [5]:
1. Older adults who require light to moderate physical assistance to perform activities
of daily living;
2. Individuals post-stroke who would benefit from unilateral stability and assistance;
3. People with incomplete spinal cord injury who are mobile but would benefit from
physical assistance.
The study also elicited the requirements of secondary users (professional and non-
professional caregivers).
Prioritized user requirements were used to inform a design brief and technical
specifications, which were developed via collaborative multidisciplinary discussions.
This led to the first design cycle, in which the a prototype was rapidly developed using
off-the-shelf technologies via co-design workshop methods. Testing the a prototype
within the first year of the project accelerated understanding of users relative to the
potential technologies to be explored and assisted the design team in developing a
common vision and shared understanding of the concept.
Application of a UCD Approach to the Development of XoSoft 47
Integration of novel technologies and refinement of the concept in line with user
requirements was the focus of the b prototype cycles, leading to the final cycle – c
prototype development. Periodic user-centered design reviews were carried out in each
cycle, focusing on primary users for the a prototype, and expanding to include sec-
ondary and tertiary users as the concept progressed to the b and c prototypes.
3 Practical Benefits and Challenges Encountered

3.1 Benefits of Implementing UCD to XoSoft
The main benefits of adopting a UCD approach in the XoSoft project were:
1. Clear and early definition of target users and their needs facilitated early design
work and prototyping;
2. Creation of user personas and a concise design brief enhanced understanding of
users among non-clinical partners and gave focus to multidisciplinary design
activities;
3. Early and frequent user feedback aided regular refinement of the concept and timely
user input on new/altered components e.g. garment redesign.
3.2 Challenges of Implementing UCD to XoSoft

Challenges associated the UCD of XoSoft have included:
1. Ethical considerations associated with involving users in research, particularly
potentially vulnerable populations like target users of XoSoft.
2. Technological challenges associated with the integration of components with
varying Technology Readiness Levels (TRLs) into a single system e.g. high TRL
components like garment materials and inertial measurement units, together with
low TRL technologies like soft sensors.
3. Time constraints, since ideally in iterative design, the cyclical design process should
be repeated for as long as required to achieve the optimal system. With a finite
timeframe for the project, rapid progress within and between cycles was required.
4 Conclusion
A user-centered approach offers many benefits in terms of structuring and imple-

menting the design process for a soft lower body exoskeleton, as seen in the XoSoft
project.
Acknowledgment. The authors acknowledge the contributions of the XoSoft consortium: Fon-
dazione Instituto Italiano di Tecnologia (Italy), Consejo Superior de Investigaciones Científicas
(Spain), Stichting Saxion (Netherlands), University of Limerick (Ireland), Zurich University of
Applied Science (Switzerland), Roessingh Research and Development BV (Netherlands), accel-
opment AG (Switzerland), Geriatrics Centre Erlangen (Germany), Össur hf (Iceland).
48 V. Power et al.
References
1. International Organization for Standardization: ISO 9241-210:2010(en) Ergonomics of
human-system interaction – Part 210: Human-centred design for interactive systems (2010)
2. Gulliksen, J., Göransson, B., Boivie, I., Blomkvist, S., Persson, J., Cajander, Å.: Key
principles for user-centred systems design. Behav. Inf. Technol. 22, 397–409 (2003)
3. O’Sullivan, L.W., Power, V., de Eyto, A., Ortiz, J.: User centered design and usability of
bionic devices. In: 3rd International Conference on NeuroRehabilitation (ICNR2016), 18–21
October 2016, Segovia, Spain
4. Maguire, M.: Methods to support human-centred design. Int. J. Hum. Comput. Stud. 55, 587–
634 (2001)
5. Power, V., de Eyto, A., Bauer, C., Nikamp, C., Schülein, S., Müller, J., Ortiz, J., O’Sullivan,
L.W.: Exploring user requirements for a lower body soft exoskeleton to assist mobility. In:
Bai, S., Virk, G.S., Sugar, T. (eds.) Wearable Exoskeleton Systems: Design, Control and
Applications. The Institution of Engineering and Technology (2018)
Preliminary Experimental Study on Variable
Stiffness Structures Based on Textile Jamming
for Wearable Robotics
Ali Sadeghi(&), Alessio Mondini, and Barbara Mazzolai
Center for Micro BioRobotics, Istituto Italiano di Tecnologia,

Viale Rinaldo Piaggio 34, 56025 Pontedera, PI, Italy
{ali.sadeghi,barbara.mazzolai}@iit.it
Abstract. Textile based technologies can generate solutions highly adaptable

with wearable robots and devices. Textile jamming (TJ) is a stiffness modulating
technique with elaborated textile materials. The fabric with embedded miniature
and rigid segments remain flexible due to the textile substrate while they present
a high variation of stiffness (up to 17 times) in their stiff mode. The resulted TJ
packs can be assembled by traditional sewing technique to the textile garments.
1 Introduction
Exosuits, an emerging approach in the field of wearable robotics, aim to bring in the
exoskeleton and rehabilitation context the results achieved in the soft robotics field.
More comfort, lower weight, less bulky, having less mechanical stress on the user skin,
muscles and joints, are some of the features of soft exoskeletons that have motivated
the recent research activities for shifting from entirely rigid wearable robots to a softer
generation [1]. Soft wearable robots exploit soft actuators [2], soft sensors [3], and
intrinsically soft structural materials (e.g. textile) in contrast with the conventional
exoskeletons that use rigid structural materials (e.g. metals), rigid sensors and actuators
(electromagnetic motors and fluidic cylinders). This revolutionary approach has created
a growing need for continuous improvement of soft robotics technologies toward their
better performance in terms of energy efficiency, power generation, response time, life
time, simplicity of integration and fabrication, cost and etc.
Material based stiffness tuning technologies can have many potential applications
in the field of wearable robotics by tuning the structural stiffness and having multi
modal robot bodies (switching between stiff and soft modes). As example, stiffness
tunable pads at the interfacing contact to the wearer body adapts to the skin in the soft
mode and provides accurate motion transmission by the robot in its stiff mode [4] or
stiffness variable joints can assist the user by decreasing the load amount on the wearer
muscles and joints [5]. A soft engaging and disengaging solution for storing and
This work has received funding from the European Union’s Horizon 2020 framework programme
for research and innovation under grant agreement No. 688175.

https://doi.org/10.1007/978-3-030-01887-0_10
50 A. Sadeghi et al.
releasing of elastic energy will be useful in realizing entirely soft unpowered

exoskeletons that traditionally require rigid clutching systems [6]. The soft exosuit
presented in [7] demonstrates the feasibility of a soft unpowered exoskeletons but still
it uses rigid metallic clutch in its structure. Among different stiffness tuning tech-
nologies, granular jamming and layer jamming mechanisms are widely exploited in
many different areas of soft robotics, such as adaptive grasping [8], medical applica-
tions [9] including wearable robotics [4, 5] due to their simplicity of implementation
and relatively high variation of stiffness. In this work we investigate and study the
feasibility of realizing the layer jamming mechanism, entirely with textile material
(membrane and internal layers) as highly compatible material with wearable robots and
exosuits. Here we present the preliminary results of our experimental approach.
2 Textile Jamming
2.1 Concept
Use of jamming to change structural stiffness is not limited to the granular medium
(granular jamming) [8] or medium of flexible layers (e.g. paper) in layer jamming [4].
In fact, a similar principle can be applied to extremely soft textile materials. A clear
example of that principle is given by the vacuum bags commonly used for garments.
A bag of soft clothes subject to a vacuum becomes a stiff block. This is due to the
reduction of bag volume and consequently decrease of frictional interaction among
textile fibers compressed together by the external environmental pressure. We named
this technique “Textile Jamming” (TJ), with the objective of developing textile-based
materials able to modulate their stiffness using a vacuum. In this concept a package of
textiles with particular yarns, weaving orientations or patterned textures should present
high softness and flexibility in normal conditions (no vacuum applied), and the
interaction of these internal patterns could result in a stiff structure when subject to
negative pressures. Elaborating the surface of textile layers with array of miniature and
rigid structures (Fig. 1) improves the frictional interaction between layers. Moreover,
during the vacuum, it provides the possibility of mechanical interlocking between the
layers. This elaboration does not affect significantly the flexibility of textile (similarly
to the assembly of rigid beads and buttons on the clothes), since the textile in between
the rigid structure remains flexible. The softness of textile and interlocking of rigid
segments also permit achieving a stiff structure even after sharp folding (Fig. 1). The
same result is not easily achievable in the layer jamming by flexible sheets as having
sharp folds is limited by the minimum bending radius of the flexible sheets.
2.2 Prototyping and Experiments

In order to quantitatively compare the performance of TJ solution with traditional
materials, we fabricated the samples of different materials in 50 50 mm2 square
packages. An air sealed cotton fabric coated with silicone (Ecoflex 0030 of Smooth-On
Co.) was used to encapsulate the internal layers. Coffee, layers of transparent PET
sheets, and woven cotton fabric elaborated by matrix of plastic beads were used as
Preliminary Experimental Study on Variable Stiffness Structures Based on TJ 51
Fig. 1. The textile jamming (TJ) concept; the package changes the stiffness due to the frictional
and interlocking interaction of miniature beads integrated on the textile under vacuum.
internal materials. Coffee was selected as reference material since it is well exploited in
previous works. In this sample 10 grams of coffee grains were packed in air sealed
cotton bag (resulted thickness was the same of TJ samples). Two series of TJ samples
with two and four layers, each one with three different size of octagonal beads (di-
ameter height of 2 1 mm – Small Beads -, 4 1.5 mm – Medium Beads - and
8 2 mm – Large Beads) were realized. In both series the internal side of textile
packaging was elaborated by the assembly of textile beads. One of these series con-
tained one more layer in the middle with symmetric sides elaborated by the rigid beads
(Fig. 1). The sample of conventional layer jamming was realized by four layers of PET
film with 0.2 mm thickness.
Fig. 2. Top: measured forces of samples for 10 mm, bottom: the percentage of stiffness
variation under different vacuum pressure
52 A. Sadeghi et al.
Following the ASTM D790 protocol for material flexural stiffness measurement a
mechanical testing instrument (Roell Z005 by Zwick) was used to characterize the
deformation force of the samples at different negative pressures (No vacuum; −0.2
Atm, −0.4 Atm, −0.6 Atm) for 10 mm deflection at constant displacement speed of
10 mm/min. A summary of the experimental results is depicted in Fig. 2. As it was
expected in general the TJ samples with elaborated textile presented a better variation
of stiffness. The result of TJ with largest beads were always better than samples with
coffee. While the results of PET layers were always lower than all the other samples.
The maximum deflection forces and stiffness variations respect to their passive mode
were measured for the TJ sample with large beads 93 N and 17 times, for coffee sample
72 N and 9 times and finally for layers of PET were 13 N and 2 times. Due to the
flexibility of textile, the deflection force of all the TJ samples under no vacuum was less
than coffee sample which means their higher flexibility in their passive mode.
3 Conclusions
In this work we experimentally evaluated the feasibility of realizing layer jamming with
entirely textile based materials. We recorded higher variation of stiffness in our TJ
samples elaborated with octagonal beads of 8 mm diameter and 2 mm height respect to
the samples made by coffee or PET films. The TJ samples are easily integrative with
textile garments by any technique such as sewing and in any configuration such as
vertically while the samples with coffee granules always have to face with the issue of
gravity and migration of particle inside the package.
References
1. Asbeck, A.T., et al.: A biologically inspired soft exosuit for walking assistance. Int. J. Robot.
Res. 34(6), 744–762 (2015)
2. Mosadegh, B., et al.: Pneumatic networks for soft robotics that actuate rapidly. Adv. Funct.
Mater. 24(15), 2163–2170 (2014)
3. Totaro, M., et al.: Soft Smart garments for lower limb joint position analysis. Sensors 17(10),
2314 (2017)
4. Bureau, M., et al.: Variable stiffness structure for limb attachment. In: 2011 IEEE
International Conference on Rehabilitation Robotics (ICORR). IEEE (2011)
5. Hauser, S., et al.: Jammjoint: a variable stiffness device based on granular jamming for
wearable joint support. IEEE Robot. Autom. Lett. 2(2), 849–855 (2017)
6. Collins, S.H., Wiggin, M.B., Sawicki, G.S.: Reducing the energy cost of human walking
using an unpowered exoskeleton. Nature 522(7555), 212 (2015)
7. Poliero, T.: Soft wearable device for lower limb assistance: assemsment of an optimized
energy efficient actuation prototype. In: Soft Robotics 2018, Livorno (2018)
8. Brown, E., et al.: Universal robotic gripper based on the jamming of granular material. Proc.
Natl. Acad. Sci. 107(44), 18809–18814 (2010)
9. Ranzani, T., et al.: A bioinspired soft manipulator for minimally invasive surgery.
Bioinspiration Biomimetrics 10(3), 035008 (2015)
Towards Embroidered Sensing Technologies
for a Lower Limb Soft Exoskeleton
M. Totaro1(&), E. Bottenberg2, R. Groeneveld2, L. Erkens2,

A. Mondini1, G. J. Brinks2, and L. Beccai1
1
Center of Micro-BioRobotics, Istituto Italiano di Tecnologia,
Pontedera, PI, Italy
massimo.totaro@iit.it
2
Smart Functional Materials Research Group,
Saxion University of Applied Sciences, Enschede, The Netherlands
e.bottenberg@saxion.nl
Abstract. In this work we describe the developments of soft mechanical

sensing technologies for XoSoft, a soft lower-limb exoskeleton for assisting
people with low to moderate levels or reduced mobility. Starting from the results
obtained in integrating capacitive strain sensors in standalone knee and ankle
modules, we embed the capacitance constituent electrodes directly in the gar-
ment fabric, still guaranteeing an electrically shielded design. Embroidery of
conductive yarns with F-head and W-head were tested, varying several
parameters, such as stitch length and distance, presser foot height, and geometry
(zig-zag patterns). Based on these techniques, a novel sensorized kneepad was
fabricated. The adopted solutions make the sensing system more robust,
improve both wearability and user comfort, and are expected to allow long term
monitoring of joint movements.
1 Introduction
In the last decade, several exoskeletons [1, 2] have been developed both by academic
and industrial research. Recently, soft exoskeletons are being investigated [3, 4],
exploiting results and technologies of soft robotics, in particular for sensing, actuation
and materials.
XoSoft [5] is a modular soft lower-limb exoskeleton to assist people with walking
impairments. It consists of ankle, knee and hip modules, which can be used individ-
ually or in combination based on the user’s requirements. As a medical device, XoSoft
will assist people with low to moderate levels of reduced mobility to improve health
and quality of life. Primary users are mainly elderly people, individuals with mobility
impairments, and stroke patients. The overall system encompasses a central processing
hub, variable stiffness joints, multi-joint actuation, smart soft mechanical sensors, IMU
units.
This work has received funding from the European Union’s Horizon 2020 framework programme
for research and innovation under grant agreement No. 688175.

https://doi.org/10.1007/978-3-030-01887-0_11
54 M. Totaro et al.
We addressed the design and development of the wearable soft sensors for the
monitoring of knee and ankle joint movement. In a first phase separate sensorized
modules have been developed [6]. The sensing system is based on capacitive strain
sensors, built from integrating stretchable conductive textile (Electrolycra®, Holland
Shielding Systems BV, Dordrecht, NL) and silicone elastomer (Ecoflex® 00–10,
Smooth-On Inc., Macungie, PA, USA). A three-electrode configuration (bottom
ground, central, and top ground) is used to shield the sensors from proximity effects
and from electromagnetic noise.
In particular, for the knee, 1 Degree of Freedom (DoF), namely the
flexion/extension angle, is measured, while for the ankle, up to 3 DoFs are monitored.
As shown in Fig. 1, the kneepad integrates three strain sensors and a wireless custom-
made electronic board. The redundancy improves the correlation with the angle and lets
to discriminate accidental contacts with external objects during the movement.
Regarding the anklepad, shown in Fig. 1, five strain sensors have been integrated:
three in the front and two in the back side, respectively. In this way, by combining
properly the sensor outputs, different DoFs can be discriminated and measured. Some
basic movements (plantar flexion/extension, adduction/abduction, rotation) have been
monitored and characterized in a first study [6].
Fig. 1. Sensorized kneepad and anklepad, developed during the first phase of XoSoft project.
The kneepad (left) integrates three textile-based capacitive strain sensors; and, the anklepad
(right) has five sensing elements, three on the front and two on the back side, respectively.

2.1 Embroidered Knee Sensor
In order to improve performances and reliability of the sensors, textile electrodes
should be part of the XoSoft garment and we have experimented different techniques
for embroidering the bottom ground electrodes on the garment. This is the base on top
of which the other layers of the each sensor can be integrated.
The machine used for embroidering with F-Head and W-Head is a ZSK JCZA
0109-550. As conductive material, a silver coated yarn (235/36, Shieldex, Palymira,
NY, USA) with resistivity <100 X/m has been used. The F-head was used to test if it
was possible to stitch the embroidered pattern using directly the conductive yarn.
Towards Embroidered Sensing Technologies for a Lower Limb Soft Exoskeleton 55
However, because of the stretchability of the substrate this process did not work
properly. The main issue with the F-head was that the conductive yarns extended also
on the backside. Since this is in contact with the skin, it should be electrically insulated.
Then, W-Head was used. During embroidering, a self-sticking water-soluble polyester
yarn was employed to prevent the textile from stretching. In the case of W-Head,
different parameters were tested, like stitch length, stroke pantograph distances, and
geometry (zigzag designs). Moreover, the conductive yarn was not directly embroi-
dered on the textile. Instead, it was laid on the garment and then stitched on using
polyester filament.
3 Results
First, the electrode was patterned on a non-stretchable textile. The stitch length is
3.6 mm, the stitch distance 1 mm, stroke zig-zag with ten increments (one increment
step corresponds to 0.1 mm), and stroke pantograph with 80 increments. In this way,
the embroidery looked irregular, probably caused by the relative large stitch distance.
Then, electrodes were embroidered on a stretchable textile with a smaller stitch length
of 1 mm. The material was still stretchable after washing off the water-soluble support
yarn. The pattern was sharper, but some imperfections were still present, especially if
the distance between the 180° turns became smaller. To improve this aspect, the
movement of the pantograph (frame) has been set larger. In this way, the main region
of the pattern was embroidered without visible problems. Only in the narrower part
Fig. 2. (A) Textile electrodes patterned using regular embroidering with F-Head. (B) Optimized
electrode embroidering of silver coated yarn with W-Head. (C) Sensorized kneepad with 3
sensors having: bottom ground electrode embroidered in the garment; Electrolycra central
electrode; and top ground electrode embroidered in elastic textile. Electrical connections to read-
out board are made by micro coaxial cables.
56 M. Totaro et al.
(i.e. the connection pattern of the electrodes, perpendicular to the main geometry) the
turns were too small. To avoid this, in the connection regions the stitches were set
perpendicular to those of the main region. In Fig. 2B the patterned electrodes using this
technique are shown. In the final version (Fig. 2B) the conductive yarn was embroi-
dered in a compact smooth electrode showing a moderate stretchability with respect to
the hosting textile but still having an elastic behavior.
Exploiting these results, a novel kneepad, shown in Fig. 2C, was developed. The
layout was unaltered with respect to the previous version having three sensors [6].
However, in this case for each sensor the bottom ground electrode was directly
embroidered in the kneepad garment. The central electrode was made of Electrolycra,
with a layer of silicone elastomer used as dielectric material. Then, the top ground
electrode were embroidered in a stretchable textile band. The latter were fixed to the
underneath layers by silicone elastomer (which act as dielectric material) in the active
sensing area, while they are sewn at the edges to the substrate garment. Also, since the
read-out electronic board is located in the hip region of the exoskeleton, making robust
and yet reliable and unobtrusive connections is a challenging task for capacitive sen-
sors. Indeed, parasitic elements and electromagnetic noise should be avoided. In this
case, micro coaxial cable (40 AWG Micro Coax, Temp-Flex, South Grafton, MA,
USA) with outer diameter of 350 µm and core of 200 µm. The final device results
having a high robustness regarding the integration in the textile and the recorded
outputs are unaffected by external noise or parasitic capacitances. A complete char-
acterization of the system is being currently performed in human subjects.
4 Conclusions
Embroidering techniques developed during the XoSoft project allow to overcome some
issues observed in the first generation of the sensing system. In particular, the
robustness is improved relevantly, since previously the sensing elements were fixed to
the garment by means of stretchable elastomers. In that case, with very high strain,
partial detachment of single sensors could occur, especially at their edges, where higher
stress is concentrated. This limited the usability of the modules for long term moni-
toring. Otherwise, by embroidering the electrodes, sensing elements are part of the
garment itself, avoiding damages after high strain but still guaranteeing shielding
through the three electrode configuration. In addition, both comfort and wearability can
be improved relevantly.
References
1. Hill, D., Holloway, C.S., Morgado Ramirez, D.Z., Smitham, P., Pappas, Y.: What are user
perspectives of exoskeleton technology? A literature review. Int. J. Technol. Assess. Health
Care 33(2), 160–167 (2017)
using an unpowered exoskeleton. Nature 522, 212–215 (2015)
Towards Embroidered Sensing Technologies for a Lower Limb Soft Exoskeleton 57
3. Yap, H.K., et al.: A soft exoskeleton for hand assistive and rehabilitation application using
pneumatic actuators with variable stiffness. 2015 IEEE International Conference on Robotics
and Automation (ICRA). IEEE (2015)
4. Panizzolo, F.A., et al.: A biologically-inspired multi-joint soft exosuit that can reduce the
energy cost of loaded walking. J. Neuroeng. Rehabil. 13 (2016). Art. no. 43
5. www.xosoft.eu
6. Totaro, M., et al.: Soft smart garments for lower limb joint position analysis. Sensors 17(10)
(2017). Art. no. 2314
Recent Results from Evaluation of Soft
Wearable Robots in Clinical Populations
Conor Walsh(&)
School of Engineering and Applied Science and Core Faculty with the Wyss
Institute, Harvard University, Cambridge, USA
walsh@seas.harvard.edu
Abstract. We have been on developing new approaches to design, manufac-

ture, and control soft wearable robotic devices and characterizing their perfor-
mance through biomechanical and physiological studies so as to further the
scientific understanding of how humans interact with such machines. Example
application areas include enhancing the mobility of healthy individuals such as
soldiers walking with a heavy load, restoring the mobility of patients with gait
deficits such as those poststroke, and assisting those with upper extremity
weakness to perform activities of daily living such as patients with a spinal cord
injury. This abstract summarizes recent results from evaluation of these devices
in clinical populations.
1 Introduction
Our long term vision is for ubiquitous soft wearable robots that can be worn all day,
every day underneath clothing, in both the workplace, community, and home envi-
ronments (Fig. 1). This would enable a change in the paradigm of how we currently
assist healthy individuals with physical tasks (i.e. time off to recover from muscu-
loskeletal injury), those with physical impairments with mobility (i.e. canes, plastic
braces and walkers), and how we rehabilitate those after injury (i.e. a limited number of
therapy sessions in a clinic environment). Our efforts center around understanding how
we can leverage the mechanical and electrical properties of textile materials to create a
new class of soft wearable robots that are lightweight and flexible (i.e. non-restrictive).
This includes integrated actuation and sensing, as well as control algorithms to ensure
synergistic assistance delivery. Our view is that these systems will not replace tradi-
tional rigid exoskeletons (that are emerging as a valuable clinical tool for those with
severe impairments) but will instead offer complementary capabilities for those who
would not benefit from rigid exoskeletons.
2 Restoring Poststroke Gait
Preliminary work by our team has demonstrated that applying well-timed, small
assistive forces to the paretic limb of stroke survivors during walking can impart
substantial biomechanical, energetic, and cognitive benefits to users as shown in Fig. 2
[1–3]. In [1], we showed the effect of the exosuit actively assisting the paretic limb of

https://doi.org/10.1007/978-3-030-01887-0_12
Recent Results from Evaluation of Soft Wearable Robots in Clinical Populations 59
Fig. 1. Vision for soft wearable robots for upper and lower limb.
individuals in nine individuals in the chronic phase of stroke recovery during treadmill
and overground walking. The level of assistance applied was relatively low (*12% of
biological joint torques], yet the exosuit assistance was able to facilitate an immediate
5.33° increase in the paretic ankle’s swing phase dorsiflexion and 11% increase in the
paretic limb’s generation of forward propulsion. These improvements in paretic limb
function contributed to a 20% reduction in forward propulsion interlimb asymmetry
and a 10% reduction in the energy cost of walking, compared to walking with the
exosuit unpowered, which is equivalent to a 32% reduction in the metabolic burden
associated with poststroke walking. In [2], we demonstrated that the same soft exosuit
targeting the paretic ankle could reduce common poststroke gait compensations.
Specifically, compared to walking with the exosuit unpowered, walking with the
exosuit powered resulted in significant reductions in hip hiking (27%) and circum-
duction (20%).
In order to better understand the biomechanical mechanisms underlying exosuit-
induced reductions in metabolic power, we evaluated the relationships between
exosuit-induced changes in the body center of mass (COM) power generated by each
limb, individual joint powers, and metabolic power [3]. Compared to walking with an
exosuit unpowered, we found that exosuit assistance produced more symmetrical COM
power generation during the critical period of the step-to-step transition (22.4 ± 6.4%
more symmetric). In addition, changes in individual limb COM power were related to
changes in paretic (R2 = 0.83, P = 0.004) and nonparetic (R2 = 0.73, P = 0.014)
ankle power.
60 C. Walsh
Fig. 2. Summary of biomechanical and physiological results with the soft exosuit in patients
poststroke from [1].
3 Restoring Grasp After Spinal Cord Injury
Beyond assisting locomotion, our team is also focused on developing inflatable, textile-
based soft wearable robots for assisting those with upper extremity weakness to be able
to achieve common activities of daily living (ADLs). Our efforts on this line of research
are currently focused on those who have suffered spinal cord injury and are left with
various degrees of impairment depending on the level of injury. Our initial efforts
focused on a soft robotic glove for restoring grasping ability using elastomeric soft
actuators [4]. More recently, we have been leveraging the unique properties of textiles
in creating fluid-powered soft wearable robotics. The advantages of textile-based soft
robots are that they can be very lightweight, easily integrated into apparel, and
mechanically transparent when unpowered. Here we summarize some of our initial
work in this area on a textile soft robotic glove [5, 6] and also recent work on a soft
wearable textile robot to assist the shoulder [7]. Our goal with this work is to create a
full upper extremity garment with arrays of textile-based inflatable actuators that can be
selectively pressurized with embedded textile strain and pressure sensors to monitor the
state of the actuators and the underlying limb.
Recent Results from Evaluation of Soft Wearable Robots in Clinical Populations 61
Our initial efforts in evaluating the soft robotic glove have focused on those who
have suffered a spinal cord injury, where we evaluated the grasping function of nine
participants with C4–C7 spinal cord injury without and with assistance provided by a
soft robotic glove through experimental sessions involving the administration of a
clinical motor function test [6]. The test we used for evaluation was the Toronto
Rehabilitation Institute Hand Function Test that comprises both manipulation of
objects encountered during ADLs and measures of grip strength and we compared two
conditions (wearing the glove powered and not wearing it). Our results showed that the
soft robotic glove had an immediate effect on key hand functions assessed during
assisted manipulation of ADL objects. The difference of the mean score between the
baseline and assisted condition was significant across all subjects and objects of the
test. An improvement of 32.45 ± 14.41% relative to the maximal test score indicates
that the glove sufficiently enhances hand function to independently perform various
ADLs. Moreover, the grip strength was also increased when using the assisting soft
robotic glove, further demonstrating the effectiveness of the device in assisting grasps.
It was also interesting to see that individuals with particularly low baseline scores (due
to higher levels of injury, C4–C5) particularly benefitted from the soft robotic glove,
with an improvement of over 50% (Fig. 3).
Fig. 3. Soft robotic glove for hand function restoration: A. Study subject during unassisted
object manipulation using a passive tendinosis grasp to lift an object (baseline condition) B.
Study subject performing an active palmar grasp to manipulate an object using the soft robotic
glove (assisted condition). C. Soft robotic glove assisting palmar grasp D. Soft robotic glove
assisting pinch grip E. Soft robotic glove assisting the grasp of a wooden block to assess grip
strength F. Improvement in mean TRI-HFT score across all subjects (N = 9, mean
difference = 2.34, 95% confidence interval from 1.6715 to 3.0069, p < 0.000001)
62 C. Walsh
Acknowledgment. This research is the result of a multidisciplinary team from Harvard

University, Boston University and Spaulding Rehabilitation Hospital with support from the
National Science Foundation (NSF #1446464 and #1454472), the National Institutes of Health
(BRG R01HD088619), Harvard SEAS and Wyss Institute.
References
1. Awad, L., Bae, J., O’Donnell, K., De Rossi, S., Hendron, K., Sloot, L., Kudzia, P., Holt, K.,
Ellis, T., Walsh, C.: Soft wearable robots improve walking function and economy after stroke.
Sci. Transl. Med. 9(400), eaai9084 (2017)
2. Awad, L., Bae, J., Kudzia, P., O’Donnell, K., De Rossi, S., Hendron, K., Holt, K., Ellis, T.,
Walsh, C.: Reducing poststroke gait compensations through targeted assistance of paretic
ankle function using a soft wearable exosuit. Am. J. Phys. Med. Rehabil. (AJPMR) 96(10),
S157–S164 (2017)
3. Bae, J., Awad, L., Long, A., O’Donnell, K., Hendron, K., Holt, K., Ellis, T., Walsh, C.:
Biomechanical mechanisms underlying exosuit-induced improvements in walking economy
after stroke. J. Exp. Biol
4. Polygerinos, P., Wang, Z., Galloway, K.C., Wood, R.J., Walsh, C.J.: Soft robotic glove for
combined assistance and at-home rehabilitation. Robot. Auton. Syst. (RAS) Spec. Issue
Wearable Robot. 73, 135–143 (2015)
5. Cappello, L., Galloway, K., Sanan, S., Wagner, D., Granberry, R., Engelhardt, S., Haufe, F.,
Peisner, J., Walsh, C.: Exploiting textile mechanical anisotropy for fabric-based pneumatic
actuators. Soft Rob. (2018, in press)
6. Cappello, L., Meyer, J., Galloway, K., Peisner, J., Granberry, R., Wagner, D., Engelhardt, S.,
Paganoni, S., Walsh, C.: Assisting hand function after spinal cord injury with a fabric-based
soft robotic glove. J. Neuro Eng. Rehabil. (2018, in press)
7. O’Neill, C., Phipps, N., Cappello, L., Paganoni, S., Walsh, C.: Soft robotic shoulder support:
design, characterization, and preliminary testing. In: 15th International Conference on
Rehabilitation Robotics (ICORR), London, July 2017
Subject-Centered Based Approaches
for Controlling Wearable Robots
Toward an Affordable Multi-Modal
Motion Capture System Framework
for Human Kinematics
and Kinetics Assessment
Randa Mallat1,2(B) , Vincent Bonnet1 , Mohamad Khalil2 ,

and Samer Mohammed1
1
LISSI Laboratory, University of Paris-Est-Créteil, Créteil, France
mallatranda@gmail.com, vincent.bonnet@u-pec.fr
2
Faculty of Engineering and Azm Center, Lebanese University, Beirut, Lebanon
Abstract. The present study aims at designing and evaluating a low-

cost, simple and portable system for human kinematics (joint angles)
and kinetics (joint torques) motion assessment. The system is based on a
single camera, a set of customized markers, low-cost inertial measurement
units and an affordable Wii Balance board. The automatically detected
and tracked marker positions and orientations were used synchronously
with other sensors data as inputs to an Extended Kalman Filter based
on a biomechanical model of the investigated tasks. The method was
validated with arm motions and with lower-limb motions of a subject
wearing an exoskeleton. External ground reaction forces measured with
a Wii Balance board were compared to the estimated ones based on the
proposed low-cost system. Comparative analysis shows good accuracy
with a low average NRMS error.
1 Introduction
Human motion analysis starts with capturing accurately kinematic variables

such as the joint angles. Quantitative kinematic data are useful in a wide range
of applications including rehabilitation and motion assistance. Usually, accurate
kinematic measurements are performed using costly and restricted to labora-
tory environments stereophotogrammetric systems. Recently, the use of small,
low-cost and light weight Inertial Measurements Units (IMU) has been widely
promoted against traditional camera based systems. An IMU might incorporate
3D accelerometers, gyroscopes and magnetometers. However, magnetometers are
usually avoided due to magnetic field distortions in daily basis environments.
Moreover, the integration of IMU data is prone to a large nonlinear and time-
dependent drift [1].
Low cost RGB camera based systems have been also investigated as an afford-
able and portable tool to track the human movements [2]. Indeed, they do not
suffer from drift error. In a previous work [3], our group has proposed to use an
https://doi.org/10.1007/978-3-030-01887-0_13
66 R. Mallat et al.
Fig. 1. (a) Arm and lower limb mechanical models. (b) Extended Kalman Filter algo-
rithm. (c) The affordable involved equipments. (d) Experimental tests on human arm
and lower limbs with and without wearing an exoskeleton. For the arm test each marker
was mounted on an IMU sensor.
Fig. 2. (a) Reconstructed elbow flexion/extension with the EKF, (b) Forearm’s marker
position, Forearm’s IMU linear accelerations and angular velocities along x, y and z
axis measured (red) vs estimated (black) with the EKF.
Toward an Affordable Multi-Modal Motion Capture System Framework 67
Fig. 3. Vertical ground reaction force and moments of the human while wearing the
exoskeleton measured (red) based on the proposed motion capture system and their
estimate (black) from the identification model.
Augmented Reality (AR) detector to estimate the position and orientation of

human segments. However, camera based systems are subject to markers occlu-
sions. In this context, we propose a new low cost and user friendly motion capture
system based on a camera and IMUs. Both systems data will be fused into an
Extended Kalman Filter (EKF). Multi-modality allows to reduce the noise effect,
specially the IMU drift, and handle markers occlusions. The method has been
validated experimentally for arm motions as well as with a subject wearing a full
lower limbs exoskeleton. In the latter, the resultant estimated kinematics were
used together with Ground Reaction Forces and Moments (GRFM) recorded
with an affordable Wii Balance Board (WBB) to show the possibility of iden-
tifying the body segments inertial parameters of a human-exoskeleton system
using the proposed low-cost motion capture system.
2 Method
2.1 Mechanical Model
Two mechanical models were used to assess arm and lower limbs motions (see
Table in Fig. 1.a). Each sensor position, orientation, angular velocity and accel-
eration were modeled by computing the Forward Kinematic Model (FKM) and
its first and second derivatives. The 3D Cartesian orientation was converted into
quaternion to avoid gimbal lock.
2.2 Measurement System
As described by Fig. 1, arm and lower limb kinematics were estimated based on
data gathered synchronously from a set of IMU and an RGB camera detecting
the pose of customized ARUCO markers located on each segment [4]. Moreover
each marker was mounted on an IMU that provides 3D linear accelerations and
68 R. Mallat et al.
3D angular velocities. A wand-based calibration method [3] was used priorly to

determine joints center positions, segments lengths and local pose of the markers
in their corresponding segments frames.
Unfortunately, as stated before, each sensor measurements is tainted with
noise, drift or occlusion for the visual information. Therefore, all data were fused
into an EKF (Fig. 1.b) to estimate joint kinematics while compensating for each
of the measurements inaccuracies. The proposed EKF aims at estimating joint
angles, velocities and accelerations gathered into one state vector xk at each
time step k. A classical constant acceleration model governs the state evolution.
3 Experimental Validation
The proposed approach was validated experimentally with both upper and lower
limbs models (Fig. 1.d). Preliminarily, the subject was asked to perform simple
elbow flexion/extension movements. Figure 2 shows the tracking of the mea-
sured markers pose, IMU linear accelerations and angular velocities respectively
with the measurement model as well as the reconstruction of the elbow flex-
ion/extension angle with the EKF.
As shown in our previous work [3], the proposed low-cost system thanks to
its portability can be used with an exoskeleton. In this experiment we assume
that the exoskeleton and the subject are rigidly attached. The estimated joint
kinematics are used to perform a dynamic identification process. Figure 3 shows
the GRFM comparison between measured from the WBB and estimated from
the proposed system during a cross validation trial. The NRMS difference was
lower than 6%.
4 Conclusion
This paper shows the possibility of fusing IMU sensors with visual markers in
an affordable and transportable motion capture system. The proposed system
is likely to improve kinematics estimates and to track more complex motions
compared to other only visual systems or IMU-based separately. It has been val-
idated experimentally for the dynamic identification of human-exoskeleton lower
limbs as well as with upper limbs model. However it needs more experimental
validations and a real comparison with a reference stereophotogrammetric sys-
tem. Nevertheless, such system, with an overall estimated price of 100 Euros,
will be without doubt a promising tool to support home rehabilitation processes.
Toward an Affordable Multi-Modal Motion Capture System Framework 69
References
1. El-Gohary, M., McNames, J.: Human joint angle estimation with inertial sensors and
validation with a robot arm. IEEE Trans. Biomed. Eng. 62(7), 1759–1767 (2015)
2. Bonnet, V., Sylla, N., Cherubini, A., Gonzàlez, A., Azevedo, C., Fraisse, P., Venture,
G.: Toward an affordable and user-friendly visual motion capture system. IEEE
International Conference on Engineering in Medicine and Biology Society (EMBS),
pp. 3634–3637 (2014)
3. Mallat, R., Bonnet, V., Huo, W., Karasinski, P., Amirat, Y., Khalil, M., Mohammed,
S.: An introduction to signal detection and estimation. In: Proceedings of the
IEEE/RSJ International Conference Robotics Automation (ICRA) (2018)
4. Munoz-Salinas, R.: ARUCO: a minimal library for Augmented Reality applications
based on OpenCv. Universidad de Crdoba (2012)
High Power Series Elastic Actuator
Development for Torque-Controlled
Exoskeletons
Mehmet C. Yildirim, Ahmet Talha Kansizoglu, Polat Sendur,

and Barkan Ugurlu(B)
Department of Mechanical Engineering, Ozyegin University, 34794 Istanbul, Turkey

barkanu@ieee.org
Abstract. This paper presents the development procedures of a high

power series elastic actuator that can be used in torque-controlled
exoskeleton applications as a high-fidelity torque source. In order to pro-
vide a high torque output while containing its weight, the main objective
was to satisfy dimensional and weight requirements within a compact
structure. A three-fold design approach was implemented: (i) The tor-
sional spring was designed using finite element analyses and its stiffness
profile was experimentally tested via a torsional test machine, (ii) ther-
mal behavior of the actuator was experimentally examined to ensure
sufficient heat dissipation, (iii) the fatigue life of the spring was com-
puted to be 9.5 years. Having manufactured the actuator, preliminary
torque-control experiments were conducted. As the result, a high-fidelity
torque control was achieved with a control bandwidth of up to 12 Hz.
1 Introduction
The potential benefits of the compliance, e.g., actuators with low mechanical
output impedance, can be exploited in the form of high fidelity torque control.
These force-controlled compliant robots enable enhanced interaction capabili-
ties and are inherently safe for human-robot interaction. Furthermore, the series
elasticity emulates the series tendons in muscular structures, and therefore, it
is highly efficient in terms of mechanical energetics [1]. In the light of these
facts, SEAs (Series Elastic Actuator) appear to be a very conceivable choice to
power torque-controlled exoskeleton systems when considering safety, depend-
ability and sustainable physical human-robot interaction. With this motivation,
we designed a SEA unit with a high torque-to-weight ratio. Despite the conven-
tional approach that is based on basic finite element analysis, we propose an
integrated approach that combines meticulous mechanical design, particularly
for the torsional spring, fatigue characteristics and the thermal behavior of the
unit [2].
This research was supported by the Scientific and Technological Research Council
of Turkey (TUBITAK) with the project 215E138.
https://doi.org/10.1007/978-3-030-01887-0_14
High Power SEA Development for Torque-Controlled Exoskeletons 71
This paper aims to report the further refinements and experiments over our
previous work presented in [2]. Section 2 explains the hardware design and torque
control, together with the results. Section 3 concludes the paper.
2 Methods
2.1 Hardware Design
(1) Mechanical Design: The SEA unit is comprised of a frameless brushless motor
(Kollmorgen, TBM-7631/7615), a strain wave gear (Harmonic Drive CSG-25,
1:100) and a 2-spoke torsional spring that acts as a torque sensor; see Fig. 1. Two
23-bit encoders (Avago Tech., AS38-H39E-B13S) were integrated to measure the
motor angle and the torsional deflection. Technical specifications of two units
with different torque output capacities are provided in Table 1. The torque-
to-weight ratios of the units are 52 and 33 Nm/kg, which are comparable to
state-of-the-art SEAs; for instance, see [3].
Table 1. CoEx-SEA actuator specifications
Specification Unit Unit-A Unit-B

Max. angular velocity RPM 26.33 44.09
Max. continuous torque Nm 164 94
Peak torque Nm 460 304
Weight kg 3.15 2.85
Dimensions (r × L) mm × mm 53.25 × 152.3 53.25 × 125
Stiffness Nm/deg 91 91
Torque resolution mNm 3.9 3.9
(2) Torsional Spring: The conventional torsional spring design is based on

empirical topology search that is followed by FEA simulations. Our past expe-
riences showed that this approach is prone to unexpected material behaviour;
e.g., instead of spring spokes, the screw holes may dominantly effect the spring
stiffness. Moreover, it may not be possible to determine the elastic/plastic region
boundaries.
To this end, we backed up finite element analyses with spring break-up exper-
iments via a torsional test machine. In these experiments, the torsional spring
specimens were subject to gradually increasing torques until they completely
break-up. In doing so, the elastic region and plastic regions boundaries were
clearly defined; we were able to validate that the spring will behave linearly
within the torsional deflection of 0◦ –2◦ . Moreover, the experimental value of
the stiffness matched well with our finite element simulations, adequately con-
firming the proposed spring topology. Figure 2(a)–(b) depict the experimentally
obtained torque-angle curves and a spring sample before and after the break-up.
72 M. C. Yildirim et al.
Fig. 1. (A) CAD drawing of the SEA unit. (B) The torsional spring mounted at the
output. (C) The actual SEA unit for the experiments.
Furthermore, the fatigue life of the springs are not chiefly investigated in the
literature. Using the actual hip joint torque data from exoskeleton-supported
able-bodied walking experiments reported in [4], we determine the fatigue life of
the spring via Palmgreen-Miner cumulative fatigue damage theory. The damage
values under this specific loading is determined to be 2e−7 ; see Fig. 2(c). This
damage value indicates that the loading could be applied for 5e6 cycles that
correspond to 83333 h (approx. 9.5 years) of continuous operation if the spring
material is AL-7075.
(3) Thermal Management: Another important characteristics of the mechan-
ical design is to see how the system emits the heat generated by the motor. To
that end, we use the actual hip joint torque data from exoskeleton-supported
able-bodied walking experiments reported in [4] and observed the thermal behav-
ior experimentally via a thermocouple and an IR camera. The maximum heat
accumulated during this experiment was recorded to be 33 ◦ C, in ambient tem-
perature of 22 ◦ C with natural convection, after 25 min of motion.
High Power SEA Development for Torque-Controlled Exoskeletons 73
Fig. 2. (a) Actual torque-angle curve of the spring. (b) A torsional spring specimen.
(c) Von misses stress (MPa) distribution of the spring under 1 Nm torque loading for
the fatigue calculation.
2.2 Torque Control

Having completed the design procedures, we conducted preliminary experiments
using the robust torque control technique proposed in [5]. As the result, the
actuator unit provided a control bandwidth up to 12 Hz; see Fig. 3. Although
Fig. 3. Sine input response of the torque controller, respectively for 6.2 Hz, 0.5 Hz, and
11 Hz.
74 M. C. Yildirim et al.
the performance was satisfactory, tracking performance of the controller will

be further enhanced via the use of other robust control techniques. During the
experiments, the output link was subjected to a stiff environment, i.e., it was
blocked with the obstacle that has a non-stiff material between link and the
environment.
3 Discussion
In this paper, we succinctly explained the development and torque control imple-
mentation of our SEA unit to actuate our next generation exoskeleton robots.
The units have relatively high torque-to-weight ratios and their mechanical
behaviour is well tested. The preliminary experiments showed that its torque
control performance is promising. In our next work, we will report our SEA-
powered exoskeleton systems.
References
1. Paluska, D., Herr, H.: The effect of series elasticity on actuator power and work
output: implications for robotic and prosthetic joint design. Robot. Auton. Syst.
54(8), 667–673 (2006)
2. Yildirim, M.C., Sendur, P., Bilgin, O., Gulek, B., Yapici, G.G., Ugurlu, B.: An
integrated design approach for a series elastic actuator: stiffness tuning, fatigue
tests, thermal management. In: Proceedings of the IEEE International Conference
on Humanoid Robots, pp. 384–389, UK (2017)
3. Negrello, F., Garabini, M., Catalano, M.G., Malzahn, J., Caldwell, D.G., Bicchi, A.,
Tsagarakis, N.G.: A modular compliant actuator for emerging high performance
and fall-resilient humanoids. In: Proceedings of IEEE Conference on Humanoid
Robotics, Seoul, Korea, pp. 414–420 (2015)
4. Ugurlu, B., Oshima, H., Narikiyo, T.: Lower body exoskeleton-supported compliant
bipedal walking for paraplegics: how to reduce upper body effort. In: Proceedings
of IEEE International Conference on Robotics and Automation, Hong Kong, pp.
1354–1360 (2014)
5. Oh, S., Kong, K.: High precision robust force control of a series elastic actuator.
IEEE/ASME Trans. Mechatron. 22(1), 71–80 (2017)
Investigation on Variable Impedance Control
for Modulating Assistance in Walking
Strategies with the AUTONOMYO
Exoskeleton
A. Ortlieb1,2(&), P. Lichard1,2, F. Dzeladini1,2, R. Baud1,2,

H. Bleuler1,2, A. Ijspeert1,2, and M. Bouri1,2(&)
1
Laboratory of Robotic Systems, Swiss Federal Institute of Technology
Lausanne (EPFL), Station 9, 1015 Lausanne, Switzerland
{amalric.ortlieb,mohamed.bouri}@epfl.ch
2
Biorobotics Laboratory, Swiss Federal Institute of Technology Lausanne
(EPFL), Station 9, 1015 Lausanne, Switzerland
Abstract. Impedance based control is an established strategy to provide assist-

as-needed power support in rehabilitation. The studied control in this paper
differs, however, from the standard approach. Instead of following a defined
spatio-temporal motion where the assisting torques depends on the capacity of
the user to follow a trajectory, the controller is divided in gait phases with
constant impedance parameters. This approach should allow more freedom of
the user and simple adaptability of assistance level to the user condition. The
simulated gait pattern relies on three states (swing, stance and a double support
phase) which are associated to different spring-like impedances. In this paper,
we investigate the relation between the gait and parameters such as the spring
stiffness and the attractive position. Results show that both the user and the
impedance’s parameters can influence gait characteristics such as step length,
cadence, walking velocity.
1 Introduction
Lower extremity exoskeletons have proven their abilities to help people with ambulation
disorders, such as complete spinal cord injury (SCI) patients [1]. Powered exoskeletons
provide a verticalized posture and assist the leg motions to reproduce gait. While
remaining barriers have to be overcome for a larger use of exoskeleton in daily living
(cost, need for an accompanying person, solicitation of both arms, etc.), their usage in
rehabilitation is still promising as much as these obstacles can be removed or lowered.
The major concern of assisting robots for the rehabilitation of neurological
impairments is to promote the sensory-motor recovery of the user, which implies a high
Research supported by ASRIMM (Association Suisse Romande Intervenant contre les Maladies
neuro-Musculaires), FSRMM (Fondation Suisse de Recherche pour les Maladies Musculaires)
and the Swiss Multiple Sclerosis Society.

https://doi.org/10.1007/978-3-030-01887-0_15
76 A. Ortlieb et al.
level of human-robot interaction. In this case, exoskeletons can be referred to as haptic

interfaces and need to meet a good backdrivability or zero-impedance mode. The
exoskeleton AUTONOMYO, see Fig. 1, is used in this study and is presented in the
following. Control strategies for the assistance and training of ambulation are keys in
rehabilitation solutions and address at least the following characteristics:
1. Initiate and stop walking on the user’s demand
2. Adjustable level of assistance to the user’s need – with adaptations at the joint level
3. The gait is driven by the user while avoiding spatio-temporal constraints on motion.
Impedance based control has been previously implemented for the rehabilitation of
gait. For instance, in [2] the authors propose a trajectory based controller on the gait
trainer Lokomat, which implements an attractive field that increases with the position
error relative to the defined trajectory.
Fig. 1. The AUTONOMYO haptic lower limb exoskeleton with six degrees of freedom
remotely located from the joints
In this paper, a variable impedance controller is proposed and described to fulfil the
three targeted objectives stated above. The controller is implemented and tested on
AUTONOMYO with different parameters to evaluate how versatile and free of con-
straints it can be.
The study presented here evaluates the correspondence between the control approach
and points 1–3. The variable impedance controller implemented on the AUTONO-
MYO device is tested on a healthy user.
Investigation on Variable Impedance Control for Modulating Assistance 77
2.1 Exoskeleton AUTONOMYO

AUTONOMYO is composed of six actuators for the hip and knee flexion/extension
and for the hip adduction/abduction at both legs. The system is interfaced at the foot,
the shank and the trunk of the body. The device weights about 25 kg and is adaptable
to user’s height from 160 cm to 205 cm. See [3] for more details.
2.2 3-Phases Variable Impedance Controller

The 3-phases variable impedance controller simulates spring-like behavior inspired
from the muscle contributions during walking. The impedance is defined for the knee
and hip joints by Eq. (1), where the stiffness k and attractive angle a0 can be adjusted to
the need of the user.
sjoint ¼ k ða a0 Þ b a_ ð1Þ
a and a_ are the measured joint angle and velocity respectively, b is a damping
coefficient and sjoint is the corresponding torque provided by the exoskeleton. The
stiffness k and the attractive angle a0 can be adjusted to the need of the user. Note that
the torque is limited to a maximal value of 25 Nm. The impedance varies depending on
the walking phase, which may correspond to the swing-, stance- or a double support-
phase. The gait is initiated, and phases are detected, through the hip flexion velocity.
Full details are provided in [4]. Each step is triggered by a flexion motion of the hip that
the user intent to perform. The actuations about the hip adduction/abduction were
locked to a 0° position to focus on the flexion/extension assistance, a light support was
needed to compensate for this mobility constraint.
Table 1. Tests and parameters

Trial# Hip Knee User implication
Swing Stance Swing Stance
k a0 k a0 k a0 k a0
1 2 40 2 −10 0.5 50 0.5 5 Comfortable
2 2 60 2 −10 0.5 50 0.5 5 Low
3 2 60 2 −10 0.5 50 0.5 5 Comfortable
4 2 60 2 −10 0.5 50 0.5 5 High
5 3 60 3 −10 0.5 50 0.5 5 Low
6 3 60 3 −10 0.5 50 0.5 5 Comfortable
7 3 60 3 −10 0.5 50 0.5 5 High
8 0.5 40 0.5 −10 0.5 20 0.5 5 Comfortable
9 0.5 40 2 −20 0.5 20 1 5 Comfortable
10 2 40 0.5 −20 1 20 0.5 5 Comfortable
11 2 40 2 −20 1 20 1 5 Comfortable
12 2 40 2 −20 1 20 1 5 High
78 A. Ortlieb et al.
2.3 Method
Trials have been performed overground over a distance of 18 m with a healthy user
(203 cm, 95 kg, 29 year old) wearing the exoskeleton. Parameters explored are
attractive angles and stiffness for the hip and knee during swing and stance phases. The
level of implication of the user described as low, comfortable or high were modulated
to see the influence to the gait characteristics. Time to travel the 18 m, number of steps,
cadence, range of motion at each joint, etc. were collected to evaluate gait character-
istics. The list of trials with parameters are shown on Table 1.
3 Results
The gait characteristics resulting from the different parameters of the controllers and of
the implication of the user are summarized in Table 2. We observe that the implication
of the user has a large effect on the walking speed and step length (Trials 4, 7 and 12).
In opposition, the contribution of the stiffness coefficients is unclear as only poor
modifications are observed in trials 8 to 11 while the stiffness is modulated during each
phases of the gait. The range of motion of both hip and knee are correlated with the
swing attractive angle, where the knee is correlated with the hip parameter.
Table 2. Gait characteristics resulting from trial conditions

Trial# Walking speed Hip range of Knee range of Cadence Step
[m/s] (km/h) motion [°] motion [°] [steps/min] length
[m]
1 0.37 (1.32) 73 49 51 0.86
2 0.25 (0.9) 78 65 30 1
3 0.40 (1.44) 83 68 40 1.2
4 0.50 (1.8) 89 71 40 1.5
5 0.29 (1.05) 85 75 31 1.1
6 0.39 (1.41) 88 70 39 1.2
7 0.46 (1.66) 86 72 43 1.3
8 0.55 (1.99) 56 62 52 1.3
9 0.47 (1.71) 49 60 51 1.1
10 0.56 (2.01) 51 54 56 1.2
11 0.51 (1.85) 51 60 55 1.1
12 0.86 (3.09) 56 45 74 1.4
4 Discussion
Results highlight the flexibility of the controller. The user is primarily in charge of the
walking performances as demonstrated by results. This aspect is key regarding the
motivation of the user and can provide feedback to therapists on the level of his
implication in the training.
Investigation on Variable Impedance Control for Modulating Assistance 79
The contribution of the variable stiffness, however, did not correlate with the
walking speed as it could have been expected. This observation can be explained by
both a decrease of the user activity while facing more assistance and a limited dif-
ference in assistance due to the torque limit implemented in the exoskeleton.
Finally, tuning the attractive angles during the swing phase seems to have a strong
positive impact on step length, but in opposition, reduces the walking cadence.
5 Conclusion
Simple (only three phases) variable impedance control can lead to a very flexible tool to
assist and certainly rehabilitate people with partial gait impairments. Such an approach
seems to fulfill anticipated characteristics of customization of the assistance of the
device while letting the user modulate walking velocity, cadence and step length.
Further studies with small groups of impaired people are planned and will address in
more details the adequacy of such control with different populations suffering from
neuromuscular and neurological disorders.
References
1. Miller, L.E., Zimmermann, A.K., Herbert, W.G.: Clinical effectiveness and safety of powered
exoskeleton-assisted walking in patients with spinal cord injury: systematic review with meta-
analysis. Med. Devices (Auckl) 9, 455–466 (2016)
2. Riener, R., Lunenburger, L., Jezernik, S., Anderschitz, M., Colombo, G., Dietz, V.: Patient-
cooperative strategies for robot-aided treadmill training: first experimental results. IEEE
Trans. Neural Syst. Rehabil. Eng. 13(3), 380–394 (2005)
3. Ortlieb, A., Bouri, M., Baud, R., Bleuler, H.: An assistive lower limb exoskeleton for people
with neurological gait disorders. In: 2017 International Conference on Rehabilitation Robotics
(ICORR), pp. 441–446 (2017)
4. Ortlieb, A., et al.: “An Active Impedance Controller to Assist Gait in People with
Neuromuscular Diseases: Implementation to the Hip Joint of the AUTONOMYO Exoskele-
ton”, Presented at the BioRob. Enschede, Netherlands (2018)
Improving Usability of Rehabilitation
Robots: Hand Module Evaluation
of the ARMin Exoskeleton
Fabian Just1(B) , Daniel Gunz1 , Jaime Duarte1 , Davide Simonetti2 ,

Robert Riener1 , and Georg Rauter1,3
1
Sensory-Motor Systems Lab, ETH Zurich and Spinal Cord Injury Center,
University Hospital Balgrist, Zurich, Switzerland
fabian.just@hest.ethz.ch
2
Research Unit of Biomedical Robotics and Biomicrosystems,
Campus Bio-Medico University of Rome, Rome, Italy
3
BIROMED Lab, DBE, University of Basel, Basel, Switzerland
Abstract. The impact of arm rehabilitation robots increases with their

usability. Hereby, usability can refer to many aspects of the robot’s func-
tionalities in relation to interaction with the patient and/or the thera-
pist. In the current case, the usability of robotic hand modules are in the
focus. Especially for patients with spastic hand function, the design of
the hand module is a critical factor for the patient set-up time. In this
paper, the development of a new hand module for the ARMin according
to usability requirements is presented. The requirements entail fast set-up
time, functional movement and force training as well as hygiene factors.
The developed hand module fulfills the requirements and is expected to
increase usability and acceptance of the device.
1 Introduction
1.1 Usability of Rehabilitation Robots
The usability of a rehabilitation robot for patient and therapist directly impacts
the acceptance of the device [1]. Therefore, rehabilitation robot development
processes should always focus on therapist and patient [1,2]. The difficulty of
using or learning to use the rehabilitation robot can also have impact on the
device’s acceptance [2]. For the ARMin rehabilitation robot, we constantly work
on new methods to improve the usability of the device [3]. In this paper, the
current hand module of ARMin (see Fig. 1) was evaluated according to main
factors to increase usability for patients and therapists.
This work was supported by ETH research grant 0-20075-15, ETH, UZH and the
CRRP Neuro-Rehab, University of Zurich.
https://doi.org/10.1007/978-3-030-01887-0_16
Improving Usability of Rehabilitation Robots 81
1.2 Usability of Hand Modules for Rehabilitation
In the literature, two main approaches to actuate the hand of the patient, hand
exoskeletons and pole devices that operate inside of the hand. Furthermore the
rehabilitation purpose of the device, the sensing method, actuation type and
power transmission are characterizing the device [4]. Hand exoskeletons are usu-
ally attached to the dorsal hand side, control each finger over the full range of
motion individual but have a higher set-up time [5,6]. Pole devices like the hap-
tic knob [7] and the Alpha-Prototype II [8] are easier to set up, but individual
finger training is hard to perform and the grasp possibilities are limited.
2 Methods
2.1 Patient Set-Up in the ARMin Rehabilitation Robot
Fig. 1. (a) The 7 DoF ARMin rehabilitation robot during a training session with
therapist and patient. (b) The current hand module of ARMin for pinching movements.
The ARMin rehabilitation robot is used in stroke rehabilitation. The hand

module of the ARMin is training pinching movements, which are necessary for
activities of daily living (ADL) like carrying and interacting with objects (see
Fig. 1b)). For setting up, the patient has to be put first in the black lower and
upper arm cuff, then his hand needs to be opened before he finally can be
attached to the fabric straps of the hand module. For spastic patients the opening
of the hand needs to be done by the therapist before the hand can be inserted
into the module.
82 F. Just et al.
2.2 Requirements for an Ideal ARMin Hand Module

From literature overview and the usage of ARMin in the clinical setting the
following requirements for an ideal hand module have been set:
(1) Fast set-up: Easy and intuitive handling and 30 s maximum set-up time
(2) Enough operating force: A grasping force of 200 N is needed to cover most
of the clinical usage [9]
(3) Functional movement: The most important hand movement, the medium
wrap (ca. 30% of all important ADL) should be covered [10].
(4) Hygiene: The device should be easily cleanable.
3 Results
The hand module as seen in see Fig. 2(a) consists of a crank slider mechanism
with the open top to easily insert a hand without opening the hand and is based
on the work by Masia et. al. [8] and Lambercy et. al [7]. The insertion of the hand
through the open top is simplified by the cone structure on the top, which even
enables insertion in strongly closed hands. The attachment to the hand module
can be done independently of ARMin everywhere in the room (see Fig. 2(c)).
Finally, the module is integrated and fixed on ARMin with two easy to operate
hooks. Different grip shapes that are easily exchangeable cover various medium
wrap grip executions (see Fig. 3). Hand modules can be exchanged during train-
ing due to a clutch at the connection of the adaptive hand module to the ARMin
gear and motor.
Fig. 2. (a) Hand module in closed position without grip. (b) Hand module in open
position with grips. (c) Subject with inserted closed hand module in hand.
Improving Usability of Rehabilitation Robots 83
Fig. 3. Different exchangeable grip shapes of the ARMin hand module enable versatile
task executions and training difficulties.
4 Discussion
The hand module prototype fulfills the requirements mentioned above:
(1) For the fast set-up especially with spastic patients the adapter hand module
(see Fig. 2(c)) can already to be connected in a non spastic position before
the patient is fixed in the exoskeleton ARMin. After inserting the hand,
the module can be easily pressed into a snap locking hook connector on
ARMin to interlink to motor and gear.
(2) The mechanical structure is designed to withstand operating forces of 200 N.
(3) A pole diameter change from 16 mm up to 42 mm covers most activities of
the medium wrap grip.
(4) The used ABS plastic is no fabric and cleanable with disinfection spray.
Nevertheless, for future work safety aspects need be addressed as possible skin
jamming, that can happen when the grips are closing.
5 Conclusion
For increasing usability of the ARMin rehabilitation robot, a novel hand module
design was presented that fulfilled all set requirements of an ideal hand mod-
ule. The usability gains for patients and therapy are expected to improve the
acceptance of the device and to foster rehabilitation outcomes for patients.
Acknowledgment. The authors would like to thank Michael Herold-Nadig and

Marco Bader for their constant, valuable support.
84 F. Just et al.
References
1. Lee, M., Rittenhouse, M., Abdullah, H.A.: Design issues for therapeutic robot
systems: results from a survey of physiotherapists. J. Intell. Rob. Syst. 42(3), 239–
252 (2005)
2. Lu, E.C., Wang, R.H., Hebert, D., Boger, J., Galea, M.P., Mihailidis, A.: The
development of an upper limb stroke rehabilitation robot: identification of clinical
practices and design requirements through a survey of therapists. Disabil. Rehabil.
Assist. Technol. 6(5), 420–431 (2011)
3. Just, F., Baur, K., Riener, R., Klamroth-Marganska, V., Rauter, G.: Online adap-
tive compensation of the ARMin rehabilitation robot. In: 2016 6th IEEE Inter-
national Conference on Biomedical Robotics and Biomechatronics (BioRob), pp.
747–752. IEEE (2016)
4. Heo, P., Gu, G.M., Lee, S.-J., Rhee, K., Kim, J.: Current hand exoskeleton tech-
nologies for rehabilitation and assistive engineering. Int. J. Precis. Eng. Manuf.
13(5), 807–824 (2012)
5. Wege, A., Hommel, G.: Development and control of a hand exoskeleton for rehabil-
itation of hand injuries. In: 2005 IEEE/RSJ International Conference on Intelligent
Robots and Systems, IROS 2005, pp. 3046–3051. IEEE (2005)
6. Chiri, A., Vitiello, N., Giovacchini, F., Roccella, S., Vecchi, F., Carrozza, M.C.:
Mechatronic design and characterization of the index finger module of a hand
exoskeleton for post-stroke rehabilitation. IEEE/ASmE Trans. Mechatron. 17(5),
884–894 (2012)
7. Lambercy, O., Dovat, L., Gassert, R., Burdet, E., Teo, C.L., Milner, T.: A haptic
knob for rehabilitation of hand function. IEEE Trans. Neural Syst. Rehabil. Eng.
15(3), 356–366 (2007)
8. Masia, L., Krebs, H.I., Cappa, P., Hogan, N.: Design and characterization of hand
module for whole-arm rehabilitation following stroke. IEEE/ASME Trans. Mecha-
tron. 12(4), 399–407 (2007)
9. Kamper, D.G., Fischer, H.C., Cruz, E.G., Rymer, W.Z.: Weakness is the primary
contributor to finger impairment in chronic stroke. Arch. Phys. Med. Rehabil.
87(9), 1262–1269 (2006)
10. Bullock, I.M., Zheng, J.Z., De La Rosa, S., Guertler, C., Dollar, A.M.: Grasp
frequency and usage in daily household and machine shop tasks. IEEE Trans.
Haptics 6(3), 296–308 (2013)
Lower Limb Exoskeletons, from Specifications
to Design
M. Bouri(&)
Laboratory of Robotic Systems, Swiss Federal Institute of Technology Lausanne

(EPFL), Station 9, 1015 Lausanne, Switzerland
mohamed.bouri@epfl.ch
1 Introduction
Lower limb exoskeletons are motorized and instrumented wearable devices rigidly
interfaced with the wearer. Their roles are to assist walking or mobilize legs to make
people walk. Differences are made between pure mobilization and assistance. In the
case of mobilization, precomputed gait trajectories [1] are used to provide a defined
desired gait when walking (Fig. 1), while in assistance of walk, the locomotion con-
troller has to adapt the joint actuation to the intent and the capabilities of the user [2–4].
2 Robotic Solutions for Rehabilitation
Rehabilitative purposes were the first tackled issues with robotic solutions. For stroke
patients, fortunately, restorative chances of locomotion are quite high, while for
paraplegic and tetraplegic patients, the focus are on strengthening muscles and
improving gait coordination, rather than total recovery. [5] reports well some evidences
related to using robotic devices for rehabilitation. Nevertheless, some authors remain
more critic in concluding about the effectiveness of robotics compared to conventional
therapies. Mehrholz, in his last review for stroke rehabilitation [6], pointed out that we
must pay attention to the time in which the effects are maintained, he concluded
“Further research should consist of large definitive pragmatic phase III trials
undertaken to address specific questions about the most effective frequency and
duration of electromechanical-assisted gait training as well as how long any benefit
may last”.
Research supported by the company Sonceboz, by ASRIMM (Association Suisse Romande

Intervenant contre les Maladies neuro-Musculaires), FSRMM (Fondation Suisse de Recherche
pour les Maladies Musculaires), the Swiss Multiple Sclerosis Society, and the Swiss National
Centre of Competence in Research (NCCR) Robotics.

https://doi.org/10.1007/978-3-030-01887-0_17
86 M. Bouri
Fig. 1. Example of precomputed joint trajectories used on the exoskeleton TWIICE [1].
3 Presentation Outline
Exoskeletons primarily targeted rehabilitation. Probably, because their use is better

controlled in known environments, with accompanying therapists and in predefined
situations and scenarios. Nevertheless, we believe, that the important driver of this
technology remains activities for daily living. This will, without any doubt, enhance the
access of people with lower limb disability to activities they did not have access before.
The development of exoskeletons should therefore be more motivated by the
requirements of people, in their daily activities. Precisely in indoor and outdoor
walking, sitting and standing and in climbing up/down stairs. The technology should be
improved to be more reliable and better accessible to potential industrial manufacturers.
In the current presentation, specifications and developments of lower limb
exoskeletons with respect to different objectives will be discussed. Examples of design
and results will be presented. The Author is involved in lower limb robotic devices
since 2002 and will share his experience with the audience.
References
1. Vouga, T., Baud, R., Fasola, J., Bouri, M., Bleuler, H.: TWIICE—A lightweight lower-limb
exoskeleton for complete paraplegics. In: 2017 International Conference on Rehabilitation
Robotics (ICORR), pp. 1639–1645. IEEE, July 2017
2. Olivier, J., Bouri, M., Ortlieb, A., Bleuler, H., Clavel, R.: Development of an assistive
motorized hip orthosis: kinematics analysis and mechanical design. In: 2013 IEEE
International Conference on Rehabilitation Robotics (ICORR), pp. 1–5. IEEE, June 2013
3. Baud, R., Ortlieb, A., Olivier, J., Bouri, M., Bleuler, H.: HiBSO hip exoskeleton: toward a
wearable and autonomous design. In: International Workshop on Medical and Service Robots,
pp. 185–195. Springer, Cham, July 2016
Lower Limb Exoskeletons, from Specifications to Design 87
4. Ortlieb, A., Bouri, M., Baud, R., Bleuler, H.: An assistive lower limb exoskeleton for people
with neurological gait disorders. In: 2017 International Conference on Rehabilitation Robotics
(ICORR), pp. 441–446. IEEE, July 2017
5. Candace Tefertiller, P.T., Pharo, B.: Efficacy of rehabilitation robotics for walking training in
neurological disorders: a review. J. Rehabil. Res. Dev. 48(4), 387 (2011)
6. Mehrholz, J., Thomas, S., Werner, C., Kugler, J., Pohl, M., Elsner, B.: Electromechanical
assisted training for walking after stroke. The Cochrane Library (2017)
Robotic and Neuroprosthetic Balance
Management Approaches
for Walking Assistance
Novel Perturbation-Based Approaches
Using Pelvis Exoskeleton Robot in Gait
and Balance Training After Stroke
Zlatko Matjačić(&), Matjaž Zadravec, Nataša Bizovičar, Nika Goljar,

and Andrej Olenšek
University Rehabilitation Institute Republic of Slovenia,

Linhartova 51, 1000 Ljubljana, Slovenia
{zlatko.matjacic,matjaz.zadravec,natasa.bizovicar,
nika.goljar,andrej.olensek}@ir-rs.si
Abstract. We have developed an innovative admittance-controlled Balance

Assessment Robot that enables movement of a pelvis in all six degrees-of-
freedom while a subject is walking on an instrumented treadmill. Further, we
have developed a number of training approaches that are targeted to diminish
specific deficiencies like gait asymmetry, insufficient weight-bearing, reduced
push-off and poor dynamic balancing capabilities. A novel approach of
precisely-timed push-like exertion of forces to the pelvis, performed similarly to
physiotherapists that physically manipulate pelvis to indirectly modify trajec-
tories of pelvis and legs of trainees, was developed. The developed approach
was implemented in a series of case studies involving high-functioning stroke
subjects in the early chronic stage. The results of prolonged training with high
volume, specificity and intensity brought about significant changes in the bal-
ancing capabilities as well as in overall walking performance.
1 Introduction
Maintaining dynamic balance during walking is immanent to movement of humans. In

contrast to numerous studies into dynamic balancing during standing there has been
little research into dynamic relationship between centre-of-mass (COM), centre-of-
pressure (COP) and ground reaction forces (GRF) during walking under perturbed
conditions. Consequently, our knowledge on neurophysiological and biomechanical
processes underlying efficient balancing during walking is limited. This has consid-
erable implications on limited success of rehabilitation of walking following various
neurological impairments.
In the recent two decades a number of rehabilitation robots intended for training of
walking has been deployed to clinical environment. Locomat (Hocoma AG), G-EO
(Reha Technologies AG) and Lyra (medica Medizintechnik GmbH) are devices that are
used to restore stepping following pre-determined trajectories in lower limbs in the
early phase of rehabilitation while devices like Andago (Hocoma AG) and E-go
This work has been supported by Slovenian Research Agency under research program P2-0228
and research project J2-8172.

https://doi.org/10.1007/978-3-030-01887-0_18
92 Z. Matjačić et al.
(medica Medizintechnik GmbH) are used to practice overground walking in latter

phase of rehabilitation. After completing rehabilitation program majority of stroke
survivors are left with specific gait deficiencies like pronounced stepping asymmetry,
incomplete weight bearing ability and reduced push-off on the impaired leg as well as
reduced balancing abilities that would need to be further addressed.
2 Methodology and ResultS
We have developed an innovative perturbing apparatus named Balance Assessment

Robot (Fig. 1) which is in essence an admittance-controlled exoskeleton that embraces
a human pelvis and enables unhindered movement of pelvis in all six degrees-of-
freedom during walking as well as application of precisely-timed force impulses that
perturb walking [1].
Fig. 1. Balance Assessment Robot.

Novel Perturbation-Based Approaches Using Pelvis Exoskeleton Robot 93
We have performed several experimental studies, applying perturbing pushes to the

pelvis of neurologically intact subjects walking on an instrumented treadmill. The
results have shown that the balancing reactions to perturbations delivered in the frontal
plane require well-coordinated responses that relate to all three planes of movement [2].
Furthermore, the results have also indicated that balancing responses to perturbing
pushes at the level of pelvis greatly vary also with speed of walking as well as with the
magnitude of perturbation. At lower speeds of walking control of horizontal component
of ground-reaction-force (GRF), achieved predominantly by the action of hip abductor
muscles (“hip strategy”) and control of COP achieved predominantly by the action of
ankle musculature (“ankle strategy”), are the main balancing strategies (Fig. 2). At
higher speeds of walking predominant response consists of adequate stepping (“step-
ping strategy”) following perturbing pushes (Fig. 3). Further experiments were done in
selected cases of post-stroke subjects in early chronic phase. Balancing responses in
post-stroke subjects are subject-specific and may be very different from normal
responses.
Fig. 2. An example of balancing responses in chronic post-stroke subject following outward

perturbation of various amplitudes. After the left heel strike perturbation to the left commenced.
The subjects reacted with lateral COPx displacement (“ankle strategy”) and an impulse-like
increase in GRFx (“hip strategy”) during the left stance and with inward positioning of the right
leg in the next step (“stepping strategy”).
A novel approach of precisely-timed push-like exertion of forces to the pelvis,

performed similarly to physiotherapists that physically manipulate pelvis to indirectly
94 Z. Matjačić et al.
modify trajectories of pelvis and legs of trainees, was developed and applied to a
selected high-functioning post-stroke subjects. The results of prolonged training with
high volume (30 training sessions), specificity and intensity brought about significant
changes in the balancing capabilities as well as in overall walking performance [3].
Fig. 3. Illustration of “stepping strategy”.
3 Conclusion
Our initial experience with long term training in selected high-functioning post-stroke
subjects in chronic stage of illness after completed rehabilitation program using
perturbation-based approach have shown that further improvement in balancing
capabilities as well as in overall walking performance can be achieved.
Novel Perturbation-Based Approaches Using Pelvis Exoskeleton Robot 95
References
1. Olenšek, A., Zadravec, M., Matjačić, Z.: A novel robot for imposing perturbations during
overground walking: mechanism, control and normative stepping responses. J. Neuroeng.
Rehabil. 13, 55 (2016)
2. Matjačić, Z., Zadravec, M., Olenšek, A.: An effective balancing response to lateral
perturbations at pelvis level during slow walking requires control in all three planes of motion.
J. Biomech. 60, 79–90 (2017)
3. Matjačić, Z., Zadravec, M., Olenšek, A.: Feasibility of robot-based perturbed-balance training
during treadmill walking in a high-functioning chronic stroke subject: a case-control study.
J. Neuroeng. Rehabil. 15, 32 (2018)
Balance During Bodyweight Supported
and Robot-Assisted Walking
Eva Swinnen1(&), Jean-Pierre Baeyens2,4, Nina Lefeber1,

Emma De Keersmaecker1, Stieven Henderix3, Marc Michielsen3,
and Eric Kerckhofs1
1
Rehabilitation Research – Neurological Rehabilitation Group,
Vrije Universiteit Brussel, Brussels, Belgium
eva.swinnen@vub.be
2
Biometry and Biomechanics Group,
Vrije Universiteit Brussel, Brussels, Belgium
3
Jessa Hospitals, Sint-Ursula Rehabilitation Center, Herk-de-Stad, Belgium
4
THIM van der Laan University College Physiotherapy, Landquart, Switzerland
Abstract. Robot-assisted gait rehabilitation improves gait- and balance related

outcome measures, but its merit over other gait rehabilitation methods is still
insufficiently proven. The trunk and pelvis are important to maintain balance
during gait. Despite, scarce research has been presented concerning the
importance of the trunk and the pelvis in different types of gait rehabilitation
methods. We investigated whether the amount of bodyweight support and the
amount of guidance force have an influence on the movements of the trunk and
the pelvis during robot-assisted walking (Lokomat-system). The results of this
study suggest that walking with the Lokomat-system leads to significant changes
in trunk and pelvis movements, which influence the training of the trunk balance
during robot-assisted walking. These results should be taken into account in the
development of gait rehabilitation robots and gait rehabilitation strategies.
1 Introduction
To regain walking ability, gait training has been proven to be an effective approach for
people with central neurological disorders and injuries, such as stroke, multiple scle-
rosis and incomplete spinal cord injury. Intensive walking training can be performed
with the use of a treadmill in combination with assistive devices, such as bodyweight
support systems. During bodyweight supported treadmill training, an overhead harness
partially supports the patient’s bodyweight while the movements of the legs are assisted
by one, two or more physiotherapists. However, for the therapists, this type of therapy
requires strenuous physical effort, often leading to shortened training sessions [1].
To address this inconvenience, gait rehabilitation robots are increasingly being
deployed. In general, different groups of gait rehabilitation robots can be distinguished.
For instance, “end-effector devices”, which move the feet, and “exoskeleton devices”,
which move the joints, such as hip, knee, and ankle, controlled during the gait phases.
These devices can be static—for example fixed on a treadmill—or dynamic—moving

https://doi.org/10.1007/978-3-030-01887-0_19
Balance During Bodyweight Supported and Robot-Assisted Walking 97
the patient around the environment. Static exoskeletons and end-effectors are usually
combined with bodyweight support systems [2].
There is some evidence that the use of robot-assisted gait training in people after
stroke has positive effects on gait, but also on balance related outcome measures [3, 4].
Earlier studies investigated the effect of bodyweight support, with an overhead sus-
pension system, on the trunk movements [5, 6] and concluded that the use of body-
weight support leads to smaller maximum trunk and pelvis movement amplitudes
compared to walking without bodyweight support; this with the exception of the
anterior-posterior movement of the pelvis. One of the questions in this context is
whether during walking with a treadmill based robotic systems—in which the trunk
and pelvis are fixed due to the bodyweight support system and/or the exoskeleton—
there is sufficient training of the trunk balance. It is clear that retraining the ability to
maintain balance in individuals with central neurological disorders is important because
balance impairments may contribute to disordered gait.
We investigated the influence of the level of bodyweight support and/or guidance
force on trunk and pelvis kinematics when healthy participants were walking with a
treadmill based exoskeleton (Lokomat-system) [7, 8].
2.1 Participants
For this study [7, 8] 18 healthy participants (15 females and 3 males) were recruited.
Exclusion criteria were neurological disorders and/or mention of cognitive disorders,
injuries of the lower extremities or back during the previous six months, abnormal
range of motion of the lower limbs or trunk and balance problems. Participants char-
acteristics are presented in Table 1. Most of the participants had limited (1 or 2 times)
preliminary experience with robot-assisted walking with the Lokomat-system.
Table 1. Characteristics of the participants

Characteristics Mean (Standard Deviation; Range)
Body mass (kg) 63.40 (8.26; 53.0–80.0)
Body height (m) 1.72 (0.08; 1.60–1.90)
BMI (kg/m2) 21.4 (1.51; 19.2–24.7)
Age (years) 27 (3.81; 21–33)
To check if participants met the inclusion criteria, they filled in a medical and
demographic questionnaire. After approval, each participant signed an informed con-
sent form. All participants were insured, and the study was approved by the local ethics
commission of the university (BUN B1432008499) and the rehabilitation center
(12.11/fys12.02).
98 E. Swinnen et al.
2.2 Materials and Methods

The gait rehabilitation robot used was the Lokomat-system (Hocoma AG, Switzerland),
which consists of an exoskeleton that is fixated around the pelvis and the lower limbs.
The mechanical limbs of the Lokomat-system guide the lower extremities of the par-
ticipants in a prescribed pattern (trajectory control). The Lokomat-system is used in
combination with a bodyweight support system that compensates part of the weight of
the participant. First, the participants walked for 5 min on the treadmill without the
Lokomat-system. Afterwards, the Lokomat-system was fixated and the Litegait
bodyweight support system was fixed on the suspension fork.
The suspension with the bodyweight support system was 0%, 30% and 50% of the
bodyweight of each participant and the guidance force was 30%, 60% and 100%. By
drawing lots, there was a randomization of the order of walking conditions. During the
entire protocol, the walking speed was 2 kmph and measurements of 30 s took place
after a 4-min acclimatization period.
The Polhemus LibertyTM (240/16) (Polhemus, Colchester, England, UK), an
electromagnetic tracking device, was used to assess the kinematics of the trunk and the
pelvis and to determine the gait cycle. The output data were calculated in a local
maxima/minima algorithm (graphical user interface, Matlab, Eindhoven, The Nether-
lands). For each variable from each subject, the mean of these maxima and minima was
used to calculate the average ROM. Consequently, the ROM in each plane, for each
subject, in each walking condition was deduced. Statistics were performed using SPSS
v20 (IBM, Chicago, IL). A repeated measures analysis of variance with Bonferroni
correction for multiple comparisons was used to analyze the outcome variables. The
significance level (a) was set at 5%. For a more detailed description of the data and
statistical analysis see [7, 8].
3 Results
Significant changes in trunk and pelvis kinematics, when comparing robot-assisted

treadmill walking with treadmill walking without robot-assistance, were found
(p 0.05). In general, during robot-assisted treadmill walking (combinations of 30%,
60% and 100% guidance force and 0%, 30% and 50% bodyweight support) there were
significant (1) smaller antero-posterior and lateral translations of the trunk and the
pelvis, (2) smaller antero-posterior flexion and axial rotation of the trunk, (3) larger
lateral flexion of the trunk, and (4) larger antero-posterior tilting of the pelvis.
The literature concerning trunk and pelvis kinematics during walking with bodyweight
support and/or robot-assistance is scarce [5, 6] and the effects of variations of guidance
force, bodyweight support and walking velocity during walking with the Lokomat-
system are still poorly understood [9, 10]. Muscle activity and kinematics seem mainly
influenced when subjects walk with high levels of bodyweight support and slow
Balance During Bodyweight Supported and Robot-Assisted Walking 99
walking speeds [5, 6, 9, 10]. From the data of the current studies it can be concluded
that compared to regular treadmill walking, walking with the Lokomat-system leads to
small but significant changes in trunk and pelvis movements with overall restrictions in
the movements of the upper body [7, 8].
These studies, investigated in healthy people, suggest that when robot-assistance is
added to the therapy, there is a change in balance training of the trunk compared to
conventional gait training on a treadmill.
The data should be further investigated in patient populations and results of these
studies should be taken into account in the development of gait rehabilitation strategies
and robots. It is important that therapists are aware of the influence of the use of
bodyweight support and robotic systems on the trunk balance during gait rehabilitation.
Acknowledgment. We are grateful to the therapists of the rehabilitation center Sint-Ursula,

Jessa Hospital, for participation and assistance in the study. This study was funded by the
ALTACRO-project (Vrije Universiteit Brussel).
References
1. Calabro, R.S., Cacciola, A., Berte, F., Manuli, A., Leo, A., Bramanti, A., et al.: Robotic gait
rehabilitation and substitution devices in neurological disorders: where are we now? Neurol
Sci. 37(4), 503–514 (2016)
2. Morone, G., Paolucci, S., Cherubini, A., De Angelis, D., Venturiero, V., Coiro, P., et al.:
Robot-assisted gait training for stroke patients: current state of the art and perspectives of
robotics. Neuropsychiatr. Dis. Treat. 13, 1303–1311 (2017)
3. Swinnen, E., Beckwée, D., Meeusen, R., Baeyens, J.-P., Kerckhofs, E.: Does robot-assisted gait
rehabilitation improve balance in stroke patients? Top. Stroke Rehabil. 21(2), 87–100 (2014)
4. Mehrholz, J., Thomas, S., Werner, C., Kugler, J., Pohl, M., Elsner, B.: Electromechanical-
assisted training for walking after stroke. Cochrane Database Syst. Rev. 5 (2017)
5. Swinnen, E., Baeyens, J.-P., Pintens, S., Van Nieuwenhoven, J., Isbroukx, S., Buyl, R.,
et al.: Trunk kinematics during walking in persons with multiple sclerosis: the influence of
body weight support. NeuroRehabilitation 34(4), 731–740 (2014)
6. Aaslund, M.K., Moe-Nilssen, R.: Treadmill walking with body weight support. Effect of
treadmill, harness and body weight support systems. Gait Posture. 28, 303–308 (2008)
7. Swinnen, E., Baeyens, J.-P., Knaepen, K., Michielsen, M., Hens, G., Clijsen, R., et al.:
Walking with robot assistance: the influence of body weight support on the trunk and pelvis
kinematics. Disabil. Rehabil: Assist. Technol. 10(3), 252–257 (2015)
8. Swinnen, E., Baeyens, J.-P., Knaepen, K., Michielsen, M., Clijsen, R., Beckwée, D., et al.:
Robot-assisted walking with the Lokomat: the influence of different levels of guidance force
on thorax and pelvis kinematics. Clin. Biomech. 30(3), 254–259 (2015)
9. van Kammen, K., Boonstra, A.M., Reinders-Messelink, H.A., den Otter, R.: The combined
effects of body weight support and gait speed on gait related muscle activity: a comparison
between walking in the Lokomat exoskeleton and regular treadmill walking. PloS ONE 9(9)
(2014)
10. van Kammen, K., Boonstra, A.M., van der Woude, L.H.V., Reinders-Messelink, H.A., den
Otter, R.: The combined effect of guidance force, bodyweight support and gait speed on
muscle activity during able-bodied walking in the Lokomat. Clin. Biomech. 36, 65–73
(2016)
Maintaining Gait Balance After Perturbations
to the Leg: Kinematic and Electromyographic
Patterns
Eleonora Croci, Roger Gassert, and Camila Shirota(&)
Rehabilitation Engineering Laboratory,

Institute of Robotics and Intelligent Systems, Department of Health Sciences
and Technology, ETH Zurich, Zurich, Switzerland
{roger.gassert,camila.shirota}@hest.ethz.ch
Abstract. Maintaining balance following gait perturbations is difficult and still

not well addressed in gait assistive devices. A challenge is in identifying per-
turbations, and whether and which responses are required to reestablish balance
and walking. Here, we investigate the timing of changes in the kinematic and
muscle activation patterns of unimpaired subjects to external perturbations. We
used the ETH Knee Perturbator to lock the knee at different points of swing
phase, and identified changes in the gait pattern with Statistical Parametric
Mapping, adjusted for data containing perturbations. We show that kinematic
patterns differ within approximately 100 ms of the perturbation, and that muscle
activity changes later, much closer to foot-strike. Our results suggest that
mechanical (joint angles and velocities) sensors are best suited to identify
external perturbations, devices should change their behavior in response to such
perturbations, and responses may not need to be initiated immediately following
the perturbation.
1 Introduction
Gait assistive devices have greatly advanced in the past decades, enabling an ever-
increasing range of activities to their users. However, most devices still lack the ability
to respond to balance-disturbing perturbations, especially during gait. For example, the
most advanced solutions available for above-knee amputees stiffen or lock the knee to
prevent joint buckling [1]. However, such solutions are very limited compared to the
abilities of unimpaired human knee joints to generate responses, and are further
insufficient to address balance-disrupting disturbances, as indicated by the high inci-
dence of falls in populations that use such devices [2].
We first need to better understand how unimpaired humans respond to different
types of balance disturbances during gait. Although many studies in gait perturbations
have been done [3, 4], they were limited by methods that can only analyze single points
Research supported by the Swiss National Center of Competence in Research on Robotics

(NCCR Robotics), and the ETH Zurich Foundation in collaboration with Hocoma AG.

https://doi.org/10.1007/978-3-030-01887-0_20
Maintaining Gait Balance After Perturbations to the Leg 101
in the gait cycle, or single variables, as opposed to the multi-variate time-series that
characterize gait. Here, we investigate the response of unimpaired human subjects to
disturbances to the knee joint during different points in swing phase. We compare the
muscle activity and kinematic responses to undisturbed walking using a multivariate
time-series technique to identify when patterns change. These changes indicate the
responses used to maintain balance, and could be used as target patterns for gait
assistive devices in response to perturbations.
2 Methods
2.1 Experimental Setup

Two subjects (30 and 25 years old, BMI of 22.1 and 22.5 kg/m2, respectively) par-
ticipated in this experiment. Subjects gave informed consent prior to their participation
(swissethics, PB_2017-00160/KEK-ZH: 2014_0049).
The ETH Knee Perturbator [5] was used to lock the right knee for 200 ms at 6
points in swing phase (10%, 22% 34%, 46%, 58% and 70%). Perturbations were
repeated 15 times each, applied in random order, and with at least 30 s between
perturbations to avoid anticipation. Data collection occurred in blocks of 20 min, with
mandatory pauses between blocks to avoid confounding effects, e.g., due to fatigue.
EMG (Noraxon TeleMyo DTS, Noraxon, Scottsdale AZ USA) from 7 leg muscles
(right and left rectus femoris (RF) and tibialis anterior (TA), and right gluteus medius
(GMed), semitendinosus (ST) and gastrocnemius (GM)) were measured at 1500 Hz.
Motion data of the pelvis and lower limbs were collected using reflective markers
(Optitrack, NaturalPoint, Covallis OR USA) and exported to Visual3D (C-Motion,
Germantown MD USA), where gait events (foot-strike and toe-off) and the angle and
velocity of the hip and knee joints were calculated. All data were exported to and
analyzed in MATLAB (The Mathworks, Natick MA USA).
EMG signals were high-passed at 50 Hz, rectified, then low-passed at 25 Hz, using
2nd-order Butterworth filters applied to have zero delay. Strides were defined between
right foot-strikes. The perturbed strides and the strides immediately before each per-
turbed stride were extracted for analysis. Data from each stride were resampled to 101
points in time.
2.2 Data Analysis

A modified version of statistical parametric mapping (SPM) was used to compare the
perturbed and baseline (stride immediately before the perturbation) strides. SPM [6] is
the multivariate time-series equivalent to a t-test, taking into account the dependencies
between time points and multiple variables used to describe a system. Paired Hotell-
ing’s T2 test (a = 0.05) were used to compare data from each of the perturbation times
to their corresponding baseline strides.
SPM outputs a test statistic as a function of time, and a threshold over which
statistical significance is achieved. We adjusted the SPM threshold to the maximum
value of the test statistic before the perturbation, to avoid the identification of
102 E. Croci et al.
differences that are statistically significant, but not relevant to the problem of identi-
fying changes post-perturbation.
For each subject and each perturbation time, we compared perturbed strides to
baseline strides based on EMG data only or kinematic data only. We measured the time
between the perturbation (as indicated by the closing of the ETH Knee Perturbator
clutch) and the first time point where the adjusted SPM indicated differences in the
signal.
Muscle activation (Fig. 1) and kinematic patterns differed between baseline and per-
turbed strides for both subjects for almost all perturbation times (Fig. 2).
Fig. 1. Sample EMG data and SPM output for subject 1. Baseline (blue) and perturbed (red;
34% swing phase) muscle activation patterns, and the output of the adjusted SPM (bottom).
Perturbation time is the vertical orange bar, and adjusted SPM threshold is the red dotted line.
Kinematic patterns differed earlier than EMG patterns in all cases. This is likely due
to the nature of the perturbations, which should cause changes to the kinematics before
there are changes to muscle activation patterns, as is generally expected to occur when
there are external disturbances to gait. Differences in movement patterns were not
Maintaining Gait Balance After Perturbations to the Leg 103
Fig. 2. Time after perturbation when perturbed strides are statistically different from baseline
strides for EMG (circles) and kinematic (triangles) data, for subjects 1 (left) and 2 (right). Shaded
areas represent the average time between perturbation start and end of swing phase.
immediately evident, generally occurring around 100 ms post-perturbation regardless

of perturbation timing in swing phase.
In contrast, EMG patterns were much slower to change, but differences still
occurred before the end of swing phase with a couple of exceptions (22% and 70% for
subject 2). This indicates that active responses were used to counteract the perturba-
tions, suggesting the need to react to external perturbations as opposed to the lack of
responses in most gait assistive devices. More detailed comparisons between the
kinematic and EMG responses across our perturbations would indicate whether mul-
tiple reaction patterns would be necessary, and what each would consist of.
Further, EMG responses tended to occur quicker with perturbations later in swing
phase. This could be related to the amount of time available between the perturbation
and the end of swing phase: the shorter the time available, the earlier reactions need to
be to adequately restore balance and the baseline walking pattern. For perturbations in
early swing, this delayed response could be further suggesting that responses do not
need to occur immediately after the perturbation, but rather that some time can be
‘spared’ and dedicated to gather more information before selecting an adequate
response pattern. However, data from more subjects and different types of perturbations
would be needed to better understand these issues.
4 Conclusion
Maintaining balance after perturbations during gait requires adequate responses that are
still not addressed by current gait assistive devices. In this paper, we showed that
mechanical (joint angles and velocities) sensors indicate the occurrence of a pertur-
bation earlier than muscle activity, regardless of perturbation timing. Our results
104 E. Croci et al.
suggest that mechanical sensors would be best suited to quickly indicate gait distur-
bances, and that assistive devices should change their behavior to help reestablish
balance and the undisturbed walking pattern, based on the perturbations timing.
Acknowledgment. The authors would like to thank Y. Bader and M.R. Tucker for help with
data collection, S. Wyss and A. Melendez-Calderon for discussions regarding the statistical
method, and O. Lambercy for support.
References
1. Bellmann, M., Schmalz, T., Blumentritt, S.: Comparative biomechanical analysis of current
microprocessor-controlled prosthetic knee joints. Arch. Phys. Med. Rehabil. 91, 644–652
(2010)
2. Kulkarni, J., Wright, S., Toole, C., Morris, J., Hirons, R.: Falls in patients with lower limb
amputations: prevalence and contributing factors. Physiotherapy 82, 130–136 (1996)
3. Schillings, A.M., van Wezel, B.M., Mulder, T., Duysens, J.: Muscular responses and
movement strategies during stumbling over obstacles. J. Neurophysiol. 83, 2093–2102 (2000)
4. Cham, R., Redfern, M.S.: Lower extremity corrective reactions to slip events. J. Biomech. 34,
1439–1445 (2001)
5. Tucker, M.R., Shirota, C., Lambercy, O., Sulzer, J., Gassert, R.: Design and characterization
of an exoskeleton for perturbing the knee during gait. IEEE Trans. Biomed. Eng., 20 January
2017
6. Pataky, T.C., Robinson, M.A., Vanrenterghem, J.: Vector field statistical analysis of
kinematic and force trajectories. J. Biomech. 46, 2394–2401 (2013)
A New Sensory Feedback System
for Lower-Limb Amputees: Assessment
of Discrete Vibrotactile Stimuli Perception
During Walking
Mariangela Filosa1, Ilaria Cesini1, Elena Martini1, Giacomo Spigler1,

Nicola Vitiello1,2, Calogero Oddo1, and Simona Crea1,2(&)
1
BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy
simona.crea@santannapisa.it
2
Fondazione Don Carlo Gnocchi, Milan, Italy
Abstract. Sensory feedback systems can improve gait performance of lower-

limb amputees by providing information about the foot-ground interaction force.
This study presents a new platform designed to deliver bilateral vibrations on the
waist of the user, synchronously with specific gait events. Preliminary percep-
tual tests were carried out on five healthy subjects to investigate the perception
thresholds on the abdominal region. The reaction time and the percentage of
correct perceptions were computed to compare three stimulation levels: 50%,
70% and 100% of the maximum vibration amplitude (i.e., 1.5 g, 1.9 g and
2.2 g). The reaction times decreased with higher activation levels. The per-
centage of correct perceptions was 40% with 50% stimulation level and higher
than 97% with 70% and 100% stimulation levels, respectively. The results
suggest that vibration amplitudes of 1.9 g provide vibrotactile stimulation that
can be effectively perceived during walking, thus used to convey sensory
information.
1 Introduction
Lower-limb loss is a major disabling condition, typically causing abnormal gait

kinematics, higher metabolic consumption, back pain and secondary joint and bone
disorders [1]. Altered sensory-motor afferent feedback is a primary determinant of these
impairments. Although amputees can rely on the forces acting at the stump-socket
interface for maintaining balance in standing or during locomotion, the development of
a sensory feedback device that can provide information about the foot-ground inter-
action is considered a promising solution to improve gait performance and increase the
outcomes of rehabilitation [2]. Vibrotactile stimulation, in particular, is among the most
This work was supported in part by the EU within the CYBERLEGs Plus project (H2020-ICT-
2016-1 Grant Agreement #731931) and in part by the Italian National Institute for Insurance
against Accidents at Work (INAIL Centro Protesi, Budrio) within the MOTU project.

https://doi.org/10.1007/978-3-030-01887-0_21
106 M. Filosa et al.
preferred solutions to provide feedback due to its minimal invasiveness, high accept-
ability and cost-effectiveness.
A previous study conducted by our team demonstrated that gait symmetry can be
improved by delivering unilateral discrete vibrotactile stimulations synchronously with
specific gait events [3, 4]. Despite the promising results, the previous study required a
multiple-day training phase to let the amputee learn how to take benefit from the
sensory information, i.e. how to associate the haptic feedback to a more physiological
gait pattern. In this study, we describe a novel sensory feedback platform for lower-
limb amputees, designed to improve the intuitiveness of the discrete event-driven
stimulation strategy by delivering bilateral short-lasting and low-intensity vibrations at
the occurrence of specific gait phase transitions, in order to provide feedback from both
the sound and the prosthetic limbs. Along with the presentation of the system, we
report a preliminary study on the vibrotactile perception capability of five healthy
subjects under dynamic conditions, with the ultimate goal to identify suitable low-
intensity stimulation levels for providing effective feedback during walking.

The sensory feedback device used in the present study consisted of three main hard-
ware modules, as shown in Fig. 1: a pair of pressure-sensitive insoles, a control board
(namely, the Vibro Board) and 12 vibrating motors (VT units) attached to a commercial
belt adjustable in size.
Fig. 1. Experimental setup. (a) A subject on the treadmill during the experimental session.
(b) Distribution of the VT units around the trunk. (c) VT units encapsulated in the PDMS matrix
and attached to the belt. (d) Instrumented shoes housing the pressure-sensitive insoles.
Each insole comprised an array of optoelectronic sensors [5]. Pressure data were
wirelessly transmitted to the real-time processor of the Vibro Board (sbRIO 9651,
National Instruments, PHX, USA), where the instantaneous centre of pressure and the
vertical ground reaction force were estimated in real-time. According to the calculated
A New Sensory Feedback System for Lower-Limb Amputees 107
biomechanical variables, three gait phases were identified, similarly to [1]. Concur-
rently with gait phase transitions, the Vibro Board generated the commands to activate
one VT unit with a specified intensity.
Each miniaturized VT unit was made of an eccentric rotating mass motor (Precision
Microdrives, London, U.K.) encapsulated in a matrix of PDMS. The smoother contact
area of the silicon layer was conceived to increase the comfort for the user without
degrading perception, thanks to the wider contact area. The selected vibrating motors
could reach a vibration amplitude of 2.6 g when operating at rated-speed. However, due
to the inertia of the eccentric mass, and the presence of the silicon layer, the actual
vibration amplitude is lower than the maximum value when activated for short times.
A preliminary analysis showed a vibration amplitude of 2.2 g with a vibration duration
of 100 ms. The VT units were attached to the belt and equally spaced at a distance of
5.5 cm from each other (consistently with the tactile spatial resolution on the abdomen
[6]).
2.2 Experimental Protocol

Five able-bodied subjects (one female, age 26.6 ± 1.1; weight 65.2 ± 10.6 kg; height
1.74 ± 0.04 m; foot size 41–43 EU) were recruited for the study. They were asked to
walk on a treadmill at self-selected speed while wearing the instrumentation. Prior to
the experiment, a short session was performed to let the subjects familiarize with the
activation of the VT units. During the perception test, each VT unit was activated
according to a pseudo-random sequence to deliver 100 ms bursts of vibration at three
different levels of intensity, (50, 70 and 100% of the maximum amplitude, corre-
sponding to vibration amplitudes of 1.5 g, 1.9 g and 2.2 g, respectively), and syn-
chronously with one of three gait events (heel-strike, foot-flat, toe-off). Every time they
perceived a vibration, the subjects were required to press a button connected to the
Vibro Board as quickly as possible and communicate which VT unit they perceived as
active.
The stimulation duration of 100 ms was selected to achieve clear perception without
overlap between successive vibrations, annoyance or adaptation effects [3, 4, 7, 8]. The
number of steps occurring between two consecutive activations was randomized to
avoid possible bias due to expectation. Each stimulation was repeated 4 times. In total,
432 activations were tested (12 VT units, 3 phase transitions, 3 intensity levels, 4
repetitions).
2.3 Data Analysis

A MATLAB routine (MathWorks, Inc.) was developed for offline data elaboration. The
percentage of correct perceptions was calculated for each stimulation intensity. The
Reaction Time (RT), i.e. the elapsed time between the VT unit activation and subject
response, was calculated. Missing recognitions were not considered in the computation
of the RTs. Mean RTs and correct perceptions of each stimulation level were extracted
for each subject and then the across-subject median values were computed.
108 M. Filosa et al.
Figure 2 shows the across-subject median and interquartile range of the RT and overall
percentage of correct perceptions, for each stimulation level. Overall, the RT resulted
lower with higher stimulation levels (700 ms, 602 ms and 562 ms, respectively with
50%, 70% and 100% stimulation levels). The percentage of correct perceptions was
40% with the lowest stimulation level and higher than 97% with 70% and 100%
stimulation levels. A similar trend was observed in a previous study [8], where the RTs
were computed to compare three vibration frequencies: 140, 180, and 220 Hz, with the
latter resulting in the shortest RT.
Fig. 2. Median ± interquartile range of reaction time (in blue) and percentage of correct
perceptions (in red) for each stimulation level.
The main purpose of this work was to analyze the subjects’ perception thresholds
on the abdomen, when vibrating stimuli are delivered during walking. So far, psy-
chophysical studies have identified the minimum skin displacement and stimulation
frequency required for detecting vibration on the abdominal region in static conditions
(4–14 µm at 200 Hz) [7]. However, it is unknown whether the ambulation condition
can affect perception.
Our results suggest that perception thresholds on the abdomen during locomotion
fall between 50% and 70%. As expected, the maximum level (100%) showed the best
performance. Nevertheless, at 70% vibrations were already clearly perceived by the
subjects. Therefore, in order to achieve a trade-off between comfort and correct per-
ception, the stimulation level shall be set at 70%, as high intensity vibrations might
induce skin adaptation effects or be perceived as bothersome by subjects, in long-term
utilization.
Future studies will investigate the capability of the device to improve the gait
performance of lower-limb amputees.
A New Sensory Feedback System for Lower-Limb Amputees 109
References
1. Tucker, M.R., Olivier, J., Pagel, A., Bleuler, H., Bouri, M., Lambercy, O., del R. Millán, J.,
Riener, R., Vallery, H., Gassert, R.: Control strategies for active lower extremity prosthetics
and orthotics: a review. J. Neuroeng. Rehabil. 12(1), 1 (2015)
2. Zambarbieri, D., Schmid, M., Verni, G.: Intelligent systems and technologies in rehabilitation
engineering. In: Teodorescu, H.N.L., Jain, L.C. (eds.), pp. 129–151 (2001)
3. Crea, S., Cipriani, C., Donati, M., Carrozza, M.C., Vitiello, N.: Providing time-discrete gait
information by wearable feedback apparatus for lower-limb amputees: usability and
functional validation. IEEE Trans. Neural Syst. Rehabil. Eng. 23(2), 250–257 (2015)
4. Crea, S., Edin, B.B., Knaepen, K., Meeusen, R., Vitiello, N.: Time-discrete vibrotactile
feedback contributes to improved gait symmetry in patients with lower limb amputations: case
series. Phys. Ther. 97(2), 198–207 (2017)
5. Crea, S., Donati, M., De Rossi, S., Oddo, C., Vitiello, N.: A wireless flexible sensorized insole
for gait analysis. Sensors 14(1), 1073–1093 (2014)
6. Cholewiak, R.W., Brill, J.C., Schwab, A.: Vibrotactile localization on the abdomen: effects of
place and space. Percept. Psychophys. 66(6), 970–987 (2004)
7. Jones, L.A., Lederman, S.J.: Human Hand Function. Oxford University Press, London (2006)
8. Sharma, A., Leineweber, M.J., Andrysek, J.: Effects of cognitive load and prosthetic liner on
volitional response times to vibrotactile feedback. J. Rehabil. Res. Dev. 53(4), 473–482
(2016)
Fast Online Decoding of Motor Tasks
with Single sEMG Electrode
in Lower Limb Amputees
Federica Barberi1, Federica Aprigliano1, Emanuele Gruppioni2,

Angelo Davalli2, Rinaldo Sacchetti2, Alberto Mazzoni1,
and Silvestro Micera1,3(&)
1
The BioRobotics Institute, Scuola, Superiore Sant’Anna, Pisa, Italy
s.micera@santannapisa.it
2
INAIL Prosthesis Center Vigorso di Budrio (BO), Bologna, Italy
3
Bertarelli Foundation Chair in Translational Neuroengineering, Center for
Neuroprosthetics and Institute of Bioengineering, School of Engineering,
École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland
Abstract. The quality of life of lower limb amputees strongly depends on the
performance of their prosthesis. Active prostheses controlled by prosthesis
sensors can participate to the movement and improve the walking performance
of the amputees. However, a promising control mechanism involves the use of
electromyography (EMG) to decode motor intentions. This approach could
timely inform the prosthesis about the steps that the patient is going to perform
much earlier compared to the feedback given by sensors. Here, we investigate
whether an EMG-based algorithm is able to detect the motor intentions of
transfemoral amputees. Subjects with a transfemoral amputation performed
different motor tasks (e.g., ground level walking, climbing up/down stairs),
while we recorded the EMG signals from surface electrodes placed on the
subject’s stump. Our decoding algorithm achieved 100% motion intention dis-
crimination. Such perfect decoding was achieved usually after less than 100 ms
from the onset of the movement, thus ensuring that the information about the
next step could be transmitted to the active prostheses with a sufficient advance
to achieve its proper control. These results showed not only the feasibility of
EMG-based online decoding of motor intentions, but also that perfect decoding
can be achieved online with as little as one recording site, ensuring a minimum
discomfort and encumbrance of the whole system.
Authors thank Scuola Superiore Sant’Anna, Pisa, and Centro Protesi INAIL, Budrio. This study
was funded by MOTU (Protesi Robotica di Arto Inferiore con Smart Socket ed Interfaccia
Bidirezionale per Amputati di Arto Inferiore), INAIL PPR-AI 1/2. A special thanks to Dr. Pericle
Randi.

https://doi.org/10.1007/978-3-030-01887-0_22
Fast Online Decoding of Motor Tasks 111
1 Introduction
Notwithstanding the progress in the development of active prosthesis in recent years,

lower-limb amputees rarely feel completely confident in their own prosthesis since they
are still not able to control it in an intuitive way [1]. As a consequence, lower-limb
amputees are more apt to suffer from neuromuscular disorders such as asymmetric
walking and balance, which often lead to joint degeneration, higher metabolic con-
sumption and back pain [2]. Lower-limb prostheses can be passive, quasi-passive, or
active [3]. Candidate control unit solutions for active prosthesis include the use of
sensors either on the healthy leg (“echo control” case), or the amputee side (“intent or
gait mode recognition” case) [4]. In the last case sensors information can be extracted
both from the prosthesis (e.g., inertial sensors, pressure sensors) and the residual limb
of the patient surface EMG (sEMG) or intramuscular EMG (iEMG) [5]. First attempts
at developing lower limb active prostheses based on sEMG signals have been suc-
cessful [6, 7] but their performances rely on the use a large set of recording sites [8].
Here, we present a new decoding algorithm of task intention detection, based on
sEMG recorded from a single electrode thus ensuring minimal discomfort for the
amputee. This information could help in achieving an automatic smooth and precise
control of the prosthesis itself, helping amputees feel more confident in their prosthesis.
2 Methods
Experiments took place at Centro Protesi INAIL in Budrio (Bologna). Seven subjects
with above-knee amputation, Medicare Functional Classification Level (K-Level) 3 or
4, and current use of a semi-passive mechanical prosthetic knee participated in this pilot
study. Subjects were informed about the study prior the experiments. With the help of a
physical therapist we identified functional residual muscles suitable for recording the
activity: Gluteus Medius, Tensor Fasciae Latae, Rectus Femoris, Adductor. The sEMG
signal was recorded at 2 kHz using the 64-channel portable device Sessantaquattro by
OTB. Two foot switches were placed under the foot. Both EMG and amplification
circuit were arranged in a bag fastened around the hip of the patient.
Subjects performed five different tasks during sEMG recording: standard walk,
ascend and descend a ramp, and climb up and down stairs. We tested all possible
transitions between these tasks, with at least five trials for each transition. sEMG
signals were pre-processed using a fourthorder Butterworth band pass filter between 10
and 250 Hz. Gait events were identified using the signal recorded by the footswitches.
The analysis focused on the first step of each task (i.e., from the first heel off to the
following heel strike). From each sEMG signal we extracted the maximum fractal
length (MFL) (1).
XL1
MFL ¼ log10 i¼1
j xi þ 1 xi j ð1Þ
where xi is the ith sample of sEMG signal amplitude, and L is the length of the
window. Linear discriminant analysis (LDA) was tested for single muscle/regressor
112 F. Barberi et al.
combination. Performances of the classifier were computed using leave-one-out

validation.
3 Results
The movement onset was identified by the lost of contact of the pressure sensor located
in the prosthesis heel (heel off). The average duration of the following gait phase varied
from 716 ms for walk on plain to 1157 ms for stair down.
Table 1 reports the times from movement onset required to decode with 100%
performance the different transitions based on the LDA of the MFL extracted from the
sEMG of single muscles. For each transition there was at least one muscle able to
perform the decoding perfectly, even for particularly different transitions as those
comparing similar behaviors as walking on plain or over a ramp. Transition to and from
stair up/down was always correctly identified by the majority of the inspected muscles.
Table 1 shows also that for 9/10 transitions (all except walk vs ramp down) there is
at least one muscle from which it is possible to achieve perfect decoding in less than
100 ms (through the use of MFL feature). This surprising result is due to the fact that at
the moment of heel off of the first step of a particular task, muscles are already set in a
task-specific tension and this can be easily detected from proper analysis of sEMG (see
Fig. 1).
Table 1. Perfect classification times

MFL sEMG decoding - time from movement onset
(ms)
Task TFL RF ADD GM
Walk vs stair up – 45 165 20
Walk vs stair down 355 365 20 130
Walk vs ramp up – – 50 30
Walk vs ramp down – – – 340
Stair up vs stair down 230 120 20 135
Stair up vs ramp up – 85 415 20
Stair up vs ramp down 200 145 295 20
Stair down vs ramp up 320 300 20 20
Stair down vs ramp down 150 75 20 20
Ramp up vs ramp down – – 60 –
Fast Online Decoding of Motor Tasks 113
MFL of adductor (mV)
Time from heel off
Fig. 1. Evolution of Maximum Fractal Length (MFL) of adductor in the first 200 ms following
heel off during the first step climbing the stairs (black lines) or descending them (purples lines).
Vertical red line at 20 ms indicates that the two movements can be always discriminated at
movement onset.
4 Discussion
We were able to achieve perfect decoding from single electrodes in a small fraction of
the task duration. This is a very important result as our final aim is to use the motor
intention decoding to control active prosthesis. We focused on single muscle decoding
as we aimed at designing a prosthesis control device causing the minimal discomfort
for the amputee. Still, decoding from the sEMG of pairs of muscles can increase the
performance both in number of tasks decoded and in decoding speed while keeping a
good level of comfort. Further studies will then focus on the possibility of decoding
task transition from couple of sEMG electrodes via Support Vector Machine decoding.
5 Conclusion
Decoding of the movement intention from single sEMG electrodes during the early
phase of the first step is possible. Further methods of analysis are being carried out to
reinforce the reported outcomes.
References
2. Ehde, D.M., Czerniecki, J.M., Smith, D.G., Campbell, K.M., Edwards, W.T., Jensen, M.P.,
Robinson, L.R.: Chronic phantom sensations, phantom pain, residual limb pain, and other
regional pain after lower limb amputation. Arch. Phys. Med. Rehabil. 81(8), 1039–1044
(2000)
3. Martinez-Villalpando, E.C.: Design and evaluation of a biomimetic agonist-antagonist active
knee prosthesis, pp. 1–102 (2012)
114 F. Barberi et al.
4. Windrich, M., Grimmer, M., Christ, O., Rinderknecht, S., Beckerle, P.: Active lower limb
prosthetics: a systematic review of design issues and solutions. Biomed. Eng. Online 15(3), 5–
19 (2016)
5. Micera, S., Raspopovic, S.: Control of hand prostheses using peripheral information, pp. 48–
68 (2010)
6. Au, S.K., Bonato, P., Herr, H.: An EMG-position controlled system for an active ankle-foot
prosthesis: an initial experimental study. Sig. Process. 375–379 (2005)
7. Wu, S., Waycaster, G., Shen, X.: Electromyography-based control of active above-knee
prostheses. Control Eng. Pract. 19(8), 875–882 (2011)
8. Hargrove, L.J., Simon, A.M., Young, A.J., Lipschutz, R.D., Finucane, S.B., Smith, D.G.,
Kuiken, T.A.: Robotic leg control with EMG decoding in an amputee with nerve transfers,
pp. 1237–1242 (2013)
A Wearable Haptic Feedback System
for Assisting Lower-Limb Amputees
in Multiple Locomotion Tasks
Ilaria Cesini1(&), Giacomo Spigler1, Sahana Prasanna1,

Domitilla Taxis2, Filippo Dell’Agnello1, Elena Martini1,
Simona Crea1,3, Nicola Vitiello1,3, Alberto Mazzoni1,
and Calogero Maria Oddo1(&)
1
The BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy
{ilaria.cesini,calogero.oddo}@santannapisa.it
2
University of Twente, Enschede, Netherlands
3
Abstract. Lower limb prosthesis performance determines the quality of life of

amputee patients. Such performance will benefit from a feedback informing the
patient about the gait phase and the overall condition of the foot. This study
reports the design and validation of a wearable haptic feedback system con-
ceived to assist lower-limb amputees in various locomotion scenarios. Three
vibrating motors were attached to a belt in textile to provide information about
the foot-ground contact, by remapping the variables detected under the foot, on
the waist of the user. Multiple activation patterns were implemented and com-
pared in a pilot study involving one able-bodied subject, during walking,
ascending and descending stairs. A novel assessment protocol was proposed to
benchmark the different stimulation patterns. The protocol resulted to be a viable
method for quicker development and testing of new strategies.
1 Introduction
Sensory information from lower-limbs is crucial for individuals to maintain balance,

ambulate and adapt to the physical environment. The main consequences of lower-limb
loss are the deterioration of the overall balance performance, decreased gait pattern
quality, higher risk of falling and energy consumption [1]. Thus, the restoration of
sensory feedback in lower-limb amputees might improve their stability and increase the
confidence of the user while using the prosthesis.
Many studies proposed noninvasive systems to compensate for sensory deficit, in
which the feedback was delivered through visual [2], auditory [2, 3] or haptic [2, 4, 5]
channels, following the principle that the subjects can incorporate the augmented
This work was supported in part by the EU within the CYBERLEGs Plus project (H2020-ICT-
2016-1 Grant Agreement #731931) and in part by the Italian National Institute for Insurance
against Accidents at Work (INAIL Centro Protesi, Budrio) within the MOTU project.

https://doi.org/10.1007/978-3-030-01887-0_23
116 I. Cesini et al.
information into their body control scheme in order to restore a more physiological gait
pattern [5]. However, most of the proposed solutions focus on ground walking, while
little attention has been devoted to exploring other application scenarios, such as
ascending and descending stairs.
The goal of this project was to develop a completely wearable haptic feedback
device to convey information about the foot-ground interactions by selecting a stim-
ulation strategy out of multiple ones. We developed a protocol to compare the efficacy
of the proposed stimulation patterns on intact subjects and to identify the most intuitive
strategy for providing missing sensory information, in order to increase amputees’
awareness of the prosthesis and facilitate smooth locomotory transitions. The validation
protocol was designed to be adapted to different locomotion tasks, i.e. over-ground
walking, ascending and descending stairs.
The present work reports the experimental setup and the preliminary tests con-
ducted with one intact subject to validate the device and the stimulation strategies, and
investigate potential criticalities, before extending the feedback system to amputees.
2.1 System Setup

The wearable haptic feedback apparatus was composed of a sensorized interface,
driving electronics for data processing and motor control and three vibrating motors
(Pico Vibe 304-116, Precision Microdrives) attached to a textile belt. In this study, we
used a pressure sensitive insole to detect foot-ground interactions. The device inte-
grated an array of optoelectronic sensors and was developed at Scuola Superiore
Sant’Anna (Pisa, Italy) [6]. The data acquired from the insole were wirelessly trans-
mitted to a central processing unit, namely the VibroMOTUr board, which was
physically attached to the belt housing the vibrotactile (VT) stimulation units. The
VibroMOTUr board calculated the vertical Ground Reaction Force (vGRF) and Center
of Pressure (CoP) coordinates from the raw sensor signals. The board generated the
haptic commands to control the VTs, based on three different stimulation strategies,
described in Sect. 2.3. The three VTs were equally spaced at 9 cm on one side of the
waist of the subject, from navel to spine to generate ipsilateral perception. The
vibrating motors were encapsulated in a polymeric matrix of PDMS (Sylgard® 184
Silicone Elastomer, Dow Corning) to provide comfortable contact with the skin and
attached to the belt through Velcro fasteners. The configuration of the setup is shown in
Fig. 1.
2.2 Experimental Protocol

One able-bodied male subject of 38 years, weight 62 kg, height 1.68 m and foot size
42 EU performed the tasks. The subject was provided with a pair of shoes, with the
right shoe embedded with the pressure sensitive insole. The VibroMOTUr board was
enclosed in a 3D-printed box that was mounted on the left side of the belt using Velcro
fasteners. The VTs were attached to the belt on the right side, i.e. the same side as the
A Wearable Haptic Feedback System 117
CoP CoP + vGRF
OR
OPTOELECTRONIC SENSOR
Fig. 1. Outline of the wearable haptic feedback apparatus, consisting of a sensorized insole, a
textile belt endowed with vibrotactile units (VTs), and control electronics (VibroMOTUr board).
sensorized shoe. The subject familiarized with the haptic feedback with an initial setup
stage, in which he perceived the intensity of the stimulation while walking freely for 2–
3 min. The experimental trials consisted in ascending 11 steps followed by a short
over-ground walk, or descending the same track (i.e., walking first, followed by
descending the stairs). A different feedback stimulation was provided on each trial,
pseudo-randomized among three stimulation strategies so that a single strategy was
repeated 5 times each ascending and descending the track (for a total of 30 trials). The
time required to complete each trial was measured to estimate the performance. In fact,
since sensory feedback is not required for intact subjects to walk confidently, we
hypothesized that a poor and un-intuitive feedback strategy might increase their cog-
nitive load, affecting gait speed. During the experiment, the subject was not aware that
the durations of the trials were a variable of interest, and he performed each task
walking at normal speed.
2.3 Vibrotactile Stimulation Strategies

Two different stimulation strategies i.e. CoP and CoP_vGRF, were compared against a
baseline, CoP_Rand. They are shown in Fig. 2(a). Specifically, in the CoP one of three
VTs was activated with maximum intensity when ground contact was detected. The
activated VT was chosen by segmenting the CoP on the right foot into three segments,
each associated to a different VT (VT1 ¼ fCoP\50 mmg, VT2 ¼ f50 mm CoP\
190 mmg, VT3 ¼ f190 mm CoPg). In the CoP_vGRF a single VT was selected and
activated as in CoP, but the intensity of the activation was modulated according to the
vGRF. CoP_Rand was identical to CoP, but the mapping between the CoP segments
and the VTs was randomized every time the CoP moved to a different segment.
118 I. Cesini et al.
(a)
(b)
Fig. 2. (a) VTs activation during two steps of walking, using the CoP and CoP_vGRF
stimulation strategies. The activation is displayed in red, green and blue, while the center of
pressure on the right foot is shown in black. (b) Average time required to walk upstairs (Up) and
downstairs (Down) in each protocol, and their combined average (Up+Down). Bars denote
standard deviation.
3 Results
The results collected from the subject are shown in Fig. 2(b). It is possible to observe
that a difference was found between the strategies in terms of the time required to
complete the tasks especially in ascending the stairs, suggesting that the lesser cog-
nitive load is associated to CoP encoding. These preliminary results show that the
proposed validation protocol is effective in comparing different stimulation strategies.
4 Conclusion
We presented a new wearable haptic feedback device to convey information about the
foot-ground interactions via three VT units, activated according to three different
actuation strategies. We further performed a preliminary validation of a novel exper-
imental protocol on one able-bodied subject to compare the stimulation patterns.
Whereas most of the sensory feedback strategies found in literature focus on
walking tasks, our stimulation protocol covers different locomotion scenarios. More-
over, we proposed a novel method to assess the performance on intact subjects, which
resulted to be viable for selecting the optimal stimulation strategy. The study will be
extended to a larger pool of subjects, both intact and amputees.
A Wearable Haptic Feedback System 119
References
1. Lauretti, C., Pinzari, G., Ciancio, A.L., Davalli, A., Sacchetti, R., Sterzi, S., Guglielmelli, E.,
Zollo, L.: A vibrotactile stimulation system for improving postural control and knee joint
proprioception in lower-limb amputees. In: 2017 26th IEEE International Symposium on
Robot and Human Interactive Communication, pp. 88–93 (2017)
2. Zambarbieri, D., Schmid, M., Verni, G.: Sensory feedback for lower limb prostheses. In:
Intelligent Systems and Technologies in Rehabilitation Engineering, pp. 129–151 (2001)
3. Yang, L., Dyer, P.S., Carson, R.J., Webster, J.B., Bo Foreman, K., Bamberg, S.J.M.:
Utilization of a lower extremity ambulatory feedback system to reduce gait asymmetry in
transtibial amputation gait. Gait Posture 36(3), 631–634 (2012)
4. Crea, S., Edin, B.B., Knaepen, K., Meeusen, R., Vitiello, N., Cipriani, C., Donati, M.,
Carrozza, M.C., Vitiello, N.: Time-discrete vibrotactile feedback contributes to improved gait
symmetry in patients with lower limb amputations: case series. IEEE Trans. Neural Syst.
Rehabil. Eng. 23(2), 198–207 (2017)
6. Crea, S., Donati, M., De Rossi, S., Oddo, C., Vitiello, N.: A wireless flexible sensorized insole
for gait analysis. Sensors 14(1), 1073–1093 (2014)
Benchmarking Wearable Robots
COVR – Towards Simplified Evaluation
and Validation of Collaborative Robotics
Applications Across a Wide Range of Domains
Based on Robot Safety Skills
Jule Bessler1(&), Leendert Schaake1, Catherine Bidard2,

Jaap H. Buurke1, Aske E. B. Lassen3, Kurt Nielsen3, José Saenz4,
and Federico Vicentini5
1
Roessingh Research and Development, Enschede, Netherlands
{J.Bessler,L.Schaake,J.Buurke}@rrd.nl
2
CEA, LIST, Interactive Robotics Laboratory, Gif-sur-Yvette, France
catherine.bidard@cea.fr
3
Teknologisk Institut, Taastrup, Denmark
{aala,kuni}@teknologisk.dk
4
Fraunhofer Gesellschaft zur Förderung der angewandten Forschung, Business
Unit Robotic Systems, Magdeburg, Germany
Jose.Saenz@iff.fraunhofer.de
5
Consiglio Nazionale delle Richerche, Milan, Italy
Federico.vicentini@itia.cnr.it
Abstract. COVR is a European project driven by five national research and

technology organizations. Through the development of an intuitive toolkit and a
range of testing protocols for validation of safety for robots sharing space with
humans, it will increase the safety of all types of collaborative robots across all
domains with special attention to rehabilitation robotics.
1 Introduction
Recent developments in robotics are leading away from an approach where humans
and robots are separated by fences towards collaborative and wearable robots intro-
ducing human-robot interaction as an essential aspect. This however inevitably raises
safety issues which present a major challenge for developers.
In our experience with end-users, robotics components manufacturers, and system
integrators, safety has become a barrier to the promotion and availability of collabo-
rative robotics technology in all domains, from manufacturing to healthcare. Reasons
for that are the complexity of technologies and limitations on knowledge of standards
as well as clarity on how to provide evidence of compliance to those standards [1, 2].
COVR has received funding from the European Union’s Horizon 2020 research and innovation
programme under grant agreement No. 779966.

https://doi.org/10.1007/978-3-030-01887-0_24
124 J. Bessler et al.
Wearable robots present a particular challenge since they are physically connected to
the user. Especially in the field of rehabilitation, collaborative robots can be subject to
various types of users at the same time. Therapists need to be considered as well as
patients, who are by definition compromised in some way and can be very vulnerable.
The EU-funded project “Being safe around collaborative and versatile robots in
shared spaces” (COVR) [3] aims to systematically break down both technical and non-
technical barriers to support more widespread use of collaborative robots in a wide
range of industries and domains (e.g. manufacturing, logistics, healthcare and reha-
bilitation, agriculture).
2.1 Toolkit
To address the barriers to the development of collaborative robots (cobots), COVR will
provide a toolkit which can support cobot manufacturers and users by identifying the
safety protocols that apply to their specific case. The toolkit shall apply to all cobots in
all applications and be easy for non-specialists to use. It will analyze the cobot and its
intended behaviors by asking a series of questions and use this information to identify
relevant standards as well as matching safety requirements and risk reduction measures.
The output includes checklists of requirements that need to be fulfilled, and instructions
for performing validation tests.
We have chosen to start our toolkit in physical human-robot interaction scenarios in
(a) manufacturing, where we see the greatest immediate need due to technologies being
already mature and deployable, and (b) rehabilitation robots, where we see the greatest
challenge for any safety certification system due to the compromised nature of the
people interacting with the cobot. We will then extend into other domains such as
logistics, agriculture and assistive robots.
2.2 Testing Protocols

In many cases, the relevant standards do not clearly define the safety-related validation
procedures in sufficient detail. Wherever this is the case, COVR uses experience on
best practice and cross-fertilization among sectors to fill the gaps. For instance, if the
standards for medical electrical equipment do not specify validation procedures or pass
values for the identified item, information from industrial experience can be used. The
aim is to complete and unify the safety validation processes and implement this in the
toolkit. Wherever COVR develops additional testing protocols, active consultation with
regional stakeholders from standardization, national agencies, accident insurance, and
safety verification bodies is crucial to create a consensus on the validity of these
protocols and the toolkit.
COVR – Towards Simplified Evaluation and Validation 125
2.3 Shared Safety Facilities

In order to create a space for conducting these protocols, COVR will establish shared
safety facilities at partner sites as a service for interested third parties. These state-of the art
cobot testing facilities will offer training, access to measurement systems for validation as
well as support in using the toolkit and applying the protocols. Training will include
methodologies (risk assessment, functional safety) and safe design (technology, envi-
ronment, interactions). Developers can get a space for building mock-ups and prototypes
to test and get guidance in implementation of risk reduction measures. After the product
specific requirements and matching test protocols have been defined by the toolkit, the
shared safety facility can be used to perform those protocols. The results can be utilized for
development and implementation of additional safety functions where necessary.
2.4 Realistic Trials/Financial Support for Third Parties

The shared safety facilities will also carry out realistic trials. Third parties seeking to
engage with COVR can get funding in the scope of the COVR Awards, to “stress test”
the toolkit and protocols with specific use-cases, and to provide background research
and experimental data for determining best practices.
There will be three open calls encouraging groups working in fields like design and
installation of pilot cobot systems, development of cobot systems or components (in-
cluding safety components or systems) to apply for funding. The calls will be almost
identical but expanding with regard to domain coverage. Initially, COVR focuses
mainly on cobots in manufacturing and rehabilitation robots and will then expand to
other domains such as agriculture and logistics. In each of the calls, awards will be
given out to applicants developing a cobot technology, product or system for use in
shared spaces and focusing on the safety of collaboration with humans e.g. by
developing safety evaluation protocols, test devices or setups. Any legal entity based in
an EU member state or associated country is eligible to apply, individually or in a
consortium. Detailed information on evaluation criteria and other updates will be
provided via the project website (www.safearoundrobots.com) before the first open call
(open 15/11/2018). The awardees will get access to shared safety facilities at partner
sites and the COVR toolkit in its current version to support the development, certifi-
cation and deployment of a cobot related product. These processes are seen as beta-
tests of all COVR elements and will be used to validate and enrich the COVR toolkit
and testing protocols.
3 (Expected) Results
The main aim of the project is to enable robotic users to find out exactly how to certify
or recertify a cobot in about 30 min. It is expected that by the end of the project, the
toolkit will offer a comprehensive list of protocols relevant to the industrial domain and
rehabilitation protocols will be available for most cases. The foundation for the
domains logistics, agricultural, commercial and assistive cobots will also have been
laid. Importantly, we expect to achieve general acceptance across all stakeholders by
regular consultations before any release of the protocols and toolkit.
126 J. Bessler et al.
Moreover, various cobot safety services will be established and suitable organi-
zations will be encouraged to become additional COVR safety hubs.
4 Discussion
COVR develops various services designed to make safety certification easier and more
transparent, and thereby increase safety around cobots. The common approach is novel:
it allows for a single point of access for users from all domains, fields of applications
and countries. This common framework will enable cross-fertilization and increase the
robustness of protocols – where a harmonized testing protocol does not exist, the
nearest sister domain can be used for inspiration. This is also supported by basing
testing protocols on cobot skills which, in contrast to product-specific metrics, are
independent of product type and field of application. A broad acceptance of testing
protocols and validation procedures can only be achieved by working together with
many types of cobot stakeholders.
There are several online tools for robot risk assessment available [4–8] however,
COVR will be the first to focus on cobots in all fields of application.
5 Conclusions
COVR is developing a new process using a toolkit framework that creates a unified
approach to safety assurance. The overall ambition is to create a workable system
which guides cobot developers and integrators through a safety assessment process that
can be used directly in safety certification in most cobot application domains. Shared
safety facilities will physically support the tests that the toolkit requires, host realistic
trials, and the services will ensure that help is on hand if needed.
References
1. Vasic, M., Billard, A.: Safety issues in human-robot interactions. In: 2013 IEEE International
Conference on Robotics and Automation, pp. 197–204 (2013)
2. Guiochet, J., Do Hoang, Q.A., Kaâniche, M., Powell, D.: Applying existing standards to a
medical rehabilitation robot : limits and challenges. In: IEEE/RSJ International Conference on
Intelligent Robots and Systems, p. 5 (2012)
3. COVR - Welcome to Safety (2018). http://safearoundrobots.com/. Accessed 30 Apr 2018
4. Kastanienbaum. http://www.kastanienbaum.com/company.html. Accessed 30 Apr 2018
5. Software About. http://www.designsafe.com/cgi-bin/index-4a.php. Accessed 30 Apr 2018
6. Intoduction to the designsafe software - YouTube. https://www.youtube.com/watch?v=
8PJBpVnpDfA. Accessed 30 Apr 2018
7. RSC-Robot safety Center. http://robotsafetycenter.com/Robot-Safety-Center.html. Accessed
30 Apr 2018
8. SAPHARI - Safe and Autonomous Physical Human-Aware Robot Interaction - Home. http://
www.saphari.eu/. Accessed 30 Apr 2018
Monitoring Upper Limbs
During Exoskeleton-Assisted Gait Outdoors
Matteo Lancini(&), Simone Pasinetti, Valeria Montini,

and Giovanna Sansoni
Department of Mechanical and Industrial Engineering,

University of Brescia, Brescia, Italy
matteo.lancini@unibs.it
Abstract. Powered exoskeleton can restore locomotion to spinal cord injury

subjects but measuring their impact on the upper limbs is critical, since repeated
excessive loads are strongly correlated to chronic pain at shoulder level.
This paper presents a novel set of instrumented crutches, able to measure
force exerted on the ground during walking sessions, thanks to embedded time-
of-flight cameras and force sensors.
The force sensors, along with an inertial module, assess the force acting on
the upper limbs, while the time-of-flight cameras detects the gait phases looking
at the feet position.
The aim is to provide an affordable measuring system, without requiring a
fully instrumented gait-lab, allowing the user-robot interaction to be measured in
a more natural setting, closer to the foreseen working condition.
The instrumented crutches are fully independent of any other instrumentation
to allow a comparison of different exoskeleton models in terms of upper limb
involvement.
1 Introduction
The loss of locomotion is one of the major impairments that could result from a spinal
cord injury. In recent years more and more exoskeletons are being developed with the
aim to overcome this limitation [1, 2]. A few are also commercially available [3], and
their efficacy and long-term health effects were investigated in different studies [4–6].
Most of these studies are performed in specialized gait lab, using motion capture,
accelerometric data and force platforms to acquire a full kinetic and kinematic defi-
nition of the human-exoskeleton system [7–9].
This approach has however been limited to indoor application, in a small, con-
strained space, and only for a limited number of walking sessions.
To overcome this problem a set of wireless instrumented crutches has been
developed, with the aim of allowing the measurement of upper limbs involvement
during assisted gait in a more natural setting, and without relying on the exoskeleton
specific model or capabilities.

https://doi.org/10.1007/978-3-030-01887-0_25
128 M. Lancini et al.
A set of forearm crutches were instrumented using a set of four 350 Ω strain gauges and
a Bluetooth data acquisition system, as described in [10], to measure the axial force
exerted by the subject. These crutches, with respect to others found in the literature, are
capable of wireless measurements.
Since the force variation on the crutches due to step initiation is not clear enough to
mark the step initiation, a Camboard Picoflexx (PMD Technologies®) Time-of-Flight
(ToF) camera was mounted on each crutch. The camera contains only the depth sensor,
which makes it small and lightweight (68 17 7.25 mm and 8 g only). The camera
has a resolution of 224 171 pixels and a viewing angle of 62° 45°. The mea-
surement range goes from 10 cm to 400 cm, and the frame rate from 5 fps to 45 fps.
The Picoflexx has a USB 2.0/3.0 interface and does not require any extra power
supply. The camera software directly computes the depth information and shows an
average measurement accuracy of 37 mm, comparable with gold standard ToF cam-
eras, such as the Microsoft Kinect v2 [11].
Fig. 1. An expert user of a Rewalk P5 device, while practicing with the exoskeleton and the
instrumented crutches presented in this work.
3 Results
An expert user of a Rewalk exoskeleton P5 model was measured during a walking

session, using the instrumented crutches, as shown here in Fig. 1. Twenty steps were
recorded and analyzed.
Preliminary results are visible in Fig. 2, were the gait events are coupled with the
point clouds acquired with the ToF camera (only the right foot is represented).
Monitoring Upper Limbs During Exoskeleton-Assisted Gait Outdoors 129
4 Discussion
Without a full kinematic and kinetic analysis of the human-robot system, an accurate
assessment of the shoulder loads during assisted gait is still not possible. The load
exerted on the crutches, however, strongly correlates with upper limbs joint reactions,
and could be used to assess the shoulder involvement while using an exoskeleton.
Fig. 2. Gait events recorded during 20 steps in a walking session of an expert Rewalk P5 user
with synchronous crutches movement, coupled with the point clouds acquired with the ToF
camera. Crutches events are in green, left foot in red and right foot in blue. In the point cloud
images, the red area represents the floor while the blue area represents the right foot.
Fig. 3. Forces acting on the crutches recorded during 20 steps in a walking session of an expert
Rewalk P5 user with synchronous crutches movement. Bold line is average, and the area is a
1-sigma confidence interval.
130 M. Lancini et al.
An example of this data could be seen in Fig. 3, where forces acting on each crutch
are plotted as function of the gait phase.
In the trade-off between the accuracy of a gait lab and easier measurements on the
field, however, is critical to take into account that the exoskeleton user requires some
time to adjust to the robot: limiting the evaluation to short walking tests could lead to
different behaviors than found in practice.
5 Conclusion
The proposed system could be used to improve the design of exoskeleton for spinal
cord injury subjects, allowing a benchmarking of different robotic solutions. While
simulations and experimental gait-lab analysis are the primary tools for assessing the
efficacy and safety of exoskeletons, a proper evaluation should take into account an
environmental setting as close as possible to everyday working conditions.
Being independent from the exoskeleton sensing system, the instrumented crutches
could be considered an external validating device for the exoskeleton under testing.
The lack of markers or connections, moreover, reduces the setup time for walking tests,
and requires no additional instructions to the exoskeleton final user.
More subjects are currently being involved in the study, and a validation of the
system by comparison with a gold standard in different conditions is clearly still
necessary and is currently being scheduled. A real-time version of the proposed
solution is currently under development.
Acknowledgment. We would like to thank the subjects of the study and the hospitals, Domus
Salutis (Brescia, Italy) and Villa Beretta (Costa Masnaga, Italy), for their involvement and
support in the project. We would also like to acknowledge the help and support provided by
Paolo Gaffurini during the tests.
References
1. Contreras-Vidal, J.L., et al.: Powered exoskeletons for bipedal locomotion after spinal cord
injury. J. Neural Eng. 13(3) (2016)
2. Onose, G., et al.: Mechatronic wearable exoskeletons for bionic bipedal standing and
walking: a new synthetic approach, Front. Neurosci. 10 (2016)
3. Esquenazi, A., Talaty, M., Packel, A., Saulino, M.: The ReWalk powered exoskeleton to
restore ambulatory function to individuals with thoracic-level motor-complete spinal cord
injury. Am. J. Phys. Med. Rehabil. 91(11), 911–921 (2012)
4. Zeilig, G., Weingarden, H., Zwecker, M., Dudkiewicz, I., Bloch, A., Esquenazi, A.: Safety
and tolerance of the ReWalk TM exoskeleton suit for ambulation by people with complete
spinal cord injury: A pilot study. J. Spinal Cord Med. 35(2), 101–196 (2012)
5. Fineberg, D.B., et al.: Vertical ground reaction force-based analysis of powered exoskeleton-
assisted walking in persons with motor-complete paraplegia. J. Spinal Cord Med. 36(4),
313–321 (2013)
Monitoring Upper Limbs During Exoskeleton-Assisted Gait Outdoors 131
6. Federici, S., Meloni, F., Bracalenti, M., De Filippis, M.L.: The effectiveness of powered,
active lower limb exoskeletons in neurorehabilitation: A systematic review. NeuroRehabil-
itation 37(3), 321–340 (2015)
7. Álvarez, M.T., et al.: Simultaneous estimation of human and exoskeleton motion: a
simplified protocol. In: IEEE International Conference on Rehabilitation Robotics,
pp. 1431–1436 (2017)
8. Xu, L., Shen, L., Qian, J., Zhang, Y., Wen, Z.: Measuring on coupling force between lower
extremity exoskeleton and subject during rehabilitation training. In: 2011 Fourth Interna-
tional Conference on Intelligent Computation Technology and Automation, vol. 1,
pp. 822–825 (2011)
9. Lonini, L., Shawen, N., Scanlan, K., Rymer, W.Z., Kording, K.P., Jayaraman, A.:
Accelerometry-enabled measurement of walking performance with a robotic exoskeleton: a
pilot study. J. Neuroeng. Rehabil. 13(1) (2016)
10. Lancini, M., Serpelloni, M., Pasinetti, S., Guanziroli, E.: Healthcare sensor system exploiting
instrumented crutches for force measurement during assisted gait of exoskeleton users. IEEE
Sens. J. 16(23) (2016)
11. Fursattel, P., et al.: A comparative error analysis of current time-of-flight sensors. IEEE
Trans. Comput. Imaging 2(1), 27–41 (2016)
What Do People Expect from Benchmarking
of Bipedal Robots? Preliminary Results
of the EUROBENCH Survey
R. Conti1(&), F. Giovacchini1, L. Saccares1, N. Vitiello3, J. L. Pons2,

and D. Torricelli2
1
IUVO S.r.l., Pontedera, Italy
roberto.conti@iuvo.company
2
CSIC, Madrid, Spain
diego.torricelli@csic.es
3
The BioRobotics Institute, Scuola Superiore Sant’Anna, Pontedera, Italy
nicola.vitiello@santannapisa.it
Abstract. In this paper preliminary results of the H2020 EUROBENCH project

are reported. EUROBENCH aims at defining a unified benchmarking framework
for bipedal robotics. In particular, in this paper the structure of the survey promoted
by the EUROBENCH consortium and its initial results are briefly reported.
Objective of the survey is addressing a comprehensive analysis of the priorities for
bipedal walking robots from the stakeholders point of view (e.g. experts, end-
users, etc.) both in the humanoids and in the wearable robotics fields.
1 Introduction
Nowadays, thanks to technological advancements, prostheses, exoskeletons and

humanoids are moving out of the laboratory into everyday applications. In the robotics
community there is a growing awareness on the importance of having clear bench-
marking methods to actually introduce wearable robotics artefacts and humanoid robots
into the society [1, 2]. As [3] sustains, benchmarking is not only the correct approach to
assess and compare the performance of different technologies but it is also a pillar tool
to define and support certifications and standardization requirements necessary to
introduce such robotic technologies in the market.
Nevertheless, a consolidated benchmarking methodology for robotics has not been
reached yet. So far, humanoid locomotion has been benchmarked on a competition-
based approach, where the capability of each robot to complete a predefined task is
investigated [1, 4], while in the wearable robotics domain performance is usually
assessed based on the actual user’s motor function recovered or augmented.
EUROBENCH project has received funding from the European Union’s Horizon 2020 research
and innovation program under grant agreement No 779963.

https://doi.org/10.1007/978-3-030-01887-0_26
What Do People Expect from Benchmarking 133
In this paper, preliminary results of the survey promoted by EUROBENCH project

are reported. The EUROBENCH project is the first European project specifically
focused on designing a unified benchmarking framework for bipedal robotic tech-
nologies. The survey involved different types of stakeholders (e.g. experts, end-users,
etc.) in both humanoid and wearable robotics fields. Collected inputs came from
opinion leaders from academy, industry and end-users, with the ultimate goal of
addressing a comprehensive overview of their actual priorities and expectations on the
benchmarking of bipedal walking robots.
The paper is organized as it follows: Sect. 2 describes in detail the objectives of the
survey and its structure. Section 3 reports the results of the survey, while Sect. 4
proposes some interpretations and preliminary conclusions of the collected data.
2 Objectives and Structure of the Survey
The proposed survey had two specific objectives: (i) establishing the benchmarking
priorities of the stakeholders in the wearable robotic and humanoids communities;
(ii) identifying the needs of the stakeholders for the unified benchmarking framework
that should be considered by the consortium of the project when developing new
benchmarking methods and test benches.
The population of the survey consisted of experts and robotic companies, defined
according to the experience of the EUROBENCH consortium. The survey is available
online at [5]. The structure of the survey consists of ten questions targeting wearable
robotic and humanoids communities. There are mainly multiple choices questions and
a few open questions. It is structured in three parts: (i) short introduction about the
H2020 EUROBENCH project and survey objectives; (ii) profiling of the survey par-
ticipants and (iii) technical questions. In the profiling section, participants are asked to
indicate their main activity areas, their role in the communities and their domains of
application. In the technical section, the questions of the survey focused on the defi-
nition of a priority list of:
(a) the motor tasks generally performed by wearable robots and humanoids;
(b) the performance indicators generally adopted to quantify qualitatively and
quantitatively the actions of wearable robots and humanoids;
(c) of the characteristics that an ideal benchmarking framework should have.
3 Preliminary Results
In this section, only some results from the survey are presented: a more exhaustive
report about the results will be presented in future articles.
As Fig. 1 depicts, a total of 74 people participated to the survey: 56 from the
academy, 14 from the industry sector and 8 from other areas (e.g. end-users, profes-
sional users). In particular, this histogram reports the main activity area of the partic-
ipants of the survey: 46 people belonging to wearable robotics, 24 in walking bipedal
robots and 9 in prosthetics.
134 R. Conti et al.
Figure 2 shows the profile distribution of the participants: 82.4% of the participants
have a R&D role; 8,1% are professional users (e.g. clinician, therapist, etc.); 6,8% are
robot manufacturers; 1,4% are technology providers and 1,4% have an academic role.
Figure 3 depicts the distributions of domains of application of the participants. The survey
clearly highlights that the healthcare sector (e.g. rehabilitation, assistance) is the most
popular domain in the bipedal robotic technologies community, followed by the manu-
facturing area and the consumer domains (e.g. home-based applications, personal care).
Fig. 1. Distribution of the activity area of the 74 participants of the proposed survey.
Fig. 2. Profile of the 74 participants of the survey.
Fig. 3. Domains of application of the survey participants and their percentages.
Among the technical questions presented in the survey, question (b) described in
Sect. 2, it has been considered as one of the most interesting result to report. Indeed, for
each community, the 3-most voted performance indicators are recapped hereafter:
What Do People Expect from Benchmarking 135
– Exoskeleton community: “perform human-like kinetics”, “safety of interaction with

human users” and “ability to perceive the environment/user intention”;
– Humanoids community: “stability”, “ability to make autonomous decisions” and
“ability to minimize (impacts of) failures”
– Prosthetics community: “perform human-like kinetics”, “ability to perceive the
environment/user intention” and “ability to make autonomous decisions”.
Comparing the results of the three communities, it is interesting to note that “ability
to perceive the environment/user intention” is extremely significative for the
exoskeleton and prosthetics communities due to the presence of the human in the loop;
advanced and adaptive control strategies for these types of devices have to take into
account for interpreting the user intention. Another important aspect shown both in
prosthetics and in humanoids fields is the “ability to make autonomous decisions”;
indeed, humanoids and prosthetic devices are usually intended to be autonomous in the
fulfilment of the assigned tasks.
4 Discussions and Conclusions
According to numerical results of the Fig. 1, participants of the survey are mainly
coming from the academic field with a ratio of 76% (the initial contact list counted a
ratio of 1:3 for the industrial sector compared to the academic one); this means that the
efficacy of the survey on the industrial sector should be improved. Possible corrective
actions to be performed in the future will be: increasing the number of companies in the
population in order to uniform the survey and obtaining a significative numerosity, and
trying to involve in different ways the companies to participate to the survey.
Predominance of participants coming from the R&D area (refer to Fig. 2) could
lead to an interesting evaluation in terms of survey population: both the initial contact
list and the dissemination channels that EUROBENCH used to communicate are
limited; only a few robot manufactures and professional users have been reached out.
Consequently, possible corrective actions to perform a representative statistical analysis
will be the increasing of the numerosity of less-represented categories like robot
manufacturer, end-users, professional users, etc.… As Fig. 3 shows, all the proposed
domains of application of the bipedal robotic technologies are represented. This is a
positive result, which proves how the survey was able to collect opinions from all the
domains of application. Predominance of the healthcare sector may be explained by the
fact that it represents the area in which the support of the bipedal robotic technologies
to the end user is more tangible, measurable and ethically accepted by our society.
However, these are the aspects that, in the near future, all the three robotic communities
should take into account to boost the development of the other domains of application
of the bipedal robotic technologies.
136 R. Conti et al.
References
1. Behnke, S.: Robot competitions ideal benchmarks for robotics research. In: Proceedings of
Workshop Benchmarks Robotics Research, pp. 13–17 (2006)
2. del Pôbil, A.P.: Why do we need benchmarks in robotics research? In: Proceedings of
3. Torricelli, D., Gonzalez-Vargas, J., Veneman, J.F., Mombaur, K., Tsagarakis, N.G., del-Ama,
A.J., Gil-Agudo, A., Moreno, J.C., Pons, J.L.: Benchmarking bipedal locomotion: a unified
scheme for humanoids, wearable robots, and humans. IEEE Robot. Autom. Mag. 22 (2015)
4. DARPA robotics challenge trials, 10 December 2010. http://www.theroboticschallenge.org/
5. http://www.eurobench2020.eu/survey
Modeling Human-Exoskeleton Interaction:
Preliminary Results
M. C. Sánchez-Villamañán(&), D. Torricelli, and J. L. Pons
The Neural Rehabilitation Group,

Spanish National Research Council, Madrid, Spain
mcarmen.sanchez@csic.es
Abstract. Physical interfaces have an important role in achieving efficient, safe

and comfortable transmission of forces between the exoskeleton and the human
body. They are normally composed of different compliant elements disposed in
series between the skin of the user and the exoskeleton frame. Modelling how
the compliant properties of physical interface will affect the transmission of
forces may be useful to improve the design process towards more effective, safe
and user-specific exoskeletal devices. As a first step in this direction, we propose
a simplified 2-dimensional model representing the interaction of a single-
actuated-joint exoskeleton with the human limb through a compliant element.
We studied the effects of stiffness value associated to the tissues in the behavior
of the whole system with simulated results of the model.
1 Introduction
Researchers and engineers are continuously proposing new innovative solutions to

improve robotic exoskeletons that assist or rehabilitate human motion [1]. Human-
centered design approaches are becoming more and more popular, due to the increasing
need of solutions that can adapt to specific user’s needs. Anthropometric and kinematic
data is normally considered when designing and selecting the mechanical components
of the exoskeletons. At the same time, adaptive control strategies are implemented to
improve the human-exoskeleton symbiotic performance [2]. In this process, the
physical elements that connect the human with the machine, such as cuffs and straps,
are crucial for transmitting the actuation forces in an efficient, safe, and conformable
manner. Appropriate quantitative method to design these components should be also
considered to achieve a completely human-centered design. The effectiveness and
efficiency of this transmission can be reduced by two main factors [3]: (i) the
misalignments between robot and human joints, and (ii) the compliant properties of
physical elements in series with human soft tissues.
In this contribution, we present a simplified model of force transmission between
the exoskeleton and the human. It characterizes a simple system composed of a
This work is supported by the project EUROBENCH (European ROBotic framework for bipedal
locomotion BENCHmarking) funded by H2020 Topic ICT 27-2017 under grant agreement no:
779963.

https://doi.org/10.1007/978-3-030-01887-0_27
138 M. C. Sánchez-Villamañán et al.
single-actuated-joint of the exoskeleton applying forces on soft tissues; connected to a

static rigid-fixed human limb. We consider different stiffness values of soft tissues and
observe the behavior of the system.
2 Material and Methods
2.1 Interaction model

The theoretical model shown in Fig. 1 is a simplification of human-exoskeleton
interaction through a one-degree-of-freedom system. Finput represents the force
developed by the exoskeleton. Considering the displacement (x) that this force pro-
vokes and the stiffness value of soft tissues (k), Fspring represents the force absorbed by
the human limb. The parameters of the system are also shown in Fig. 1: the inertia (I),
the mass (m) and center of mass of the exoskeleton, the position of the cuff (L) and the
imposed trajectory (Ѳ) of the robotic joint. Thus, the 2-dimensional model represents
the forces developed in the sagittal plane of motion.
Fig. 1. The human-exoskeleton interaction model is composed by a one-degree-of-freedom

system and a single-actuated-joint exoskeleton with a human shank.
2.2 Equations
Assuming the deformation angle of the linear spring is a negligible value, the input
trajectory (Ѳ) is defined as a low amplitude sinusoidal signal. We consider this trajectory,
the inertia of the system and gravity and obtain the input force of the one-degree-of-
freedom model (Fig. 1). The following equations can be derived to characterize the
dynamic behaviour of the system:

d2h
Finput ¼ m g d sin h I 2 =L cos h ð1Þ
dt
Finput Fspring ¼ m €x ð2Þ
Fspring ¼ k x ð3Þ
Pspring ¼ Fspring x_ ð4Þ

Modeling Human-Exoskeleton Interaction 139
In which Eq. (1) describes the horizontal projection of the force developed by the
structure of the exoskeleton, Eq. (2) represents the dynamic behaviour of the system,
Eq. (3) describes the output force developed by the spring due to deformation length
(x) from its equilibrium position and Eq. (4) represents the power developed by the
spring.
3 Results
The previous equations are solved with defined parameter values (see Table 1) using
Matlab. Stiffness value of soft tissues (k2) was assumed from the experimental tests
developed by Frauziols et al. [4].
Fig. 2. (a) Frequency response of the input force developed by the exoskeleton and transmitted
to the one-degree-of-freedom system. (b) Frequency response of the force developed by the
spring with k2. (c) Power developed by the spring and (d) spring deformation considering k1, k2,
k3 and f1.
As shown in Fig. 2(a), we computed the frequency response for the input force
developed by the exoskeleton, for f1, f2, f3 for the sinusoidal input trajectory (Ѳ). In
Fig. 2(b), for each frequency, the fundamental frequency, resulted from the input force
of the system, and the first harmonic, that characterizes the behavior of the spring, are
shown. The highest amplitude values in both Fig. 2(a) and (b) correspond to the
140 M. C. Sánchez-Villamañán et al.
fundamental frequencies with f3 (1.66 106 and 2.41 106). In Fig. 2(c) and (d) the
spring power and the spring deformation are shown for k1, k2, k3 when the joint of the
exoskeleton performs the sinusoidal input trajectory (Ѳ) with the frequency value of
knee joint during flexion while walking (f1) [5]. Maximum spring power values are
2.61 10−3 Nm/s, 2.10 10−5 Nm/s and 3.66 10−7 Nm/s while maximum spring
deformation values are 2.74 10−3 m, 5.06 10−5 m and 6.03 10−6 m corresponding
to k1, k2 and k3 respectively.
4 Discussion
With the results shown in Fig. 2 the role of soft tissues in force transmission can be
argued. The cuff of the exoskeleton interacts with tissues of variable stiffness values
because human topography is non-continuous and non-homogenous. Deformation
values are higher at human limb locations where the tissues are softer while the
absorbed power of these tissues also increases (see Fig. 2(c) and (d)). As a result, force
transmission from the exoskeleton to the user’s joint can depend on the position of the
cuff due to the stiffness of the tissues it covers. In the performed simulations of the
model, we considered a soft tissues stiffness value from experimental results [4] and
two more values, lower and higher, which are constant.
Table 1. Model parameters

Parameter Value
m 1.6 kg
I 0.05 kgm2
d 0.13 m
L 0.26 m
k1, k2, k3 100 N/m, 1400 N/m, 10000 N/m
f1, f2, f3 1 Hz, 1.5 Hz, 2 Hz
Ѳ 0.09 sin(2pft) [rad]
The frequency of input trajectory also affects the forces developed by the system.
Thus, the movement the exoskeleton will assist and its control strategy have to be
considered in order to obtain an efficient force transmission. We considered typical
human walking frequency; however, tests with higher frequency values are interesting
for the study of the stability of the system.
5 Conclusion
We have shown first results of the characterization of the behavior of soft tissues
though a preliminary model of a simple system of single-actuated-joint exoskeleton and
the human shank. Cuffs composed by materials of different stiffness values, depending
on the stiffness value of the tissues they cover, could properly transmit forces from the
Modeling Human-Exoskeleton Interaction 141
exoskeleton to the human. Next steps will be: (i) validating the presented model
through experimental tests, where the force of the spring will be measured with a load
cell, (ii) including more realistic linear and non-linear stiffness values, considering
different values for cuffs and human soft tissues, (iii) including damping (it was not
considered yet due to the simplicity of the model), and (iv) consider dynamic situation
when the shank of the subject is free to move and study the implications for control
purposes.
References
1. Contreras-Vidal, J.L., Bhagat, N.A., Brantley, J., Cruz-Garza, J.G., He, Y., Manley,
Q., Nakagome, S., Nathan, K., Tan, S.H., Zhu, F., Pons, J.L.: Powered exoskeletons for
bipedal locomotion after spinal cord injury. J. Neural Eng. 13(3), 031001–0310017 (2016)
2. del Ama, A.J., Moreno, J.C., Gil-Agudo, Á., Pons, J.L.: Hybrid FES-robot cooperative
control of ambulatory gait rehabilitation exoskeleton for spinal cord injured users. Biosyst.
Biorobotics 11(27), 155–159 (2014)
3. Torricelli, D., del Ama, A.J., González, J., Moreno, J., Gil, A., Pons, J.L.: Benchmarking
lower limb wearable robos: emerging approaches and technologies. In: 8th ACM International
Conference on PErvasive Technologies, Corfu (2015)
4. Frauziols, F., et al.: In vivo identification of the passive mechanical properties of Deep soft
tissues in the human leg. Strain 52(5), 400–411 (2016). Advances in experimental mechanics
for biomedical soft tissues and materials
5. Winter, D.A.: Biomechanics and Motor Control of Human Movement. Wiley, Hoboken
(2005)
Human-in-the-Loop Bayesian Optimization
of a Tethered Soft Exosuit for Assisting Hip
Extension
Myunghee Kim1(&), Ye Ding1,2, Charles Liu1, Jinsoo Kim1,2,

Sangjun Lee1,2, Nikolaos Karavas1,2, Conor Walsh1,2,
and Scott Kuindersma1
1
John A. Paulson School of Engineering and Applied Sciences,
Harvard University, Cambridge, MA, USA
myheekim@uic.edu, myunghee.kim.phd@gmail.com
2
John A. Paulson School of Engineering and Applied Sciences and Wyss
Institute for Biologically Inspired Engineering,
Harvard University, Cambridge, MA, USA
Abstract. Advances in wearable devices have led to an increased need to

develop sophisticated and individualized control strategies. To address this
problem, several researchers have begun exploring human-in-the-loop opti-
mization methods that automatically adjust control parameters in a wearable
device using real-time physiological measurements. A common physiological
measurement, metabolic cost, poses significant experimental challenges due to
its long measurement times and low signal-to-noise ratio. This study addresses
the challenges by using Bayesian optimization—an algorithm well-suited to
optimizing noisy performance signals with very limited data—to perform con-
trol adaptation online. When applied to a soft exosuit designed to provide hip
assistance, optimized control parameters were found in 24 min with a significant
reduction in metabolic cost. These results suggest that this method could have a
practical impact on improving the performance of wearable robotic devices.
1 Introduction
It has been shown that wearable robotic devices reduce the energy expenditure of
human walking [2–5]. However, response variance between participants for fixed
control strategies can be high, leading to the hypothesis that individualized controllers
could further improve walking economy [4, 5]. Recent studies on human-in-the-loop
(HIL) control optimization have shown great potential to reduce metabolic cost by
customizing assistance methods [2, 3, 6]. They also elucidated several practical
This material is based upon the work supported by the Defense Advanced Research Projects
Agency, Warrior Web Program (contract no. W911NF-14-C-0051), the Wyss Institute, the
John A. Paulson School of Engineering and Applied Science at Harvard University and the
Technology for Equitable and Assessable Medicine (TEAM) initiative at Harvard.M. Kim and Y.
Ding—Contributed equally to this work.This manuscript is a shortened version of the recently
published article [1]. Most contents are excerpted from the article including figures.

https://doi.org/10.1007/978-3-030-01887-0_28
Human-in-the-Loop Bayesian Optimization of a Tethered Soft Exosuit 143
challenges, such as long experimental protocols and low signal-to-noise ratios. In this
work, we addressed such challenges by employing a sample-efficient and noise tolerant
method to identify near optimal parameters of wearable devices.
We developed a method that could rapidly identify optimal control parameters in multi-
dimensional space to minimize the metabolic cost of walking (Fig. 1). This method
uses Bayesian optimization, a sample-efficient and noise tolerant global optimization
strategy, well suited for finding the extrema of objective functions that are noisy and
expensive to evaluate [7]. We applied the method to identify the peak and offset timing
of hip extension assistance that minimizes the energy expenditure of walking with a
textile-based wearable device (Fig. 2). Other parameters, namely peak force and onset
timing, were fixed to 30% body weight and at the maximum hip flexion event, based on
previous hip studies [5] and pilot studies.
We initialized the optimization by obtaining the metabolic cost of a prescribed
number of assisted conditions with respect to pseudo-randomly selected control
parameters from an evenly distributed parameter space. Based on this information, the
optimization iteratively estimated the subject’s metabolic cost distribution using a
Gaussian process and selected control parameters of the next iteration by maximizing
expected improvement. For each iteration, the metabolic cost was estimated by fitting a
first-order dynamical model to two minutes of transient metabolic data [2]. After a
prescribed number of iterations, the control parameters corresponding to the minimum
value of the metabolic landscape (mean of the metabolic cost distribution) represented
the optimal values.
Fig. 1. Experimental setup of human-in-the-loop (HIL) optimization. HIL optimization is a

method that automatically adjusts the control parameters of an assistive device based on real-time
measured physiological signals to minimize the metabolic cost of human walking.
144 M. Kim et al.
Fig. 2. Soft exosuit and assistive hip force profile. (A) The hip soft exosuit. A hip extension
moment was generated by pulling the inner cable to create a tension between two anchor points.
(B) Parameterization of hip force profile. The hip force profile was chosen to be a combination of
two parameterized sinusoidal curves joined at the peak. Peak force was set to 30% of body
weight and onset timing was fixed to the time of maximum hip flexion. Peak and offset timing
were actively optimized. (C) Optimized force profiles for subject 3, 4 and 6. Dashed and solid
lines are reference and measured forces normalized by body mass, averaged across ten strides
during the last minute of the validation condition. The maximum hip flexion event was used to
initialize the gait cycle in this study.
Fig. 3. Experimental results. (A) The metabolic energy cost of walking for each condition.
Optimal: minimum mean value of the posterior distribution (metabolic landscape). Validation:
5-min walking with optimized assistance. No-suit: 5-min walking with regular pants. Bars are
means, error bars are standard errors and asterisks denote statistical significance. (B) Feasible
parameter region and optimal timing values for all subjects. Optimized timings were varied
across subjects and three subjects shared optimal settings at the latest peak and offset timing.
We conducted a single-day experiment on eight subjects optimizing the assistance

timings as they walked on a treadmill at 1.25 m s−1. After optimization, we performed
a validation test confirming the optimal condition evaluated during the optimization
process and compared both with a no-suit condition. The primary analysis for outcome
includes (1) the net metabolic cost of walking (2) the convergence time across subjects,
and (3) subject-specific optimal timings.
Human-in-the-Loop Bayesian Optimization of a Tethered Soft Exosuit 145
3 Results
Optimal peak and offset timings were found over an average of 21.4 ± 1.0 min,
ranging from 18 to 24 min. and reduced metabolic cost by 17.4 ± 3.2% compared to
walking without the device (mean ± SEM). Subject-specific optimal peak and offset
timings spread over about half the search range of each control parameter (Fig. 3B).
The variability in the optimized assistance profiles (Fig. 2C) demonstrated the
importance of the individualization. The majority of optimal timings were on the
boundaries of the parameter ranges, with three subjects having their optima at the latest
peak and offset timing. It may suggest that with a larger parameter search area, further
reductions in metabolic cost could be obtained.
4 Discussion
HIL optimization holds promise to improve the performance of wearable robotic

devices for a wide range of tasks. The presented method shows a substantial metabolic
reduction and suggests the possibility of optimizing complex wearable devices using
low-dimensional control parameterization. The short convergence time would enable
researchers to apply this method to individualize control parameters during strenuous
tasks or for people with limited physical strength or endurance [8].
This sample efficient method can be further refined to be applicable for lager
parameter optimization such as multi-joint assistance parameter optimization. One
direction would be coupling estimation and parameter selection, allowing the algorithm
to spend less time refining the metabolic estimates for parameters that are unlikely to
improve performance over the best values observed so far. For this purpose, we can use
an early stopping mechanism and combine it with a standard Bayesian optimization
scheme and metabolic cost estimation method.
Acknowledgment. The authors would like to thank Chih-Kang Chang, Asa Eckert-Erdheim,
Maria Athanassiu, Brice Mikala Iwangou, Nicolas Menard, and Sarah Sullivan for their con-
tributions to this work.
References
1. Ding, Y., et. al.: Human-in-the-loop optimization of hip assistance with a soft exosuit during
walking. Sci. Robot. 3(15), eaar5438 (2018)
2. Zhang, J., et al.: Human-in-the-loop optimization of exoskeleton assistance during walking.
Science 356(6344), 1280–1284 (2017)
3. Lee, S., et al.: Autonomous multi-joint soft exosuit with online optimization reduces energy
cost of loaded walking. JNER 15(1), 66 (2018)
4. Mooney, L.M., et al.: Autonomous exoskeleton reduces metabolic cost of human walking
during load carriage. JNER 11(1), 80 (2014)
5. Ding, Y., et al.: Effect of timing of hip extension assistance during loaded walking with a soft
exosuit. JNER 13(1), 87 (2016)
146 M. Kim et al.
6. Kim, M., et al.: Human-in-the-loop Bayesian optimization of wearable device parameters.

PLoS ONE 12(9), e0184054 (2017)
7. Brochu, E., et al.: A tutorial on Bayesian optimization of expensive cost functions, with
application to active user modeling and hierarchical reinforcement learning. arXiv preprint
arXiv:1012.2599 (2010)
8. Awad, L., et al.: A soft robotic exosuit improves walking in patients after stroke. Sci. Transl.
Med. 9(400), eaai9084 (2017)
A Review of Performance Metrics for Lower
Limb Wearable Robots: Preliminary Results
D. Torricelli1(&), D. Pinto-Fernandez1, R. Conti2, N. Vitiello3,

and J. L. Pons1
1
Neural Rehabilitation Group, Spanish National Research Council,
Madrid, Spain
{diego.torricelli,jose.pons}@csic.es, david.
pinto@cajal.csic.es
2
IUVO S.r.l, Pontedera, Italy
roberto.conti@iuvo.company
3
The BioRobotics Institute, Scuola Superiore Sant’Anna, Pontedera, Italy
nicola.vitiello@santannapisa.it
Abstract. This paper reports a preliminary overview on the existing metrics

and benchmarks for the assessment of lower limb wearable robotics perfor-
mance. This analysis was conducted to identify the current necessities, prefer-
ences, and deficiencies on this topic within the scientific robotics community.
Based on these results, we aim to establish a robust set design criteria of a
comprehensive benchmarking scheme for wearable robotic technology consid-
ering real life scenarios.
1 Introduction
Assessing the performance of wearable robots is a necessary step to demonstrate the

ability of research prototypes to work out of the lab and meet users’ expectations and
needs [1]. To get to this point, tests should not only be performed in typically con-
trolled environment, e.g. treadmill walking, but should take into consideration the great
variability of motor skills composing typical activities of real life scenarios. To
understand the state of the art on this topic, we performed an extensive review of the
literature, focused on the following question: “how is the performance of wearable
robots currently assessed?”. In this paper we report the preliminary results of this
analysis and discuss the main relevant findings emerged so far.
This work is supported by the project EUROBENCH (European Robotic Framework for Bipedal
Locomotion Benchmarking) funded by H2020 Topic ICT 27-2017 under grant agreement no:
779963.

https://doi.org/10.1007/978-3-030-01887-0_29
148 D. Torricelli et al.
We performed an initial search on the Scopus database using the following query string
on paper title, keywords and abstract: (locomot* OR gait* OR walk* OR “body
transport*”) AND (test* OR assess* OR measure* OR benchmark* OR evaluat*)
AND (“wearable robot*” OR exoskelet* OR “powered ortho*”). The resulting 917
papers were filtered by reading titles and abstracts. We excluded publications that did
not assess robotic skills or did not propose metrics for performance evaluation. To the
resulting 309 papers we added other 6 publications resulting from a further search
query aimed to find papers covering specific scenarios: (TITLE-ABS-KEY(balanc* OR
“soft ground*” OR “irregular terrain*” OR slip*)) AND (TITLE(test* OR assess* OR
measure* OR benchmark* OR evaluat*)) AND (TITLE-ABS-KEY (“wearable robot*”
OR exoskelet* OR “powered ortho*”)).
Fig. 1. Percentage of publications (and corresponding number of papers) covering the different
motor skills considered.
These 315 papers were further analyzed, based on the information contained in the
abstracts, focusing on two main aspects: the type of motor skills considered and the
variables used to demonstrate the system performance. As show in Fig. 1, motor skills
have been classified in three main categories, based on our previous taxonomy [2]:
1. Walking skills, which included walking on flat ground, treadmill, slopes, irregular
terrain, soft ground, stairs, slippery surfaces, and during pushes, as well as obstacle
avoiding, and weight bearing.
2. Balance skills, which included balancing on moving surfaces, slopes, during pushes
and during manipulation skills.
A Review of Performance Metrics for Lower Limb Wearable Robots 149
3. Other skills, which included relevant motor functions not included in the previous
two categories, such as lateral stepping, crouching/kneeling, sit-to-stand, and
running.
The performance variables were clustered into the following three categories (see
Fig. 2):
1. Goal-level variables, which included the maximum speed, time, and distance
achievable by the robot, as well as stability, energy efficiency, versatility, and
dependability.
2. Kinematics/Kinetics variables, which included the spatiotemporal parameters,
kinematics and kinetics variables, symmetry, limb coordination, and human
likeness.
3. Human-robot interaction variables, which included metabolic costs, ergonomics,
comfort, muscle efforts, interaction forces, cognitive efforts, and safety.
Fig. 2. Percentage of publications (and corresponding number of papers), proposing the

different performance variables considered.
150 D. Torricelli et al.
3 Results
Preliminary results are shown in Figs. 1 and 2. The most popular motor skill was flat
ground walking, considered by around 42% of the papers reviewed, followed by
treadmill walking, which covers the 18% of the literature. Four motor skills are in the
range between 3% and 6%: slopes (3.5%), stairs (5.3%), weight bearing (5.6), and sit-
to-stand (6.6%). All other motor skills are below 2%.
The analysis on performance indicators showed a more heterogenous picture.
Kinematics and kinetics are considered by half (49%) of the reviewed papers, being the
preferred solution when evaluating the performance of a wearable robot. Goal-level
indicators are considered by the 27% of the literature. Of these, maximum speed, time
and distance are preferred to stability, efficiency or versatility. As for the human-robot
interaction perspective, the most popular indicator is the metabolic cost (13.8%), fol-
lowed by muscle effort (7%) and interaction forces (5%).
4 Discussion
Flat ground and treadmill walking represent the great majority of the motor skills
assessed. This is not surprising, considering that this is the primary functional goal for
exoskeletons. It is worth noting that many other tasks such as slopes, stairs, turning, or
irregular terrain are considered by one order of magnitude less papers. In our opinion,
this gap is not justified by a lower relevance of these motor skills. Everyday life
locomotion is composed of several occurrences of these motor skills, comparable to
steady-state walking. Similarly, balance has been highly disregarded during the eval-
uation of wearable robot performance, while this is an extremely important aspect for
both locomotion and standing purposes. We believe that future research should
strongly focus on these skills when testing the ability of wearable robot technology.
From the performance indicators standpoint, there is a clear bias towards the use of
simple indicators such as speed, time and distance, or the use of traditional gait analysis
metrics based on kinematic, kinetic and spatiotemporal variables. These approaches are
indeed effective in quantifying the walking task from a global perspective. At the same
time, we think that these measures, most of them inherited from traditional human gait
analysis, are not able to fully grasp and quantify the multifaceted performance of
symbiotic systems, composed of human and robotic system. Most of these aspects, all
listed in Fig. 2, are currently considered by less than 6% of the literature reviewed.
These figures have to considerably increase if we, as engineers and scientists, want to
prove the ability of a wearable robot to coordinate with the human, safely and effec-
tively, in daily living scenarios.
5 Conclusions and Future Work
We provided here some preliminary results on a currently ongoing review work on the
state of the art of the assessment of lower limb exoskeletons in research.
A Review of Performance Metrics for Lower Limb Wearable Robots 151
So far, we focused on two aspects, i.e. motor skill and performance variables,
independently. In the future, we will perform a cross-analysis to identify which per-
formance indicators are most used for each motor skill. This will permit to identify a set
of skill-dependent metrics should be considered when evaluating wearable robot
performance.
References1
1. Torricelli, D., del Ama, A.J., González, J., Moreno, J., Gil, A., Pons, J.L.: Benchmarking
lower limb wearable robos: emerging approaches and technologies. In: 8th ACM International
Conference on Pervasive Technologies, Corfu (2015)
2. Torricelli, D., Gonzalez-Vargas, J., Veneman, J.F., Mombaur, K., Tsagarakis, N., Moreno, J.C.,
et al.: Benchmarking bipedal locomotion. a unified scheme for humanoids, wearable robots,
and humans. IEEE Robot. Autom. Mag. 22, 103–115 (2015)
3. http://www.benchmarkinglocomotion.org/david-pinto-werob2018-article-references/
1
The complete list of the 315 papers reviewed, which for reason of space could not be included in this
2-page abstract, is available at [3].
Flexible and Transparent Technologies
for Innovative Wearable Robotics
Development of Polymer Optical Fiber Sensors
for Lower Limb Exoskeletons Instrumentation
Arnaldo G. Leal-Junior1, Anselmo Frizera1(&), Carlos Marques2,

and Maria José Pontes1
1
Graduate Program of Electrical Engineering,
Federal University of Espírito Santo, Vitória, Brazil
arnaldo.leal@aluno.ufes.br, frizera@ieee.org,
mjpontes@ele.ufes.br
2
Instituto de Telecomunicações,
Campus Universitário de Santiago, 3810-193 Aveiro, Portugal
carlos.marques@ua.pt
Abstract. This paper presents the instrumentation of a lower limb exoskeleton

with polymer optical fiber (POF) sensors. The robotic device presents a
potentiometer and an electronic strain gauge (ESG) for the assessment of angle
and human-robot interaction forces, respectively. Such devices are compared
with the proposed POF curvature sensor and a POF-based strain gauge
(POF-SG). The results show a root mean squared error (RMSE) between the
POF curvature sensor and the potentiometer of 1.80° in a measurement ranging
from 10° to 80°, whereas a RMSE of 1.31 Nm was obtained between the ESG
and POF-SG in a range of 0 to 14 Nm. Such results demonstrate the feasibility
of POF sensors as alternative solution for the instrumentation of wearable
robots.
1 Introduction
LOWER limb exoskeletons are applied as a gait assistance device and also for advanced
rehabilitation therapies as an option for human resources optimization in physiotherapy
[1]. The advantages of lower limb exoskeletons in rehabilitation are related to their
higher repeatability, the sensor feedback provides quantitative feedback of the patient
recovery and there is the possibility of customizing the treatment with different levels
of difficulty [2].
Generally, the sensors applied to measure the exoskeleton’s joint angle are encoders
and potentiometers, which need mechanical supports precisely attached to the robotic
device due their sensibility to misalignments [3] and may result in a less compact
This research is financed by CAPES (88887.095626/2015-01), FAPES (72982608), CNPq

(304192/2016-3 and 310310/2015-6). C. Marques acknowledges the financial support from FCT
through the fellowship SFRH/BPD/109458/2015, program UID/EEA/50008/2013 by the
National Funds through the Fundação para a Ciência e a Tecnologia/Ministério da Educação e
Ciência, and the European Regional Development Fund under the PT2020 Partnership
Agreement.

https://doi.org/10.1007/978-3-030-01887-0_30
156 A. G. Leal-Junior et al.
system. Inertial measurement units (IMUs) can also be employed for this end. How-
ever, they are sensitive to electromagnetic field changes, which is undesirable in
devices that presents electromechanic actuators, causing electromagnetic perturbations
that affect IMU performance [4].
In addition, the assessment of human-robot interaction forces are also important on
the development of control strategies for these devices [3]. Electronic strain gauges
(ESG) are commonly applied on the interaction forces estimation [3]. Nevertheless,
there are some drawbacks in this technology: temperature sensitivity, necessity of
frequent calibration and electromagnetic field sensitivity [5].
Optical fiber sensors are compact, lightweight, chemically stable and present
electromagnetic field immunity [6]. Such advantageous features are especially desirable
for wearable robots, which point towards the application of compact structures and
present electromechanic actuators. In addition, polymer optical fibers (POFs) present
higher strain limits, higher fracture toughness, flexibility and impact resistance than
their silica counterparts [6]. The material advantages of POFs enable their application
as curvature sensors in angles as high as 90° and also enable their embedment in
different materials and structures. For these reasons, and, in order to overcome some of
the current technologies limitations, this paper presents the development and applica-
tion of POFs for angle and human-robot interaction forces assessment in a lower limb
exoskeleton.
The wearable device employed in this work is the Advanced Lower Limb Orthosis for
Rehabilitation (ALLOR) [5]. The device is fixed on a chair and allow the user to
perform knee flexion and extension movements. ALLOR provides the assistance within
a predefined angular range. Among the different sensors of the device, there is a
potentiometer that is connected to the motor unit through a toothed belt for angle
measurement. Furthermore, an ESG is employed for the human-robot interaction force
assessment.
The POF sensors proposed are a curvature sensor in the device’s knee joint and two
POF-based strain gauges (POF-SGs), embedded in each shank support of the ALLOR.
The light source of each sensor is a 650 nm@3mW laser and the power variation is
measured by the photodiode IF-D91 (Industrial Fiber Optics, USA).
The operation principle of the POF curvature sensor is the optical power attenuation
caused by the fiber bending, where the operation principle of this sensor is further
described and modelled in [7]. In order to increase the sensor sensitivity and linearity, a
lateral section with 12 mm length and 0.6 mm depth is made on the fiber to remove its
cladding and part of the core [7].
Regarding the POF-SG, the sensor is based on the alignment difference between
two POFs. There are two fibers, where the POF connected on the light source is the
‘illuminated’ fiber. The POF connected to the photodetector is the ‘non-illuminated’
Development of Polymer Optical Fiber Sensors for Lower Limb 157
fiber, where both POF-SGs are embedded in the ALLOR shank support as shown in
Fig. 1 and the interaction force between the user and the robotic device leads to a
misalignment between the illuminated and non-illuminated POFs, resulting in an
optical power variation, acquired by the photodiode. In order to validate the proposed
optical fiber sensors, flexion/extension cycles are made in the range of about 10° to 80°
at two constant angular velocities, namely 0.16 rad/s and 0.21 rad/s and two assistance
levels for the admittance controller. The system with the ALLOR, POF sensors and
electronic sensors is presented in Fig. 1.
Fig. 1. ALLOR robotic device with the electronic and POF sensors.
Figure 2(a) presents the results obtained for the POF curvature sensor, whereas Fig. 2
(b) shows the results of the POF-SGs compared with the ESG. The shank supports are
positioned in opposite sides to obtain the response with one POF-SG in the flexion
movement and the other in extension movement. The results show that both POF
curvature sensor and POF-SG presented good accuracy. The root mean squared error
(RMSE) for the curvature sensor is 1.80°, whereas the POF-SG presented a RMSE of
1.30 Nm. However, these sensors only measures the angle or torque in a single plane
and misalignments in the exoskeleton can lead to errors in the measurement. In
addition, the interaction forces can be measured in different points of the exoskeleton
structure. In this case, the distance between the point of force application and the sensor
position has to be considered, in order to reduce the errors in the torque estimation.
(a)
(b)
Fig. 2. (a) POF curvature sensor response compared with the potentiometer. (b) POF-SG
response compared with the ESG.
4 Conclusion
The application of POF sensors in a lower limb exoskeleton is demonstrated in this

work. The developed sensors are a strain gauge for acquiring human-robot interaction
forces and a curvature sensor for measuring knee joint angles of the robotic device. Both
sensors presented good accuracy with errors below 6% when compared with the elec-
tronic sensors already installed in the device, which are a potentiometer and ESG. POF
sensors presents additional advantages related to their flexibility and electromagnetic
immunity that enable their application not only in rigid robots, but also in the new
generation of robots based on soft and flexible structures.
Development of Polymer Optical Fiber Sensors for Lower Limb 159
References
1. Cai, L.L., et al.: Implications of assist-as-needed robotic step training after a complete spinal
cord injury on intrinsic strategies of motor learning. J. Neurosci. 26(41), 10564–10568 (2006)
2. Kwakkel, G., Kollen, B.J., Krebs, H.I.: Effects of robot-assisted therapy on upper limb
recovery after stroke: a systematic review. Neurorehabil. Neural Repair 22(2), 111–121
(2008)
3. Moreno, J.C., et al.: Wearable Robot Technologies (2008)
4. Leal Jr. A., Frizera Neto, A., Pontes, M.J., Botelho, T.R.: Hysteresis compensation technique
applied to polymer optical fiber curvature sensor for lower limb exoskeletons. Meas. Sci.
Technol. (2017)
5. Leal Jr. A.G., et al.: Polymer optical fiber strain gauge for human-robot interaction forces
assessment on an active knee orthosis. Opt. Fiber Technol. 41, 205–211 (2018)
6. Peters, K.: Polymer optical fiber sensors—a review. Smart Mater. Struct. 20(1), 13002 (2011)
7. Junior, A.G.L., Frizera, A., Pontes, M.J.: Analytical model for a polymer optical fiber under
dynamic bending. Opt. Laser Technol. 93, 92–98 (2017)
T-FLEX: Variable Stiffness Ankle-Foot
Orthosis for Gait Assistance
Miguel Manchola, Daya Serrano, Daniel Gómez, Felipe Ballen, Diego Casas,
Marcela Munera, and Carlos A. Cifuentes(B)
Department of Biomedical Engineering, Colombian School of Engineering Julio

Garavito, Bogotá, Colombia
{miguel.sanchez-m,daya.serrano,daniel.gomez-v,felipe.ballen,
diego.casas-b}@mail.escuelaing.edu.co
{marcela.munera,carlos.cifuentes}@escuelaing.edu.co
Abstract. The development and a preliminary evaluation of a new

active ankle-foot orthosis for gait assistance called T-FLEX are pre-
sented in this paper. The purpose of this device is to support patients
with locomotion disabilities during rehabilitation treatment of the ankle
joint. This device is based on bio-inspired actuation, in which the stiffness
can be adjusted according to a gait phase detection, thereby reproducing
the behavior of antagonistic muscles. A preliminary trial with a healthy
subject (kinematic analysis) reveals an increase in the range of motion
in ankle kinematics, which is desirable for ankle rehabilitation and assis-
tance.
1 Introduction
Approximately 795,000 people suffer stroke in the US each year [1] and the
reported incidence rate of stroke for Colombia in 2014 was 1.31 in 1000 inhabi-
tants, many of which end up with mobility impairments [2]. One of these post-
stroke disabilities is drop foot, which is a neuromuscular disorder that deterio-
rates the patients’ walking ability to move their foot along the sagittal plane.
An ankle-foot orthosis (AFO) is the most commonly used treatment option
for drop foot patients. An AFO is defined as a mechanical device used to prevent
or correct ankle and foot deformities and to improve their functions [4]. As
variable-impedance orthosis has certain clinical benefits for the treatment of
drop-foot gait compared to both unassisted gait and conventional AFOs [5], the
objective of this work is to develop a novel active AFO based on variable stiffness
actuation, and preliminary evaluate its control system on a healthy subject.
The powered orthotic device T-Flex employs an electric actuation mechanism
for producing assistive and resistive movement. This mechanism consists of two
servomotors (Dynamixel MX106T, Robotis, USA) placed on the posterior and
anterior parts of the user’s shank (Fig. 1). The motors are attached in series with
This work was supported by Colciencias (grant 801-2017) and Colombian School of
Engineering Julio Garavito.
https://doi.org/10.1007/978-3-030-01887-0_31
T-FLEX: Variable Stiffness Ankle-Foot Orthosis for Gait Assistance 161
Fig. 1. Experimental setup using the AAFO T-Flex, Inertial sensing and EMG
capturing systems.
tendons (fishing rod coiled around a 2.85 mm Filaflex thread, whose dynamic
behavior makes them bio-inspired), which are attached to the patient’s heel and
forefoot by means of a customized insole.
The primary goal of the design is to reproduce a variable stiffness profile by
manipulating the bio-inspired tendons. These tendons are handled in such a way
that they resemble the behavior of two antagonist muscles, as ankle impedance
is varied in response to walking phase and step-to-step gait variations. T-Flex
implements the gait phase detection by means of an inertial sensing system
(BNO055, Bosch Sensortec, Germany).
Control strategies are mainly targeted towards the prevention of drop foot in
swing phase and slap foot at heel strike. During midstance phase, both tendons
remain at maximum tension to provide stability. Subsequently, joint impedance
is minimized so as not to impede powered plantar flexion movements during
late stance. Finally, during the swing phase, the posterior motor lifts the foot to
provide toe clearance.
2 Methods
One neurologically intact adult who did not suffer from any ankle-related injury
participated in this study (male, 22 yr, 75 kg, 1.77 m). Zero and variable stiffness
control strategies were evaluated on a conventional treadmill and the results were
compared to the condition without wearable device (5 min for each condition at
self-selected gait speed: 0.64 m/s). The order of the evaluation conditions was
randomized, according to the experimental protocol approved by a local ethics
committee.
Gait evaluation included simultaneous recording of kinematic data using an
inertial sensing system (100.21 Hz), and muscular activation of anterior tibialis
(AT) and medial gastrocnemius (MG) using an surface EMG (sEMG) system
(512 Hz) (Consensys IMU Development Kit, Shimmer Sensing, Ireland). The
experimental setup is shown in Fig. 1.
162 M. Manchola et al.
The recording of kinematic data was undertaken following the protocol

described in [6] and the results were filtered using a 2nd order bandpass But-
terworth filter with cut-off frequencies of 1 and 50 Hz. Ranges of motion (ROM)
along the sagittal plane were determined on the basis of the difference between
the maximum and minimum of each joint angular position.
3 Results
Figure 2(a) shows ankle kinematic behavior for each evaluation condition,
expressed in terms of the ankle dorsi-plantar flexion ROM. Figure 2(b) shows
root mean square (RMS) values acquired after thorough processing of the sEMG
signals for each selected muscle. Table 1 shows ROMs of the knee and hip joints
for the flexo-extension and add-abduction degrees of freedom.
Table 1. ROMs of knee and hip flexo-extension (FE) and add-abduction (AA) [degrees]
Movement Unassisted Passive Active

Right Hip FE 26.58 13.56 16.66
Left Hip FE 25.22 10.53 10.90
Right Knee FE 40.88 32.23 42.97
Left Knee FE 60.05 23.68 27.97
Right Hip AA 15.32 20.39 14.13
Left Hip AA 13.93 9.09 10.20
Right Knee AA 48.60 41.05 42.74
Left Knee AA 48.60 41.05 42.74
4 Discussion
Analysis of kinematic data reveals that the kinematic behavior of the left ankle
does not experience a drastic change among evaluation conditions, whereas that
of the right ankle presents a noticeable change, i.e. an increase of the dorsi-
plantar flexion ROM at this joint while using the device (Fig. 2(a)). This change
is an indication that the device could be beneficial for patients that have a
reduced ROM related to foot-drop. Although the device is built to improve the
ankle kinematics, an assessment of the knee and hip kinematic was performed
(Table 1). Clearly, there is a change in kinematic gait performance using the
device, as a decrease of ROM was found for the most knee and hip joint move-
ments. This kinematic behavior could be attributed to a compensation strategy
at the knee and hip joints [7].
By comparing the unassisted and the assisted gait condition on passive mode,
it emerges that the RMS value of the EMG does not present any significant
T-FLEX: Variable Stiffness Ankle-Foot Orthosis for Gait Assistance 163
Fig. 2. (a) Ankle kinematics. ROMs along the sagittal plane at the ankle joints.
(b) Analysis of sEMG. RMS values of anterior tibialis and medial gastrocnemius.
Abbreviations: RMG: Right medial gastrocnemius, LMG: Left medial gastrocnemius,
RAT: Right anterior tibialis, LAT: Left anterior tibialis
difference between these conditions, except for the left MG (Fig. 2(b)). Since
the active orthosis is worn on the user’s right leg, muscle electrical activity on
right MG might have increased as an effect of its implementation. Furthermore,
the fact that the use of T-Flex on active mode causes an increase of muscle
electrical activity becomes visible, as the RMS value in every monitored muscle
is higher while the orthosis is assisting the patient actively, compared to the
other evaluation conditions (except for the left MG in passive assistive gait).
Thus, in overall terms, the use of a variable stiffness orthotic device causes an
increase of muscle electrical activity, as previous research has already found [8].
This findings are in agreement with the higher ROM found in the ankle.
5 Conclusion
Overall, although the device improves kinematic performance at the ankle joint,
it affects hip and knee kinematics. Besides, concerning muscle electrical activity,
it was found that the use of the device on active mode increased the RMS value
of all studied muscles. The preliminary evaluation of this control system was an
important step and brings the ongoing development process closer to the goal of
creating an intuitive and robust control methodology for an ankle rehabilitation
robot. For future works, a statistical study with an appropriate sample and the
recording of kinetic data is expected to be conducted, in order to perform a
better evaluation of whether there are significant differences in the gait pattern
for the three studied conditions.
164 M. Manchola et al.
References
1. Benjamin, E.J., Blaha, M.J., Chiuve, S.E., Cushman, M., Das, S.R., Deo, R., et al.:
Heart disease and stroke statistics-2017 update: a report from the american heart
association. Circulation 135, e146–e603 (2017)
2. Chirveches-Calvache, M.A.: Pontificia Universidad Javeriana, Estimación de la carga
de enfermedad cerebrovascular en Colombia para el ańo 2014 (2016)
3. Perry, J., Burnfield, J.M., Cabico, L.M.: Gait Analysis: Normal and Pathological
Function. SLACK, Thorofare (2010)
4. Wu, K.K.: Foot Orthoses: Principles and Clinical Applications. Williams & Wilkins,
Baltimore (1990)
5. Yamamoto, S., Ebina, M., Iwasaki, M., Kubo, S., Kawai, H., Hayashi, T.: Compar-
ative study of mechanical characteristics of plastic AFOs. J. Prosthet. Orthot. 5,
47–52 (1993)
6. Vargas-Valencia, L., Elias, A., Rocon, E., Bastos-Filho, T., Frizera, A.: An IMU-to-
body alignment method applied to human gait analysis. Sensors 16, 2090 (2016)
7. Singer, M.L., Kobayashi, T., Lincoln, L.S., Orendurff, M.S., Foreman, K.B.: The
effect of ankle-foot orthosis plantarflexion stiffness on ankle and knee joint kinemat-
ics and kinetics during first and second rockers of gait in individuals with stroke.
Clin. Biomech. 29, 1077–1080 (2014)
8. Harper, N.G., Esposito, E.R., Wilken, J.M., Neptune, R.R.: The influence of
ankle-foot orthosis stiffness on walking performance in individuals with lower-limb
impairments. Clin. Biomech. 29, 877–884 (2014)
A Series Elastic Dual-Motor Actuator
Concept for Wearable Robotics
Tom Verstraten(B) , Raphaël Furnémont, Pablo López-Garcı́a, Stein Crispel,

Robotics and Multibody Mechanics Research Group (R&MM), Faculty of Mechanical

Engineering, Vrije Universiteit Brussel and Flanders Make, Pleinlaan 2, 1050 Elsene,
Belgium
Tom.Verstraten@vub.be
Abstract. Series Elastic Actuators (SEAs) are used extensively in the

field of wearable robotics because of their energy efficiency. Redundant
drivetrains enable a further reduction in electrical energy consumption,
as they use the actuator’s motors in a more energy efficient way. In
this work, we present a Series Elastic Dual-Motor Actuator (SEDMA), a
kinematically redundant actuator with series elasticity. We simulate its
use in an ankle prosthesis and compare its energy efficiency to that of a
traditional SEA. Results indicate an energy reduction of 16% compared
to the SEA.
1 Introduction
The use of Series Elastic Actuators (SEAs) has sparked major advances in the
development of robotic prostheses, exoskeletons and other wearable robots. A
properly tuned series spring can reduce motor speed requirements and, as a side-
effect, the energy-efficiency of the actuator. In wearable robotics, however, the
motor also tends to be operated at low powers [1], where it is the least efficient.
Moreover, in an SEA, the motor bears the entire output torque, resulting in the
need for high gear ratios which lead to a high reflected inertia of the drivetrain.
This reflected inertia is the cause of high electrical power peaks which often
occur during late stance and swing [2]. A potential solution to these problems is
the introduction of a redundant drivetrain. Continuously variable transmissions,
for example, have been identified as having great potential for ankle and knee
prostheses and exoskeletons, especially when combined with energy buffers such
as springs and flywheels [3,4]. In this work, we study the energetic benefits of an
actuation concept which combines springs and kinematic redundancy: the Series
Elastic Dual-Motor Actuator (SEDMA).
Tom Verstraten is a postdoctoral fellow of the Research Foundation Flanders (FWO).

Part of this work was funded by the European Commission ERC Starting grant
SPEAR (no. 337596).
https://doi.org/10.1007/978-3-030-01887-0_32
166 T. Verstraten et al.
2 Series Elastic Dual-Motor Actuator

A schematic of the SEDMA is shown in Fig. 1. The dual-motor actuator itself
was described in [5], where its energy efficiency was studied. Its main component
is a planetary differential. The inputs of the differential are its ring and sun gear
wheel, its output is the carrier. A drivetrain is connected to each input, according
to the specifications listed in Table 1. Finally, the output is connected to the
load through a series elastic element with stiffness ks = 382 Nm/rad. This is the
energy-optimal value for an ankle prosthesis designed for walking according to
[6].
For comparison with the SEDMA, the electrical power and energy consump-
tion of a reference SEA driven by a 200 W Maxon RE50 DC motor with 257:1
reduction will also be calculated. This drivetrain is the smallest drivetrain which
enables the SEA to perfectly track the imposed output [6]. Its specifications can
be found in Table 1 as well.
Fig. 1. Schematic of the dual-motor actuator with its main parameters.
3 Methods
3.1 Equations
The relationship imposed by the series spring is
To = TC = kS (θC − θo ) (1)
The two input speeds θ̇S and θ̇R of the sun and ring motor, respectively, are
linked to the output speed of the carrier θ̇C :
1 θ̇
1 ρ S
θ̇C = (2)
nC nS (1+ρ) nR (1+ρ) θ̇R
A Series Elastic Dual-Motor Actuator Concept for Wearable Robotics 167
Table 1. Specifications of sun, ring and reference drivetrain.
Sun (S) Ring (R) Reference

Gear ratio nλ 1:1 (direct) 33:1 257:1
Gear efficiency ηλ 100% 75% 68%
2
Drivetrain inertia Jλ (gcm ) 33.5 35.1 575
Viscous friction νλ (Nms) 4.63E-7 4.63E-7 1.46E-6
Nominal current Iλnom (A) 3.47 3.47 10.8
Max. speed ωλmax (rpm) 12 000 12 000 9500
Winding resistance Rλ (Ω) 0.611 0.611 0.103
Torque const. kT λ (mNm/A) 25.9 25.9 38.5
where nS and nR are the reductions of the gearboxes of the sun and ring drive-
train specified in Table 1; nC = 40:1 that of the gearbox on the carrier (efficiency
ηC = 98%). The reduction offered by the planetary differential is characterized
by ρ. In this work, ρ = 7.
The motor torques are given by

CC CS 1
TmS = δS JS θ̈S + νS θ̇S + TC (3)
nC nS ρCP G + 1

CC CR ρCP G
TmR = δR JR θ̈R + νR θ̇R + TC (4)
nC nR ρCP G + 1
where we defined the efficiency functions of the gearboxes
−sign(Toλ θ̇λ )
Cλ = ηλ (5)
and the efficiency function of the planetary differential:

θ̇
sign TC nS −θ̇C
S
CP G = ηP G (6)
with λ = R, S, C an index corresponding to the ring, sun and carrier, respectively,

and with ηP G = 97% the meshing efficiency of the composing gear pairs. An
explanation and derivation of these formulas can be found in [5]. The parameters
of the actuator are listed in Table 1. The parameters δλ (λ = R, S) represent
the brakes: when the ring brake is engaged, δR = 0; an engaged sun brake
corresponds to δS = 0. In all other cases, δλ = 1.
The electrical power Pelec,λ of a drivetrain is the product of its current Iλ
and voltage Uλ . They are determined by means of the motor model

Iλ = Tmλ /kT λ
(7)
Uλ = kT λ θ̇λ + Rλ Iλ
168 T. Verstraten et al.
Finally, the electrical energy consumption of the SEDMA is

(Pelec,S + Pelec,R ) dt (8)
cycle
3.2 Optimization
The optimization software GPOPS is used to optimize the SEDMA and the
reference SEA for walking at a natural cadence. The imposed output torque and
speed correspond to Winter’s gait data for the human ankle, for a 75 kg person
[7]. The speed distribution and the braking instants are optimized according to
the method presented in [8]. Motor voltage is limited to 48 V. According to the
specifications of the manufacturer, motor speed is limited to ωλ,max and the rms
value of Tmλ to Tλnom (see Table 1).
250
Biological
200
SEA (ref.)
SEDMA
150
Power (W)
100
50
0
Sun brake engaged Ring brake engaged
-50
0 20 40 60 80 100
% gait cycle
Fig. 2. Electrical power consumption of the reference SEA (red) and the SEDMA
(green). The biological power profile of the ankle is shown in blue. The period during
which only the sun motor is used (ring brake engaged) is shaded; only the ring motor
is used in the white area (sun brake engaged). The power consumption of the SEDMA
is particularly lower than that of the reference SEA during late terminal stance.

The electrical powers of the reference SEA and the SEDMA are shown in Fig. 2.
Interestingly, the two motors of the SEDMA are never used at the same time. The
SEDMA offers a considerable reduction in electrical power during late terminal
stance (42–52% GC). This is reflected by an overall decrease in electrical energy
consumption. The SEDMA consumes 30.3 J/cycle, a reduction of 16% w.r.t. the
reference SEA (36.2 J/cycle). This can be explained by the fact that the two
motors of the SEDMA can be optimized for a specific part of the gait cycle.
This has several advantages. First, it enables the use of smaller and therefore
more efficient gearboxes (see Table 1). Secondly, the motors are operated less
A Series Elastic Dual-Motor Actuator Concept for Wearable Robotics 169
in their inefficient low-power regions. And finally, high inertial torques can be
avoided by deploying a low-inertia drivetrain in the low-torque phases of the gait
cycle. A high-inertia drivetrain (i.e. high gear ratio, high motor torque, so high
output torque) can be deployed when higher torques are required. Although this
advantage is not exploited by the proposed SEDMA – high power peaks still
occur during swing in Fig. 2 – an optimized design can probably do better.

The optimization presented in this work proves the potential of dual-motor archi-
tectures for reducing the energy consumption of series elastic actuators. In future
work, we will evaluate other combinations of motors, gearboxes and springs.
By optimizing them concurrently, we expect even greater reductions in energy
consumption.
References
1. Sugar, T.G., Holgate, M.: Compliant mechanisms for robotic ankles. In: ASME
International Design Engineering Technical Conferences and Computers and Infor-
mation in Engineering Conference, V06AT07A002 (2013)
2. Verstraten, T., Geeroms, J., Mathijssen, G., Convens, B., Vanderborght, B., Lefeber,
D.: Optimizing the power and energy consumption of powered prosthetic ankles with
series and parallel elasticity. Mech. Mach. Theor. 116, 419–432 (2017)
3. Aló, R., Bottiglione, F., Mantriota, G.: An innovative design of artificial knee joint
actuator with energy recovery capabilities. J. Mech. Robot. 8(1), 011 009 (2015)
4. Lenzi, T., Cempini, M., Hargrove, L.J., Kuiken, T.A.: Actively variable transmission
for robotic knee prostheses. In: IEEE International Conference on Robotics and
Automation (ICRA), pp. 6665–6671 (2017)
5. Verstraten, T., Furnémont, R., Lopez-Garcia, P., Rodriguez-Cianca, D., Cao,
H.-L., Vanderborght, B., Lefeber, D.: Modeling and design of an energy-efficient
dual-motor actuation unit with a planetary differential and holding brakes. Mecha-
tronics 49, 134–148 (2018)
6. Verstraten, T., Flynn, L., Geeroms, J., Vanderborght, B., Lefeber, D.: On the elec-
trical energy consumption of active ankle prostheses with series and parallel elastic
elements. In: IEEE International Conference on Biomedical Robotics and Biomecha-
tronics (BioRob) (2018, in Press)
7. Winter, D.A.: Biomechanics and Motor Control of Human Movement, 4th edn.
Wiley, New York (2009)
8. Furnémont, R., Mathijssen, G., Verstraten, T., Jimenez-Fabian, R., Lefeber, D.,
Vanderborght, B.: Novel control strategy for the +SPEA: a redundant actuator
with reconfigurable parallel elements. Mechatronics 53, 28–38 (2018)
Towards Design Guidelines for Physical
Interfaces on Industrial Exoskeletons:
Overview on Evaluation Metrics
M. Sposito1,2(B) , S. Toxiri2 , D. G. Caldwell2 , J. Ortiz2 , and E. De Momi1

1
Department of Electronics, Information and Bioengineering (DEIB),
Politecnico di Milano, Milan, Italy
matteo.sposito@iit.it
2
Advanced Robotic Department (ADVR), Istituto Italiano di Tecnologia,
Genova, Italy
Abstract. On exoskeletons, physical interfaces with the body are one

of the key enabling component to promote user acceptance, comfort and
force transmission efficiency. A structured design workflow is needed for
any application-driven product, such as industrial exoskeletons. In this
paper, we review objective and subjective evaluation metrics that can be
applied to physical interfaces. These indexes can be evaluated to create
an ordered list of requirements to guide their future design. Pressure
magnitude, duration, distribution, direction and time to don and doff are
relevant objective indexes related to interfaces. Pain, comfort and ease of
operation are subjective indexes. We propose that collecting a suitable
set of metrics will lay the foundation for effective design guideline for
industrial exoskeletons.
1 Introduction
Force augmenting exoskeletons are useful devices in factories. There is grow-

ing evidence of their effectiveness on lowering risk of work-related muscu-
loskeletal disorders (WMSDs) and increasing comfort of operation in certain
physically demanding tasks characterized by poor ergonomics (e.g. manual
material handling, overhead assembly) [1]. Therefore, several companies in
conjuction with research institutes put their efforts to develop exoskeletons.
There are some models commercially available such as Model Y (Atoun, Nara
City, Japan), HAL for Labour Support (Cyberdyne, Tsukuba, Japan), Ekso
Works (Ekso Bionics, Richmond, CA, U.S.A.) and Laevo Back Support (Laevo,
Delft, The Netherlands) [2].
For widespread adoption of exoskeletons in industrial environments, several
aspects need to be optimized. Here we focus on exoskeleton physical interfaces.
With physical interfaces we refer to braces, cuffs or any other attachment to
This work has been founded by the Italian Workers’ Compensation Authority
(INAIL).
https://doi.org/10.1007/978-3-030-01887-0_33
Towards Design Guidelines for Physical Interfaces on Industrial Exoskeletons 171
the wearer’s body. An interface is responsible for the transmission of assistive

forces from the actuators and the overall wearing comfort. In [3], up to 50%
of exoskeleton power was reportedly lost due to the physical interface dynam-
ics, dissipating the force in shear stresses, compression and misalignment over
the body. Moreover, this inefficiency generates discomfort to the end user, com-
promising acceptance of the device. Therefore, design criteria for exoskeleton
interfaces are desiderable.
However, listing mechanical and comfort requirements for an exoskeleton
interface is challenging. Will the mechanical requirements for an optimal force
transmission agree with user comfort? What are the key factors affecting com-
fort? What metrics should be used for subjective evaluations? This paper is
divided in: state of the art on attachment design and evaluation methodologies,
list of objective and subjective evaluation metrics and conclusions.
2 State of the Art

2.1 Design
Interface design for industrial exoskeletons does not typically differ from attach-
ments design for medical exoskeletons, even if the two devices address different
users. In fact, people in need of a medical exoskeleton typically have muscular
impairments, while industrial exoskeletons are worn by healthy subjects. In the
first case, it is important to have it secured to the limb, as the robot applies larger
torques, most of the torque that a limb needs to be moved. In the latter case,
devices have to deal with healthy muscles that vary shape and stiffness during
movements. In fact [4] shows evidence that pressure exerted on tissues varies with
different movements, i.e. the back thigh plate recorded up to 9.5 N single-point
pressure during squat and leading leg lunge movements. Therefore, a structured
set of requirements derived from the indexes that are used to evaluate physical
interfaces is desiderable. Indeed technological advancements and research have
been mainly focused on exoskeleton actuation and mechanical design, with com-
paratively few advancements in attachments [4]. Having a priority ordered list
of requirements based on metrics will help further technical advancements of the
interfaces and ultimately promote the adoption of industrial exoskeletons.
2.2 Evaluation Methodologies

Interfaces to the human body, as a part of the exoskeleton itself, are tested and
assessed with the whole device. Static (donning and doffing) and dynamic tests
are run. In [4] and [5] wearing comfort is assessed through pressure acquisition at
the interfaces with the body. To record pressures, in [4], a custom sensing system
is used at every attachment point. In [5], an external commercial pressure mat is
used. [6] presents a viable methodology to acquire crucial information on mutual
influence of kinematic constraints, reaction forces, attachment pressure and sub-
jective exoskeleton performances. Pressure on the limb is obtained from pres-
sure fed to air cushions mounted on the inside of the attachments. Evaluation of
172 M. Sposito et al.
exoskeletons also needs to take into account comfort related to wearability, that
is another challenging task. In [7], the authors present a framework to evaluate
lower limb exoskeleton, focusing on wearing comfort and interfaces indexes, i.e.
ease of donning/doffing, aestethic design or attachment types (straps or velcro
bands).
3 Metrics
Two types of indexes are commonly used to evaluate exoskeleton performances:

objective and subjective indexes. The former refer to values that are measured
by sensors, the most used are Circumferential and Single-Point Pressure Mag-
nitude, Distribution, Duration, Direction and time to don/doff the attachment.
The latter refers to values that are obtained by mean of interviews and convey
information about perceived sensation such as Perceived Pain, Pressure Detec-
tion Threshold (PDT) and Pressure Tolerance Threshold (PTT), Perceived Com-
fort, Mental Load, Physical Load and Ease of Use.
Circumferential and single-point pressures are measured by resistive or capac-
itive sensors (e.g. Force-Sensing Resistors FSR) placed between the limb and the
interface. However both type of sensors are affected by different problems that
become relevant to wearable pressure sensing: FSRs suffer from drift caused by
prolonged pressure and need a rigid interface to rest upon, capacitive sensors
need high frequency and complex electronics.
Large variability (intra and inter-subject) in shape and dimension due to
muscular contraction; pressure values at rest should be taken into account [4].
In addition, pressure thresholds are not clearly defined. 32 mmHg (4.3 kPa) is
enough to block superficial capillary flow, but pressure during seat is well above
that limit (22 kPa), thus suggesting a compensatory effect of the vascular sys-
tem [8]. Pressure distribution and direction are central to force transmission.
Wrong direction could lead to shear stresses, interface movements and misalign-
ments. Wrong distribution could lead to dissipation of forces in soft tissues [8].
Duration of pressure is related to tissue damage and discomfort. The body can
adapt to prolonged single-point and circumferential pressures leading to reduced
awareness of tissue damage. For this reason circumferential pressures are more
dangerous as pressure thresholds blocking blood flow are lower in comparison
to single-point ones [8]. Intermittent pressure at low frequencies (e.g. 0.3 Hz)
becomes immediately unbearable due to accumulation of sensory stimuli (Tem-
poral Summation of Pain) [8]. The last listed index is the time to don and doff
the device, which can be used to guide fixation mechanism and position design.
Subjective indexes are mostly related to perceived wearing comfort. Fabric
pattern, breathability or cuff ergonomics are among key factors that enhance or
decrease user acceptance of the whole device. Subjective indexes are obtained
by means of structured interviews. We can divide these indexes in Usability
(Mental Load, Physical Load and Ease of Use), Pain (PDT, PTT), and Comfort
(Perceived Comfort). Usability index quantifies how the system is natural and
easy to use. Perceived ease to don and doff the attachments can be evaluated
Towards Design Guidelines for Physical Interfaces on Industrial Exoskeletons 173
with the System Usability Scale [5]. Since industrial exoskeletons should improve
ergonomics of different working tasks, it is central that these devices do not
hinder workers in their routine. An indication on how interfaces affects usability
is the alteration of the ability to fulfill a certain task. Nasa Task Load Index
(TLX) is used in [6] to quantify how pressure on limbs alters the perception
of ease to accomplish a determined movement. In addition, it is proven that
a preferred interval of attachment pressure exist, in which users scored higher
values in TLX metric [6]. Pain and discomfort can be evaluated through Visual
Analougue Scale (VAS) metric, thus estimating PDT and PTT for each interface
and user, perceived ease to don and doff, fabric tactile feedback and perceived
breathability.
4 Conclusion
Exoskeleton mechanical design and actuation have been object of technical and
technological advancements, however this has not happened for physical inter-
faces [4]. In this overview, we briefly present the state of the art metrics to eval-
uate performances of attachments for industrial exoskeletons. Evaluation metric
are divided into objective (values from sensors) and subjective (perceived sen-
sations) indexes, used to quantify operational capabilities of interfaces. Since a
viable ordered guideline for attachments design has not been proposed yet, our
future work will focus on experimental evaluation of how attachments dimen-
sions, positions, stiffness and materials affect objective and subjective indexes.
This will translate objective indexes and feedback from users into design guide-
lines (e.g. shape, dimensions and position of body attachments) to improve
dynamic and static interaction between human body and exoskeletons, increas-
ing attachments comfort, ease of use, force transmission efficiency and ultimately
promoting exoskeletons adoption.
References
1. de Looze, M.P., Bosch, T., Krause, F., Stadler, K.S., O’Sullivan, L.W.: Exoskele-
tons for industrial application and theri potential effects on physical work load.
Ergonomics 59 (2016)
2. Sugar, T., Veneman, J., Hochberg, C., Shourijeh, M.S., Acosta, A., Vazquez-Torres,
R., Marinov, B., Nabeshima, C.: Hip Exoskeleton Market - Review of Lift Assist
Wearables. Wearable Robotics Association
3. Yandell, M.B., Quinlivan, B.T., Popov, D., Walsh, C., Zelik, K.E.: Physical interface
dynamics alter how robotic exosuits augment human movement: implication for
optimizing wearable assistive devices. J. NeuroEng. Rehabil. (2017)
4. Levesque, L., Pardoel, S., Lovrenovic, Z., Doumit, M.: Experimental comfort assess-
ment of an active exoskeleton interface. In: Robotics and Intelligent Sensors (IRIS)
(2017)
5. Huysamen, K., de Looze, M., Bosch, T., Ortiz, J., Toxiri, S., O’Sullivan, L.W.:
Assessment of an active industrial exoskeleton to aid dynamic lifting and lowering
manual handling tasks. Appl. Ergon. (2018)
174 M. Sposito et al.
6. Schiele, A., van der Helm, F.: Influence of attachment pressure and kinematic con-
figuration on pHRI with wearable robots. Appl. Bionics Biomech. 6 (2009)
7. Bryce, T.N., Dijkers, M.P., Kozlowski, A.J.: Framework for assessment of the usabil-
ity of lower-extremity robotic exoskeletal orthoses. Am. J. Phys. Med. Rehabil.
(2015)
8. Kermavnar, T., Power, V., de Eyto, A., O’Sullivan, L.W.: Computerized cuff pres-
sure algometry as guidance for cirfumferential tissue compression for wearable soft
robotic appliance: a systematic review. Soft Robot. 5 (2018)
Design and Control of a Transparent
Lower Limb Exoskeleton
Wilian M. dos Santos1,2(B) and Adriano A. G. Siqueira1,2

1
Department of Mechanics Engineering, Engineering School of São Carlos,
University of São Paulo (USP), São Carlos 13566-590, Brazil
{wilianmds,siqueira}@sc.usp.br
2
Center for Advanced Studies in Rehabilitation,
and Center for Robotics of São Carlos,
University of São Paulo (USP), São Carlos 13566-590, Brazil
Abstract. This paper deals with the design and evaluation of a modular
exoskeleton for rehabilitation of lower limb movements. The exoskeleton
is composed of lightweight tubular structures and six free joints that pro-
vide actuation and configuration modularity to the system. Experiments
considering the interaction between a healthy subject and the exoskele-
ton are performed to evaluate the influence of the exoskeleton structure
on kinematic and muscular activity profiles during walking. Also, an
optimal impedance controller for exoskeletons was evaluated considering
the modular exoskeleton.
1 Introduction
Recently, a large number of lower limb exoskeletons for assistance and rehabilita-
tion have been developed and reported in the literature [1]. The main objective
of the assistive exoskeletons is to support patients who have suffered complete
spinal cord injury (SCI), in which there is no possibility of recovery of the move-
ments. On the other hand, exoskeletons for rehabilitation, developed for stroke
or incomplete SCI patients, seek to promote cortical motor reorganization in
order to improve patients’ walking patterns.
However, few of the already proposed lower limb exoskeletons have flexibility
to be used with different kinds of actuators or be set-up to assess only indepen-
dently one or more joints of the patient. In this paper, it is presented a novel
modular lower limbs exoskeleton for rehabilitation, which presents modular char-
acteristics both in the possibility of working one or more joints of the patient
and in the form of activation of joints of the robotic system.
This work was supported by Coordination for the Improvement of Higher Educa-
tion Personnel (CAPES), Support Program for Graduate Studies and Scientific and
Technological Research for Assistive Technology in Brazil (PGPTA), under grant
3457/2014.
https://doi.org/10.1007/978-3-030-01887-0_34
176 W. M. dos Santos and A. A. G. Siqueira
2 Modular Lower Limb Exoskeleton

The modular lower limb exoskeleton, shown in Fig. 1, is composed of a pelvic
belt, VelcroR
straps to attach the human leg to the exoskeleton structure and a
pair of custom shoes [2]. The exoskeleton’s joints were designed to allow an easy
and fast coupling of different modules to alter its functionality, characterizing the
modularity of actuation, that is, it can provide assistance by means of an active
device, for example, a motor coupled to a gearbox, or by a passive device, such
as springs and dampers. The joints are designed to cover all ranges of motion
required for different tasks. The maximum range of motion is ±120◦ , however, it
can be limited by adjustable end-stops in order to prevent joint hyperextension.
Fig. 1. Modular lower limb exoskeleton.
The exoskeleton links are telescopic and can therefore be adjusted longitudi-
nally to align the joints of the exoskeleton with the patient joints. The adjustment
is obtained by means of two hose clamps positioned at the outer tube ends of the
telescopic mechanism. Furthermore, the proposed telescopic tubular links allow
removing the exoskeleton joints corresponding to the patient joints that will not
be treated during the session, thus satisfying the requirement of modularity of
configuration.
The interfaces connecting exoskeleton and patient (pelvic belt and Velcro R
straps) are attached to the links of the exoskeleton by means of linear bearings.
The linear bearing has the function of compensating misalignments between
exoskeleton and patient joints. In many cases, considering the movement of a
patient’s natural gait and the possible movements of an exoskeleton, there is a
misalignment between rotation axis of the patient joints and the corresponding
axis of the exoskeleton.
Design and Control of a Transparent Lower Limb Exoskeleton 177
3 Exoskeleton Evaluation and Control

A set of experiments were carried out to evaluate the influence of the exoskeleton
structure on kinematic and muscular activity profiles during walking. A healthy
subject (male, 29 years, 84 kg, 1.77 m) was instructed to walk on the treadmill
for 2 min at a comfortable speed in two conditions: first, not wearing the EXO-
TAO, and, in the sequence, wearing the exoskeleton. The average speeds were
3.3 km/h and 2.5 km/h, respectively. Then, the subject wearing the exoskeleton
was instructed to walk for 2 min at a velocity of 3.3 km/h. The first and last
twenty steps were discarded in the analysis, 40 steps were considered for each
condition.
The kinematic profiles for the lower limbs were evaluated by processing the
data from 7 IMU sensors from XSens Technologies, Netherlands. The sensors
were fixed at torso (B), thigh (T), shank (S), and foot (F). We used a Trigno
Wireless EMG system (Delsys Inc., Natick, MA, USA) to measure electromyo-
graphical (EMG) signals of five lower limb muscles: rectus femoris (RF), vastus
medialis (VM), tibialis anterior (TA), biceps femoris (BF), and gastrocnemius
lateralis (GL).
Fig. 2. Joint angles.
Figure 2 shows the angles of hip (θH ), knee (θK ) and ankle (θA ) joints in
the sagittal plane, for the three conditions, with (2.5 and 3.3 km/h) and without
wearing the exoskeleton. Note that, the mean kinematic profiles are similar, how-
ever the variability of the data with the user wearing the exoskeleton is greater.
Figure 3 shows the normalized EMG signals, which correspond to the muscular
activity for the three conditions. In the same way as to the kinematic data, the
greater variability is observed for the condition where the user is wearing the
exoskeleton.
178 W. M. dos Santos and A. A. G. Siqueira
Fig. 3. Normalized EMG signals.
Additionally, an optimal impedance controller for exoskeletons was evaluated

considering the modular exoskeleton. The proposed optimal solution is based on
the estimation of torque of the patient during the gait. A model predictive control
is then performed to obtain the optimal stiffness parameters of the exoskeletons
impedance control along the step. Figure 4 shows the mean values knee angles,
θK , estimated user torques, τuser , and torques generated by the actuator, τa ,
during the swing phase of the gait, considering the case where the user was
active and passive. Three metrics are defined to evaluate the proposed control
Fig. 4. Knee angles, user estimated torques and torques generated by the actuator.
Design and Control of a Transparent Lower Limb Exoskeleton 179
strategy: the root mean square (RMS) of the errors, E RM S , user torques, τuser
RM S
,
and robot torques, τaRM S . As can be seen in Table 1, the lower the level of user
participation, measured by the estimated torque, the greater the robot assistance
torque and tracking error allowed, that is, the robot will assist in the movement,
however, the greatest effort should be made by the user. When there is a greater
participation of the user, the robot assistance torque is lower. Therefore, the
robot has the main objective of minimizing the tracking error.
Table 1. Indexes
User E RM S [deg] τuser

RM S
[Nm] τaRM S [Nm]
Active 5.85 4.99 3.44
Passive 9.59 2.28 5.48
4 Conclusions
The experiments with healthy subjects show the proposed exoskeleton has low
influence on kinematic and muscular activity profiles during walking and can be
used to evaluate optimal impedance controllers for exoskeletons.
References
1. Contreras-Vidal, J.L., Bhagat, N.A., Brantley, J., Cruz-Garza, J.G., He, Y., Manley,
Q., Nakagome, S., Nathan, K., Tan, S.H., Zhu1, F., Pons, J.L.: Powered exoskeletons
for bipedal locomotion after spinal cord injury. J. Neural Eng. 13(3), 1–16 (2016)
2. dos Santos, W.M., Nogueira, S.L., de Oliveira, G.C., Pea, G.G., Siqueira, A.A.G.:
Design and evaluation of a modular lower limb exoskeleton for rehabilitation. In:
2017 International Conference on Rehabilitation Robotics (ICORR), London, pp.
447–451 (2017)
Development and Testing of Full-Body
Exoskeleton AXO-SUIT for Physical
Assistance of the Elderly
S. Bai1(B) , S. Christensen1 , M. Islam1 , S. Rafique2 , N. Masud2 ,

P. Mattsson2 , L. O’Sullivan3 , and V. Power3
1
Department of Materials and Production, Aalborg University,
9220 Aalborg, Denmark
shb@mp.aau.dk
2
Department of Electronics, Mathematics and Natural Sciences,
University of Gavle, Gavle, Sweden
3
University of Limerick, Limerick, Ireland
Abstract. This paper presents the design and preliminary testing of a

full-body assistive exoskeleton AXO-SUIT for older adults. AXO-SUIT
is a system of modular exoskeletons consisting of lower-body and upper-
body modules, and their combination as full body as well to provide flex-
ible physical assistance as needed. The full-body exoskeleton comprises
27 degrees of freedom, of which 17 are passive and 10 active, which is able
to assist people in walking, standing, carrying and handling tasks. In the
paper, design of the AXO-SUIT is described. End-user testing results
are presented to show the effectiveness of the exoskeleton in providing
flexible physical assistance.
1 Introduction
Exoskeleton technology has advanced into many application domains, including

medical care/rehabilitation, industrial applications and for military uses [1]. In
recent years, there has been interest in wearable exoskeletons to meet the chal-
lenges and opportunities due to the aging population. The aging demographic
changes worldwide are expected to bring a strong demand for robotic technolo-
gies like exoskeletons to assist elderly people so they can remain active and have
a high quality of life. This has resultant benefits for sustainable welfare services.
It is noticeable that most exoskeletons have been developed for either upper
limbs or lower limbs, with very few designed for the full body [2–4]. The focus of
this research was to develop an exoskeleton, namely, AXO-SUIT, to provide the
general physical assistance for older adults in the community. The AXO-SUIT,
addressing how the physical assistance/rehabilitation/compensation can be pro-
vided to supplement capabilities in a natural manner, is a full-body exoskeleton
The work reported here is supported by the EU AAL Programme, Innovation Fund
Denmark, Vinnova, Agentschap Innoveren & Ondernemen and Enterprise Ireland.
https://doi.org/10.1007/978-3-030-01887-0_35
Development and Testing of Full-Body Exoskeleton AXO-SUIT 181
comprising lower and upper body modules to also form a full body exoskeleton
to physically assist the elderly.
2 Design and Development of AXO-SUIT

AXO-SUIT is designed to give assistive physical supplementary strength at the
joint level. To make the system effective, flexible and reliable, a modular app-
roach was adopted for the AXO-SUIT exoskeleton. The AXO-SUIT exoskeleton
is shown in Fig. 1. The system consists of two main subsystems, namely, the
lower- and upper-body system, i.e. LB-AXO and UB-AXO. Each subsystem
is able to work independently to provide assistance as needed. Moreover, each
subsystem has a number of modules, which enables the joint level assistance.
The total weight of the FB-AXO is 25 kg, which includes a 3 kg Li-ion bat-
tery that can power the suit continuously for approximately 1 hour. The design
specifications for the different modules are listed in Table 1, which are devel-
oped on an extensive end-user requirement analysis conducted in four European
countries [5].
Spine
Module
UB-AXO
LB-AXO
(a) (b)
Fig. 1. AXO-SUIT exoskeleton, (1) CAD model, (b) a prototype putting on human
body, in which lengths of all links for limb segments (forearm, upper arm, thigh and
shank) are adjustable to fit the sizes of individuals. The spine module can also be
reconfigured to change its total length
182 S. Bai et al.
2.1 The Lower-Body Subsystem LB-AXO
The LB-AXO subsystem is designed to support the weight of the wearer and pro-
vide supplementary assistance to perform a range of basic motions for activities
of daily living. These motions include walking on flat ground, standing stably in
free space, sit-to-stand transfers (and vice versa) and traversing up/down stairs.
A lightweight design giving up to 50% physical assistance is developed so that it
complies with the low-risk physical assistant robot as defined in EN ISO 13482.
The mechanical design also incorporates compliant actuation for enhanced
safety. The compliant elements enable the exoskeleton to behave similar to
human joints thus making the system more comfortable and safer to use.
2.2 The Upper-Body Subsystem UB-AXO
The UB-AXO subsystem has a total of 15 degrees of freedom, three at the

spine module, five at each shoulder module and one at each elbow module. The
shoulder module is designed to match the three degrees of freedom of human
glenohumeral joint. The shoulder abduction/adduction and flexion/extension
joints are powered, while shoulder internal/external rotation joint is passively
supported by a double parallelogram linkage [6]. The elbow mechanism is a single
powered joint that supports flexion/extension of the human elbow.
Each active joint in UB-AXO and LB-AXO is powered by a brushless
dc-motor with harmonic gear. The harmonic gear was selected for its back-
drivability, which allows the user to move even if the motors are powered off.
Force sensors designed at the AAU lab are used at the wrist of the UB-AXO to
detect the arm motion, while commercial load cells are mounted on the LB-AXO
detect the leg movements.
3 System Testing
The FB-AXO has been tested with a selection of end-users and under an ethical
approval by Ethics Committee for Region Nordjylland, Denmark. All subjects
of testing were provided with written informed consent. To test the system in
a feasible, safe and ethical manner, the final physical testing of AXO-SUIT was
split into two distinct levels:
• Level 1, in which participants were healthy adults aged 18 years and over.
This involved longer testing protocols (1 h), some of which included objective
measurement of users’ muscle activation via EMG.
• Level 2, in which participants were healthy adults aged 50 years and over.
This involved shorter (0.5 h), simplified physical testing protocols which were
more feasible to implement with these users.
All of the AXO-SUIT subsystems - upper body (UB-AXO), lower body (LB-
AXO), and full body (FB-AXO) - underwent Level 1 testing, while Level 2
testing was also performed on a limited number of participants and conditions
on safety grounds.
Development and Testing of Full-Body Exoskeleton AXO-SUIT 183
Table 1. Range of Motions (RoM) and actuation of AXO-SUIT
Module Joint RoM Actuation

◦ ◦
Spine lumbar flex./ext. 30 /−30 Rubber disks
axial rot. 30◦ /−30◦ Rubber disks
lateral flex. 30◦ /−30◦ Rubber disks
Shoulder protration/retraction 122◦ /−122◦ Passive
abd./add. 120◦ /−80◦ EC-i40 and LCS-17-100
int./ext. rot. 90◦ /−50◦ Passive joint
flex./ext. 170◦ /−10◦ EC-i40 and LCS-17-100
Elbow flex./ext. 145◦ /0◦ EC-i40 and LCS-17-50
Hip flex./ext. 122◦ /−122◦ EC-60, 100 W
medial/lateral rot. 45◦ /−45◦ Passive joint
◦ ◦
abd./add. 80 /−80 Passive joint
Knee flex./ext 122◦ /0◦ EC-60, 100 W
◦ ◦
Ankle dorsi/plantar flex. 25 /−30 EC-60, 100 W
inversion/eversion 35◦ /−35◦ Passive joint
Table 2 details mean levels of maximum muscle activities for the deltoid
(n = 9) with and without the UB-AXO. The tests show that the exoskeleton
assistance reduces muscle activities in some subjects, but not in others. It is
noted that the measurements were obtained with the UB-AXO only. The user
had to carry the weight of the UB-AXO, which in return increased the activities
of body as a whole.
Table 2. EMG measurement of AXO-SUIT
Max muscle Max muscle Changes in

effort w/o exo effort w. exo efforts
Bicep Lift 9.15 ± 5.96 13.84 ± 19.92 −3.69 ± 13.96
Lower 11.36 ± 8.98 14.32 ± 15.27 −2.96 ± 6.29
Pour 2.61 ± 1.91 2.55 ± 1.59 0.06 ± 0.32
Carry 16.44 ± 9.97 17.92 ± 12.73 −1.48 ± 2.76
Deltoid Lift 8.99 ± 6.84 14.28 ± 23.83 −5.29 ± 16.99
Lower 11.71 ± 9.72 13.91 ± 18.17 −2.2 ± 8.45
Pour 2.79 ± 2.11 2.21 ± 1.27 0.58 ± 0.84
Carry 13.14 ± 6.86 13.94 ± 7.23 −0.8 ± 0.37
For the FB-AXO, tests included walking on flat ground while carrying a
6 kg load, standing stably in free space, walking up/down stairs, and carrying
184 S. Bai et al.
load of 6 kg load. The tests indicated that the FB-AXO can assist user in these
tasks. The stair walking demonstrated that the FB-AXO has a good terrain
adaptability accounting for the influences of terrains such as stairs, slopes, other
uneven terrains on stable walking [7].
4 Conclusion
In this paper, the design of a modular full-body assistive exoskeleton called

AXO-SUIT is presented. The AXO-SUIT exoskeleton enables flexible physical
assistance in such a way that it can be used as a whole body system, or as either
upper-body or lower body subsystem to assist persons in different needs. The
mechanics, electronics and sensors for the construction of the exoskeleton are
described. End-user testing was conducted and the results show effective assis-
tance with some users. The testing reveals also some limitations of the system,
which will be further improved.
References
1. Bogue, R.: Exoskeletons and robotic prosthetics: a review of recent developments.
Ind. Robot Int. J. 36(5), 421–427 (2009)
2. Proietti, T., Crocher, V., Brami, A.R.: Upper-limb robotic exoskeletons for neu-
rorehabilitation: a review on control strategies. IEEE Rev. Biomed. Eng. 9, 4–14
(2016)
3. Fontana, M., et al.: The body extender: a full-body exoskeleton for the transport
and handling of heavy loads. IEEE Robot. Autom. Mag. 21, 34–44 (2014)
4. Sankai, Y.: HAL: hybrid assistive limb based on cybernics. In: Kaneko, M., Naka-
mura, Y. (eds.) Robotics Research. Springer Tracts in Advanced Robotics, vol. 66,
pp. 25–34. Springer, Heidelberg (2010)
5. O’Sullivan, L., et al.: End user needs elicitation for a full-body exoskeleton to assist
the elderly. Procedia Manuf. 3, 1403–1409 (2015)
6. Christensen, S., Bai, S.: Kinematic analysis and design of a novel shoulder exoskele-
ton using a double parallelogram linkage. ASME J. Mech. Robot. 10(4), 041008
(2018)
7. Bai, S., Low, K.H.: Terrain evaluation and its application to path planning for
walking machines. Adv. Robot. 15(7), 729–748 (2001)
Wearable Robotics for Rehabilitation
and Assistance in Latin America
Artificial Vision Algorithm for Object
Manipulation with a Robotic
Arm in a Semi-Autonomous
Brain-Computer Interface
M. A. Ramı́rez-Moreno, S. M. Orozco-Soto, J. M. Ibarra-Zannatha,

and D. Gutiérrez(B)
Center for Research and Advanced Studies, Mexico, Mexico

dgtz@ieee.org
Abstract. We propose an artificial vision algorithm for a semi-

autonomous brain-computer interface (BCI). The interface was designed
in such a way that users are able to manipulate a robotic arm to pick
up an object from a table and place it in one of two possible locations
indicated as goal disks, and the manipulation is performed without any
concern about continuous control of the final effector. The implemented
algorithm is used to obtain, in real time, the positions of the object and
the disks in reference to the robot frame. The main techniques used in
the proposed algorithm were color segmentation and homography trans-
formation. The implementation of the algorithm allows to obtain the
positions of all the items in the table, and it successfully performs pick
and place tasks, setting the items on different positions across the table.
1 Introduction
A brain-computer interface (BCI) is a system that enables a person to manipu-

late a device for specific tasks using its brain activity. Research on BCI is mainly
focused in helping disabled people regain to some extent their lost mobility [1].
In BCI, brain signals are first acquired and filtered, then spatial or temporal
features of interest are extracted from them. Such features are detected online
in order to decipher the user’s intent. Then, the device is manipulated by the
user according to the result of the feature detection process. Some devices that
have been controlled by BCIs include spellers, robotic arms, electric wheelchairs
and prostheses [2–4].
Some BCI systems have been designed under the scheme of a continuous
control exerted by the user, by controlling low-level details of the manipulated
device, i.e. the position of the final effector of a robotic arm. A widely used
command for BCI control is motor imagery, as it generates recognizable brain
patterns [5]. As this type of scenario forces the users to maintain a continuous
high-attention level, this might lead to the generation of mental fatigue and
frustration [6].
https://doi.org/10.1007/978-3-030-01887-0_36
188 M. A. Ramı́rez-Moreno et al.
Recent research has been developed on semi-autonomous BCI systems, where

low-level details are handled by the system through different types of algorithms,
depending on the particularities of the system itself. This control scheme is
simpler and attempts to prevent mental fatigue and frustration on users, while
increasing overall performance [6]. In these type of systems, the user executes
high-level commands, and the system assists the user to complete the desired
tasks. In [7], a study on two groups of BCI users was made. One group used a
process control oriented BCI (low-level) and the other a goal selection oriented
BCI (high-level). The results showed that goal selection BCI was more accurate,
easier to learn and required less mental effort compared to the process control
BCI. These types of BCIs have been implemented to drive autonomous cars
[8], perform manipulation tasks using humanoid robots [9] and controlling a
multifunctional assistive system for disabled patients [10].
In this work, an artificial vision algorithm for a semi-autonomous BCI is pre-
sented. This interface simplifies the use of a BCI system that is specifically aimed
to assist users in pick and place tasks by controlling a robotic arm. Information
about the target and goal positions is provided by the proposed artificial vision
algorithm. The details on the implemented inverse kinematics solution for the
robotic arm can be found in [11].
An ATW-1200 Acteck web camera was used to record images at 30 fps with a
resolution of 640 × 480 pixels. Images obtained with the camera were processed
and analyzed using OpenCV-Python software. The Dynamixel AX-18A Smart
Robotic Arm, with five degrees-of-freedom (DOF), was used in these experiments
for the pick and place trials.
Fig. 1. View of the setup for the object manipulation trials as seen by the camera.
Artificial Vision Algorithm for Object Manipulation 189
The robotic arm was fixed over a white table, centered at one end of the table.
A plane in the table was delimited as a 400 × 400 mm square. Square markers
of different colors (cyan, orange, magenta and yellow) with size of 30 × 30 mm
were placed at the corners of the delimited square. A blue cylindrical object with
height of 6 mm and radius of 13 mm was used as target, and two (green and red)
disks with radius of 42 mm and were used as goal markers for the manipulation
trials. The camera was fixed in a high angle, so that all markers and items were
inside its field of view. This setup, as seen by the camera, is shown in Fig. 1.
3 Results
Specific pick and place tasks using the robotic arm were defined to evaluate
its performance. In order to perform object manipulation tasks, the real-world
coordinates of the items (positioned on the table) in reference to the robot
frame were calculated. This was achieved through the following homography
transformation:
u x
=H , (1)
v y
where H = K[R|t] is a matrix that defines the change of perspective as if the
image was seen exactly from above, K is the calibration matrix of the camera,
R and t are, respectively, the rotation matrix and translation vector applied on
the image to perform this transformation, [u v]T represents the positions of
selected points in the image, and [x y]T represents their position in real-world
coordinates.
In our case, to obtain H, both real-world and image coordinates of the cen-
troids of the square markers are needed. First, the contours of the markers were
detected through color segmentation and binarization. After a marker was iden-
tified, its centroid in the image was calculated. Since the markers have known
dimensions, the positions of their centroids in real-world coordinates are known
as well. These positions were defined as: cyan (15,15), orange (385,15), magenta
(15,385) and yellow (383,385). Then, H is obtained using OpenCV command
findHomography and it is used to transform the image as seen in Fig. 2.
Next, the origin of the image (located at the corner of the cyan marker)
was rotated (Ry,π ) and translated [185 −385 0]T . Using this transformation,
coordinates in the image were converted to coordinates that are relative to the
robot frame. There, x axis goes from −20 to 20, y axis goes from 0 to 40, and
the robot is located at 0,0. Then, color segmentation and binarization was used
to obtain the position of the centroids of the target and the goal disks relative
to the robot frame. The obtained contours of the items in the table, as well as
their centroids are shown in Fig. 2.
The cylindrical object was placed on 46 different positions across the table,
while the two goal disks were placed in random locations. The positions of the
items were distributed to span all the workspace of the robot. After the artificial
vision algorithm calculated the positions of the target and the disks, the robot
was commanded to reach for the object, grab it, and place it in either the left
or the right disk.
190 M. A. Ramı́rez-Moreno et al.
Fig. 2. Homography transformation applied on the recorded image. Contours and cen-
troids of the target and goal disks are shown in yellow.
4 Conclusion
The presented algorithm was able to identify and correctly calculate the positions
of the items in robot frame. Also, several object manipulation tasks inside the
workspace of the robotic arm were performed successfully. This algorithm will
be implemented in a BCI, where users will perform these pick and place tasks
using only their brain activity to determine the destination of the task (left
or right disk). The motivation of selecting either left or right disk is due to the
widespread use of left and right hand imaginary movements as control commands
in BCI. Also, classification of imaginary movement does not depend on stimulus
presentation, allowing an adaptation of these system to a self-paced BCI.
The integration of the presented algorithm to a BCI will contribute to the
development of a more complex system which could assist a user to perform
object manipulation tasks without the need of a continuous high-attention state.
Future work will include an evaluation of the proposed system on BCI users
and results will be compared to those obtained by continuously controlling the
position of the final effector.
References
1. Wolpaw, J.R., Birbaumer, N., McFarland, D.J., Pfurtscheller, G., Vaughan, T.M.:
Brain-computer interfaces for communication and control. Clin. Neurophysiol. 113,
767–791 (2002)
2. Cecotti, H.: A self-paced and calibration-less SSVEP-based brain-computer inter-
face speller. IEEE Trans. Neural Syst. Rehabil. Eng. 18, 127–133 (2010)
3. Wang, C., Xia, B., Li, J., Yang, W., Xiao, D., Velez, A.C., Yang, H.: Motor imagery
BCI-based robot arm system. In: Seventh International Conference on Natural
Computation (2011)
Artificial Vision Algorithm for Object Manipulation 191
4. Murphy, D.P., Bai, O., Gorgey, A.S., Fox, J., Lovegreen, W.T., Burkhardt, B.W.,
Atri, R., Marquez, J.S., Li, Q., Fei, D.-Y.: Electroencephalogram-based brain-
computer interface and lower-limb prosthesis control: a case study. Front. Neurol.
8, 696 (2017)
5. McFarland, D.J., Miner, L.A., Vaughan, T.M., Wolpaw, J.R.: Mu and beta rhythm
topographies during motor imagery and actual movements. Brain Topogr. 12, 177–
186 (2000)
6. Grainmann, B., Allison, B., Mandel, C., Lüth, T., Valbuena, D., Gräser, A.:
Non-invasive brain-computer interfaces for semi-autonomous assistive devices. In:
Robust Inteligent Systems, Chap. 6 (2008)
7. Royer, A.S., Rose, M.L., He, B.: Goal selection vs. process control while learning
to use BCI. J. Neural Eng. 8, 1–20 (2012)
8. Göhring, D., Latotzky, D., Wang, M., Rojas, R.: Semi-autonomous car control
using brain computer interfaces. Intell. Auton. Syst. 12, 393–408 (2013)
9. Bell, C.J., Shenoy, P., Chalodhorn, R., Rao, R.P.N.: Control of a humanoid robot
by a noninvasive brain-computer interface in humans. J. Neural Eng. 5, 214–220
(2008)
10. Valbuena, D., Cyriacks, M., Friman, O., Volosyak, I., Gräser, A.: Brain-computer
interface for high-level control of rehabilitation robotic systems. In: 2007 IEEE
10th International Conference on Rehabilitation Robotics, ICORR 2007, pp. 619–
625 (2007)
11. Ramı́rez Moreno, M.A., Gutiérrez, D.: Modeling a robotic arm with conformal
geometric algebra in a brain-computer interface. In: Proceedings of the 2018 Inter-
national Conference on Electronics, Communications and Computers, pp. 11–17
(2018)
Design Specifications and Usability Issues
Considered in the User Centered Design
of a Wearable Exoskeleton for Upper Limb
of Children with Spastic Cerebral Palsy
Alberto I. Perez-Sanpablo1(&), Catherine Disselhorst-Klug2,

Juan M. Ibarra Zannatha1, Josefina Gutierrez-Martínez3,
Alicia Meneses Peñaloza3, Elisa Romero-Avila2,
and Santos M. Orozco-Soto1
1
Automatic Control Department, CINVESTAV, Mexico City 07360, Mexico
{alberto.perez,jibarra}@cinvestav.mx,
sorozco@ctrl.cinvestav.mx
2
Rehabilitation and Prevention Engineering Department,
RWTH-Aachen University, Aachen 52074, Germany
{disselhorst-klug,romero}@ame.rwth-aachen.de
3
National Institute of Rehabilitation, Mexico City 14389, Mexico
{jgutierrez,ameneses}@inr.gob.mx
Abstract. This work focuses on a user centered design of a wearable

exoskeleton for upper limb of children with spastic cerebral palsy. Characteri-
zation of normal kinematics and electromyographic activity during a set of six
relevant proposed activities based on a review of literature is performed on a
sample of 9 healthy children. Further research is needed to characterized
activities on children with spastic cerebral palsy and complete the design of the
exoskeleton. Proposed methodology can be used to design more significant
technology for the user.
1 Introduction
Cerebral Palsy (CP) is a common neurological disorder [1]. Motor alterations of CP

include spasticity which is a motor disturbance related to pathological increase in the
speed-dependent stretch reflex of the spastic muscle [2] limiting movement capacity.
This work focusses on the design of a wearable exoskeleton for upper limb of CP
children with spasticity. Movement of upper limb is highly complex due to its ver-
satility, therefore in order to reduce complexity, the present work is focused on the
design of an exoskeleton to assist the motion of the elbow joint. However, design of an
Research supported by Mexican National Council of Science and Technology (CONACyT) and
the German Ministry of Education and Research under the project BMBF FONCICYT 267734.
Alberto I. Perez-Sanpablo—On leave from the National Institute of Rehabilitation, Mexico City,
Mexico.

https://doi.org/10.1007/978-3-030-01887-0_37
Design Specifications and Usability Issues Considered 193
exoskeleton for a relatively simple joint became challenging on due to the high number
of tasks where elbow joint participates and due to the complexity added by presence of
spasticity. Although some exoskeletons aimed to be used as assistive devices for
patients with neuromotor alterations exists [3], spasticity represents a limitation that
prevents their use. In order to counteract spasticity, detection is fundamental. Although
several clinical and instrumented approaches to assess spasticity have been proposed
[4], only a few try to deal with spasticity during active movement [5]. Analysis of
muscle activation by surface electromyography (sEMG) seems to be an useful strategy
to detect spasticity during active movement [6].
Nowadays there are many research groups working on the development of
exoskeletons to be used as assistive devices. However, there are barriers which limits
the adoption of those devices, such as the lack of acceptance by patients and health
professionals. This is caused due to the mismatch with user objectives. A user-centered
design approach can help to resolve this problem. Therefore, knowledge of charac-
teristics of users, as well as activities and usability are fundamental.
As a result, we propose the design of an exoskeleton for the assistance of CP
children with spasticity during active movements of the elbow joint. Exoskeleton
control and spasticity prevention will be achieved by online monitoring of sEMG and
elbow kinematics. The present work is related to the definition of user, activity and
usability characteristics and to characterization of normal muscle activity and elbow
mobility during those activities.
To know patient’s characteristics is difficult due to the variety of terms used to describe
health status and components. The International Classification of Functioning, Dis-
ability and Health (ICF) was developed with the aim of providing a unified conceptual
framework for that [7]. Clinical scales can be used to evaluate the functionality of the
upper extremities and consequently the usability of assistive devices. Items of clinical
scales can also be represented in terms of ICF codes. Consequently, definition of
characteristics of user, activity and usability was based on a previously proposed
functional assessment battery for elbow functionality for children which includes lifting
an object to waist level, placing an object with both hands above the head, throwing an
object, using a knife to cut food, grasping and drinking from a glass and hammering
[8]. This battery was proposed based on a review of the literature aimed to identify the
most relevant categories with component activities and participation for CP children
with spasticity as well as validated clinical scales to assess upper limb functionality [8].
Muscle activity and elbow mobility was characterized by analyzing interaction of
major muscles involved in elbow motion (biceps brachii, brachioradialis, triceps
medialis and triceps lateralis). Surface electromyography of healthy children was
registered by sEMG using SX230FW preamplifiers connected to a datalogger at
1000 Hz (Biometrics ltd, UK). Elbow kinematics was measured by using an SG110
electrogoniometer (Biometrics ltd, UK). Measuring protocol was approved by insti-
tutional ethics and research committee. Measures were performed after signature of
informed consent assent. Elbow angle was low-pass filtered (4th order Butterworth
194 A. I. Perez-Sanpablo et al.
6.6 Hz). Elbow velocity was calculated as the low-pass filtered version of elbow
position gradient. Electromiography signals were bandpass filtered (18th order But-
terworth, 10–450 Hz), rectified, smoothed (moving average, 80 ms window size) and
normalized to obtain muscle activation profiles. Maximum elbow velocity during
extension and flexion movement was calculated for each activity as well as frequency
content represented by mean 3 dB frequency of kinematics power spectrum. Based on
elbow kinematic data, muscular activity for each muscle were categorized into ten joint
angle position categories (0°–130°) and five joint angle velocity categories (0°/s–
750°/s) similar to previous works [9, 10]. Differences between categorized sEMG data
were analyzed by a Wilcoxon test (p < 0.05).
3 Results
Surface electromyography of nine healthy children (mean age 10 years old) was reg-
istered. Active elbow motion presented different ranges of motion, at several angular
velocities. Throwing an object presented highest maximum velocity during elbow
extension (748°/s) and elbow flexion (377°/s). Using a knife to cut food presented
lowest elbow extension maximum velocity (223°/s), while hammering showed lowest
elbow flexion maximum velocity (218°/s). Frequency of 3 dB was between
2.3 ± 0.7 Hz (lifting an object to waist level) and 4.6 ± 0.9 Hz (throwing an object).
Significant differences of muscular activation were observed for activity, muscle, angle
and velocity in both concentric and eccentric contractions (p < 0.05). Increment of
velocity of elbow motion showed a bigger effect on sEMG activity than increment of
joint angle. Four muscles presented increasing sEMG activity with increasing velocity
during concentric contractions at elbow flexion and extension respectively. Biceps and
brachioradialis presented same behavior during eccentric contraction at elbow exten-
sion. Triceps lateralis presented an increment on sEMG activity with joint position
during elbow flexion while triceps medialis showed it during elbow extension.
4 Discussion and Conclusions
A set of activities to define user, activity and usability characteristics in a well delimited
and relevant context was used for characterizing elbow motion of healthy subjects.
Although some rough data about elbow functionality can be found in literature, its
relevance and applicability for the designing of an exoskeleton as the one proposed is
questionable since most of them involve characterization during sport activities and not
during activities of daily life (ADL) [11]. As a consequence, the relevance of the
proposed activities and methodology not only for the present work but also for other
applications can be foreseen. Results agree with previous reports on adults [9] and
children using roughly analysis [10]. Differences on sEMG activity across synergistic
muscles which could be due to arm position, tasks configurations and control strategies
should be further investigated. However, results suggest existence of similar sEMG
Design Specifications and Usability Issues Considered 195
activity behavior during ADL which could be useful for control purposes. Further
research is needed to characterize activities on CP children with spasticity and to
complete the design of the proposed exoskeleton.
References
1. Bar-on, L., Krogt, van der M.M., Buizer, A.I., Desloovere, K., Harlaar, J., Sloot, L.H.:
Motorized versus manual instrumented spasticity assessment in children with cerebral palsy,
145–151 (2016)
2. Lance, J.W.: What is spasticity?. Lancet (London, England), vol. 335, no. 8689. England,
p. 606, March 1990
3. Dunaway, S., Dezsi, D.B., Perkins, J., Tran, D., Naft, J.: Case report on the use of a custom
myoelectric elbow–wrist–hand orthosis for the remediation of upper extremity paresis and
loss of function in chronic stroke. Mil. Med. 182(7), e1963–e1968 (2017)
4. Bar-On, L., Aertbeliën, E., Molenaers, G., Dan, B., Desloovere, K.: Manually controlled
instrumented spasticity assessments: a systematic review of psychometric properties. Dev.
Med. Child Neurol. 56(10), 932–950 (2014)
5. Bar-On, L., Molenaers, G., Aertbeliën, E., Monari, D., Feys, H., Desloovere, K.: The relation
between spasticity and muscle behavior during the swing phase of gait in children with
cerebral palsy. Res. Dev. Disabil. 35(12), 3354–3364 (2014)
6. Becker, S., Von Werder, S.C.F.A., Lassek, A.-K., Disselhorst-klug, C.: Time-frequency
coherence of categorized sEMG data during dynamic contractions of biceps, triceps and
brachioradialis as an approach for spasticity detection. In: International Society of
Biomechanics, pp. 1–24 (2017)
7. World Health Organization: International Classification of Functioning, Disability and
Health: Children & Youth Version: ICF-CY. WHO Press, Geneva (2007)
8. Pérez-Sanpablo, A., et al.: Proposal of a functional assessment battery for elbow
functionality for the de-sign and evaluation of assistive technology for children with spastic
cerebral palsy. In: Biomedical Engineering/Biomedizinische Technik:52nd DGBMT Annual
Conference, BMT2018 Biomedical Technology, Conference Proceedings (2018)
9. Von Werder, C., Sylvie, C.F.A., Kleiber, T., Disselhorst-Klug, C.: A method for a
categorized and probabilistic analysis of the surface electromyogram in dynamic contrac-
tions. Front. Physiol. 6(UNSP), 30 (2015)
10. Pérez-Sanpablo, A., et al.: Interaction of synergistic and antagonistic muscles of elbow joint
during activities of daily living in healthy children. In: Biomedical
Engineering/Biomedizinische Technik:52nd DGBMT Annual Conference, BMT2018
Biomedical Technology, Conference Proceedings (2018)
11. Fleisig, G.S., Laughlin, W.A., Aune, K.T., Cain, E.L., Dugas, J.R., Andrews, J.R.:
Differences among fastball, curveball, and change-up pitching biomechanics across various
levels of baseball. Sport. Biomech. 15(2), 128–138 (2016)
Stance Control with the Active Knee
Orthosis ALLOR for Post-Stroke Patients
During Walking
A. C. Villa-Parra1(B) , J. Lima2 , D. Delisle-Rodriguez3,4 , A. Frizera-Neto4 ,

and T. Bastos4
1
Biomedical Engineering Research Group (GIIB) and Career of Electronic
Engineering, Universidad Politecnica Salesiana, Cuenca, Ecuador
avillap@ieee.org
2
Postgraduate Program in Biotechnology, Universidade Federal do Espirito Santo
(UFES), Vitoria, Brazil
3
Centre of Medical Biophysics, University of Oriente, Santiago, Cuba
4
Postgraduate Program in Electrical Engineering, Federal University of Espirito
Santo (UFES), Vitoria, Brazil
Abstract. Strategies for gait rehabilitation that employ active orthosis

and exoskeletons have been proposed to improve the mobility and to
accelerate functional recovery of post-stroke patients. The challenge for
these devices is to encourage active participation of the patient and for
this purpose, impedance and damping modulation can be applied at the
device’s joints during gait. Here, we present the protocol and results
of the application of stance control in an active knee orthosis, which
works under impedance and damping adjustment based on gait phases.
Experimental results of this pilot study carried out on three post-stroke
patients showed that our active orthosis offers knee support in 50% of the
gait cycle. A positive effect of the controller on the patients, regarding
safety during the gait was also found, with a score of 4.64 in a scale
of 5, using the Quebec User Evaluation of Satisfaction with Assistive
Technology (QUEST 2.0).
1 Introduction
Some robotic exoskeletons and active orthosis have been proposed recently to
improve rehabilitation treatments for post-stroke patients, which apply func-
tional compensations at the lower limb during the gait [1]. Preliminary findings
report promising results, with sub-acute stroke patients experimenting added
benefit with these devices [1]. Clinical and biomechanical research that involve
robotic exoskeletons remark that these devices must work in constant interaction
with the patients to provide more natural movements, and enabling them to take
an active part of the training/rehabilitation, facilitating their involvement in an
This work was supported by CNPq (304192/2016-3), CAPES (88887.095626/2015-01),
FAPES (72982608), Brazil, and by SENESCYT, Ecuador.
https://doi.org/10.1007/978-3-030-01887-0_38
Stance Control with the Active Knee Orthosis ALLOR 197
attempt to improve their neural plasticity [2]. Then, control strategies that con-
sider both the ability and impairment of the patient must be taken into account
[3]. In this sense, impedance controllers are of interest for the implementation
of proper gait training and rehabilitation plans, which can be applied to these
rehabilitation devices, and also using residual motor skills of the patients. This
work presents the results of a pilot study, which applied a control strategy imple-
mented in an active knee orthosis for gait rehabilitation of post-stroke patients
based on a method for on-line knee impedance modulation, and evaluates its
safety and usability by patients.

2.1 Advanced Lower-Limb Orthosis for Rehabilitation (ALLOR)
ALLOR is a two degree of freedom orthosis developed at UFES/Brazil. This
orthosis is composed of an active knee joint and a passive hip, which moves in the
sagittal plane during walking. The hip joint has a manual flexion/extension angle
regulator (0◦ to 80◦ ) to establish a safe range of motion according to the user
requirements. The components of the active knee joint are a brushless flat motor,
a harmonic drive gearbox, an analog pulse-width modulation (PWM) servo drive
and a computer PC/104. ALLOR is equipped with strain gauges, a precision
potentiometer, Hall effect sensors and an instrumental insole. ALLOR provides
both mechanical power to the knee joint and feedback information related to
torque, knee angle, angular speeds and gait phases. ALLOR is mounted on the
left leg of the user with the axis of rotation of the orthosis’ joints aligned with
the axis of the user knee and hip. To ensure the alignment and to sustain the
load of ALLOR structure, a backpack, wraps (around the wearer’s waist) and
braces at the thigh and shank are used. The total weight of ALLOR is 3.4 kg
and it is adaptable to different anthropometric setups (heights: 1.5 to 1.85 m
and weights: 50 to 95 kg). During walking it is possible to use ALLOR without
support devices, however, clinical condition takes into account the need of using
the orthosis plus walker, canes, parallel bars or crutches. In ALLOR, the stance
control [4] and others control strategies [5] were developed in Simulink/Matlab
using a real-time target library.
2.2 Subjects
Three post-stroke patients (1 female, 2 males, 54.67 ± 3.06 years) from CREFES
of Espirito Santo state (Brazil) participated in this research. Written informed
consent was obtained from each subject before participation. Eligibility criteria
for inclusion were: (a) Hemiparetic gait of left side; (b) Cognitive skills and lan-
guage to follow the experiment instructions; (c) Volunteers without some type
of cardiorespiratory disease that interfered with the protocol; (d) Volunteers
without additional neurological or orthopedic disease that hinders ambulation;
(e) Volunteers without detectable cognitive alterations based on the Mini Men-
tal State Examination (MMSE); (f) Height range between 1.5 to 1.85 m and
maximum weight of 95 kg (ALLOR adjustment limitations).
198 A. C. Villa-Parra et al.
2.3 Protocol
The Ethics Committee of the Federal University of Espirito Santo approved this
protocol, with number: 64801316.5.0000.5542. To the experiments, ALLOR was
mounted on the patients to perform three level-ground walking trials in a dis-
tance of 10 m employing the stance control strategy based on a knee impedance
modulation reported in [4]. A knee velocity based-pattern was employed to allow
both body support in stance phase of gait and free movement of the leg in
swing phase. The trials were carried out at speed determined by the patient,
and were performed with the acquisition hardware attached to a four wheel
walker. Pictures of this pilot study with the post-stroke patients are illustrated
in Fig. 1. To measure patient’s satisfaction with the use of ALLOR, the Que-
bec User Evaluation of Satisfaction with Assistive Technology (QUEST 2.0)
was used, which allows evaluating issues related to assistive technology (dimen-
sions, weight, adjustments, safety, durability, simplicity of use, comfort and effec-
tivenes). The score for each question varied from 1 to 5 (1: “not satisfied at all”;
2: “not very satisfied”; 3: “more or less satisfied”; 4: “quite satisfied”; 5: “very
satisfied”) and the average score for valid questions answered was considered.
Walker Passive joint
PC/104
Driver
AcƟve joint
Instrumented
Insole
Fig. 1. Post-stroke patients wearing ALLOR during the pilot study. A four-wheel
walker as a balance assistive device was used as support.
3 Results
No dangerous situation and no adverse effects were reported during or after the
experiments, and all subjects completed the experiment. Figure 2 shows the per-
formance of ALLOR operation for Patient 1, which demonstrates the efficiency
of the stance control based on knee impedance modulation. This Figure shows
the plantar pressure, footswitch signal, knee trajectory, torque and impedance
modulation during the experiment with ALLOR. It can be observed that the
stance control with ALLOR can successfully support the knee joint during stance
phase of gait. For all these cases, the gait phases were recognized according to
Stance Control with the Active Knee Orthosis ALLOR 199
the plantar pressure, and knee torque did not exceed 5 Nm. In addition, ALLOR
did not demand from de patients a knee torque greater than that for healthy
subjects who used ALLOR in a before experiment. Table 1 presents the mean
and standard deviation for temporospatial parameters and maximum flexion
during swing phase for the three post-stroke patients. Regarding cadence, the
results show better performances for patients than for healthy subjects. On the
other hand, the percentage of stance phase of the gait cycle for Patient 1 and 3
was close to the healthy subjects. The opposite for Patient 2, who decreased the
stance phase. Regarding the kinematics, the maximum knee flexion was lowest
for the three patients compared to the results for healthy subjects. Regarding
the QUEST survey, the user satisfaction (mean and standard deviation) with
ALLOR controlled by the proposed strategy was scored by patients as: dimen-
sions: 3.91(0.00), weight: 3.91(0.82), adjustment: 4.22(0.94), safety: 4.64(0.47),
durability: 3.91(0.82), ease of use: 5.00(0.00), comfort: 4.31(0.47), effectiveness:
3.91(0.82), in a range of 0 to 5. In relation to user satisfaction, some hardware
adjustments in ALLOR are needed, in order to obtain a more robust orthosis
and use new materials to make the structure more compact and decrease weight.
We verified that the total time required to mount ALLOR on the user is approx-
imately 8 min. However, users no-familiarized with ALLOR spend from 20 to
25 min to wear it.
Fig. 2. Plantar pressure, gait phases, knee angle and knee torque during gait with
impedance modulation for stance control for the post-stroke patient 1.
Table 1. Results of post-stroke patients wearing ALLOR
Patient Cadence Stance phase Max Flexion

(steps/min) (% gait cycle) swing phase (◦ )
1 34.72 46.35(7.18) 13.82(3.30)
2 37.07 28.19(9.26) 8.61(5.78)
3 33.68 51.81(5.72) 15.21(1.60)
200 A. C. Villa-Parra et al.
4 Conclusion
Results demonstrate that the impedance modulation allows developing a satis-
factory gait for post-stroke patients using the stance control strategy proposed.
It is necessary more clinical trials with a larger sample to determine how ALLOR
influence their gait.
References
1. Louie, D.R., Eng, J.J.: Powered robotic exoskeletons in post-stroke rehabilitation of
gait: a scoping review. J. NeuroEng. Rehabil. 13, 53 (2016)
2. Tucker, M.R., et al.: Control strategies for active lower extremity prosthetics and
orthotics: a review. J. NeuroEng. Rehabil. 12, 1 (2015)
3. Cao, J., et al.: Control strategies for effective robot assisted gait rehabilitation: the
state of art and future prospects. Med. Eng. Phys. 36, 1555–1556 (2014)
4. Villa-Parra, A.: Knee impedance modulation to control an active orthosis using
insole sensors. Sensors 17, 2751 (2017)
5. Villa-Parra, A., et al.: Control of a robotic knee exoskeleton for assistance and
rehabilitation based on motion intention from sEMG. Res. Biomed. Eng. (2018, in
press)
Gait Phase Detection for Lower Limb
Prosthetic Devices
Pablo E. Caicedo1(B) , Carlos F. Rengifo2 , Luı́s E. Rodrı́guez3 ,

and Wilson A. Sierra3
1
Corporacion Universitaria Autonoma del Cauca, Popayán, Colombia
pablo.caicedo.r@uniautonoma.edu.co
2
Universidad del Cauca, Popayán, Colombia
caferen@unicauca.edu.co
3
Escuela Colombiana de Ingenierı́a, Bogotá, Colombia
{luis.rodriguez,wilson.sierra}@escuelaing.edu.co
Abstract. A prosthesis is an electronic-mechanical device that allows

the replacement of a lost limb functionality. These features require strict
control of the energy used for increment the operating time and patient’s
safety. In lower limb prosthetic control, it is fundamental to detect each of
the phases of the human gait cycle. For example, during the swing phase,
the prosthesis must duplicate the movement of the healthy limb, and in
load phase, this movement must be adapted. This article presents the
algorithm for spatiotemporal human gait parameter using Teager-Kaiser
energy operator and its partial validation.
1 Introduction
A prosthesis is a device that allows the replacement of a lost limb functional-
ity [1]. Lower limb prostheses supply two fundamental functions: they support
the weight of the human body, and they move the patient’s center of mass dur-
ing a gait cycle [2,3]. These features require a strict control of the energy used
in order to increment the operating time of the prosthesis and to ensure the
patient’s safety [4].
In lower limb prosthetic control, it is fundamental to detect each of the
phases of the human gait cycle [5]. For example, during the swing phase, the
prosthesis must duplicate the movement of the healthy limb, and in load phase,
this movement must be adapted [5].
To detect the phases of a walking cycle different measuring devices have been
used. The most frequently applied are inertial units, force sensor, pressure insoles
and EMG data loggers [6]. The inertial units allow the gait phase detection and
kinematic parameter measurement (joint angle, limb acceleration, etc). The force
sensors allow measuring of ground reaction forces. The pressure insoles allow the
estimation of the center mass position. Finally, EMG data loggers record the
muscle electrical activity.
The algorithm proposed in this paper uses an inertial sensor to detect the
phases of the gait cycle and to estimate both temporal and spatial gait variables.
https://doi.org/10.1007/978-3-030-01887-0_39
202 P. E. Caicedo et al.
These variables are: (i) cadence, (ii) swing time, (iii) stride length. Gait phase
detection is made by sensing heel strike and toe off events.

2.1 Materials
The algorithm design process used a dataset of ten participants without
any reported gait diseases. The age average of the participants (five male,
five females) is 22 ± 1.15 years old. The male participant height average is
175.2 ± 9.52 cm; similarly, the female participant height average is 161 ± 6.12 cm.
In the verification procedure, every attendant did nine gait experiences over an
equally-spaced floor marks, 65 cm for males and 60 cm for females; the partici-
pants carefully stepped on the marks with the same feet segment; this mean that
the stride length for men is 130 cm and 120 cm for women. There were three gait
velocities in the experiment slow, medium and fast. The medium speed is the
usual participant pace, the low and the high ones were not controlled during the
experiment. The temporal gait parameters validation was made by comparison
with the commercial motion capture system Technaid S.L.
For validation, three inertial measurement units (IMU) was placed in the
participant: both feet and lower back. Inertial units are XSens Awinda 2017
with a sampling frequency is 100 Hz.
2.2 Methods
Identification of heel strike and toe-off events was made through three steps as
reported in [7]: (i) Angular velocity smoothing, (ii) Local maxima an minima
estimation and (iii) Event identification.
After the event estimation, the temporal gait parameters are assessed as
follows: (i) Stride time is the elapsed time between two consecutive heel strike,
(ii) cadence is the relation of number of steps and the elapsed time between the
first ipsilateral heel strike and the last contralateral heel strike.
For stride length estimation, there needs five steps, the first three procedures
are: (i) Acceleration signal smoothing: This is done with a 5-order butterworth
filter with a 5 Hz cut-off frequency (ii) Energy signal of acceleration norm: The
acceleration-norm energy signal is calculated through Teager-Kaiser operator
(Eq. 1) and (iii) Determination of plantar foot phase: In the plantar phase, the
foot is in full contact with the floor. The inertial unit attached to it does not
move, and it is only affected by the gravity force; therefore, the energy must be
minimum. A temporal and amplitude threshold of the energy signal is applied.
Amplitude threshold (10% of the global maximum) identifies the minimal region
of the energy signal and the temporal threshold (250 ms1 ) allows the elimination
of positive false in the identification.
2
ψ (n) = X(n) − X(n + 1) X(n − 1) (1)
1
The temporal threshold is a time window during which only one plantar phase can
occur.
Gait Phase Detection for Lower Limb Prosthetic Devices 203
The last two process for stride length estimation are: (iv) Reprojection of
acceleration signal: Each of the inertial signals (signals between plantar phases)
is reprojected to the global coordinate system. This reprojection is carried out
in two steps: (i) Rotation of the signal to the coordinate system defined by
the orientation of the inertial sensor during the plantar foot phase, this is made
through an integration algorithm using quaternion; the mathematical foundation
of this algorithm is presented in Eqs. 2 and 3 and (ii) Rotation of the forward
signal to the global coordinate system uses the Eqs. 4 through 8 and (v) Linear
velocity and linear distance estimation: Since the foot at the beginning and the
end of the cycle is in the plantar phase; the linear velocity is considered equal
to zero at those specific moments. According to this, the linear velocity is the
global coordinate system acceleration integration minus the drift. Mathematical
foundations are showed in Eqs. 9 and 10
1
q̇(n) = Ω (ω(n)) qT (n) (2)
2
⎡ ⎤
0 ωi ωj ωk
⎢ ωi 0 ωk −ωj ⎥
Ω (ω(n)) = ⎢
⎣ ωj −ωk 0 ωi ⎦
⎥ (3)
ωk ωj −ωi 0
Where ω is the angular velocity of the gyroscope in the inertial unit, q is
T
the resulting quaternion for the algorithm. Initial condition is q(0) = [1, 0, 0, 0] .
Second reprojection
Apii (t1 )
Z= (4)
Apii (t1 )
T
Z × [1, 0, 0]
Y=

(5)
T
Z × [1, 0, 0]
X=Y ×Z (6)
T
RG = [X, Y, Z] (7)
AG
ii = RG Apii (8)
Where X, Y, Z are the axis of the global coordinate system. Apii (t1 ) is the
acceleration measured during plantar foot phase. AG ii is the acceleration in the
global coordinate system
tf
1
b= AG
ii dt (9)
(tf − t1 ) t1
Where: b is the drift contribution, t1 is the initial moment of the plantar
phase and tf is the initial moment of the next plantar phase.
The linear velocity is showed in Eq. 10.
n

G
υii (n) = Aii − b dt (10)
t1
204 P. E. Caicedo et al.
The distance estimation is done by integrating the linear velocity. The con-
tribution of the drift is only eliminated on the Z axis since only on this axis does
the condition of zero-displacement exist.
3 Results
Figure 1 shows the statistical distribution of the relative error for stride length
estimation in three cases; the best estimate (lower error), the most common case
and the worst stride length assessment (greater error) of one of the 90 gaits in
the validation data.
In the results, the outliers for the worst-case scenario are 1156% and 100%
for two different strides. But in Fig. 1 these are eliminated.
Table 1 shows the error statistics for cadence and stride time estimation.
The column named “Alg.” presents the cadence result for the algorithm and
the“Tech” column shows the result for commercial Technaid-brand device. The
last column “%Dif” is the percentage error between the columns“Alg.” and
“Tech”.
Results shows that the difference between the results, is in the expected range
for reported data [6,8]. These also suggest a correct gait segmentation for the
data.
Fig. 1. Relative error statistical distribution
Table 1. Stride time error and cadence error statistics
Right foot Left foot Cadence

Tr. Alg. Tech. % Dif. Alg. Tech. % Dif. Alg. Tech. % Dif.
1 1.1807 1.15 2.6654 1.1598 1.15 0.8515 105.3624 104.3 1.0186
2 1.1533 1.13 2.0605 1.1451 1.15 0.4240 107.8353 105.5 2.2136
3 1.1756 1.18 0.3718 1.1911 1.15 3.5735 105.7927 103.2 2.5123
4 1.1500 1.13 1.7658 1.1466 1.10 4.2377 107.9473 107.9 0.0438
5 1.1832 1.15 2.8862 1.1842 1.15 2.9749 104.7981 104.3 0.4775
6 1.2306 1.20 2.5528 1.2153 1.20 1.2781 100.9127 100.0 0.9127
Gait Phase Detection for Lower Limb Prosthetic Devices 205
Acknowledgment. The authors thank to Corporacion Universitaria Autonoma del

Cauca, Universidad del Cauca, Escuela Colombiana de Ingenieria and Innovaccion
research project for the financial support of this project.
References
1. Goršič, M., Kamnik, R., Ambrožič, L., Vitiello, N., Lefeber, D., Pasquini, G., Munih,
M.: Online phase detection using wearable sensors for walking with a robotic pros-
thesis. Sensors (Switzerland) 14(2), 2776–2794 (2014)
2. Liu, Y., Chen, Y., Shi, L., Tian, Z., Zhou, M., Li, L.: Accelerometer based joint step
detection and adaptive step length estimation algorithm using handheld devices. J.
Commun. 10(7), 520–525 (2015)
3. Chen, B., Zhong, C.-h., Zhao, X., Ma, H., Guan, X., Li, X., Liang, F.-y., Cheng,
J.C.Y., Qin, L., Law, S.-w., Liao, W.-h.: A wearable exoskeleton suit for motion
assistance to paralysed patients. J. Orthop. Transl. 11, 7–18 (2017). http://
linkinghub.elsevier.com/retrieve/pii/S2214031X16303023
4. Anam, K., Al-Jumaily, A.A.: Active exoskeleton control systems: state of the art.
Procedia Eng. 41(Iris), 988–994 (2012). https://doi.org/10.1016/j.proeng.2012.07.
273. http://linkinghub.elsevier.com/retrieve/pii/S1877705812026732
5. Long, Y., Du, Z.-j., Wang, W., Dong, W.: Development of a wearable exoskeleton
rehabilitation system based on hybrid control mode. Int. J. Adv. Robot. Syst. 13(5)
(2016). http://journals.sagepub.com/doi/10.1177/1729881416664847
6. Karuei, I., Schneider, O.S., Stern, B., Chuang, M., MacLean, K.E.: RRACE: robust
realtime algorithm for cadence estimation. Pervasive Mob. Comput. 13, 52–66
(2014). https://doi.org/10.1016/j.pmcj.2013.09.006
7. Caicedo-Rodrı́guez, P.E., Rengifo-Rodas, C.F., Rodrı́guez-Cheu, L.E.: A human gait
temporal parameters calculation algorithm. In: VII Latin American Congress on
Biomedical Engineering CLAIB 2016, Conference Proceedings, vol. 60, pp. 285–288.
Springer (2016)
8. Rampp, A., Barth, J., Schülein, S., Gaßmann, K.G., Klucken, J., Eskofier, B.M.:
Inertial sensor-based stride parameter calculation from gait sequences in geriatric
patients. IEEE Trans. Biomed. Eng. 62(4), 1089–1097 (2015)
Lower Limb Exoskeletons in Latin-America
Antonio J. del-Ama1(&), Jose M. Azorín3, José L. Pons2,

Anselmo Frizera4, Thomaz Rodrigues4, Ángel Gil-Agudo1,
Javier O. Roa5, and Juan C. Moreno2
1
Biomechanics and Assistive Technology Unit,
National Hospital for Paraplegics, Toledo, Spain
ajdela@sescam.jccm.es
2
Neural Rehabilitation Group of the Spanish National Research Council,
Madrid, Spain
3
Miguel Hernández University, Elche, Spain
4
Universidade Federal do Espírito Santo, Vitória, Espírito Santo, Brazil
5
TechnAid S.L, Arganda del Rey, Spain
Abstract. This article surveys the main lower limb exoskeletons developed, or
under development, in Latin-America, under the REASISTE Ibero American
Network. There are several groups working in this field, which approaches and
results are comparable to those reported by other groups in Europe or North
America (ALLOR, CPWalker, BioMot, Kinesis and CHIEF exoskeletons), the
overall activity in this field is comparatively limited. Moreover, we have noticed
a lack of clinical experimentation, which further prevents the advancement of
the filed in Latin-America. The specific conditions of the healthcare systems, as
well as the differences among cultures may yield valuable information towards
the rethinking of the design of the exoskeletons.
1 Introduction
Neurological diseases such as stroke and spinal cord injury (SCI) produces, among
other consequences, alterations on motor an sensory pathways, leading to a decrease on
quality of life. In the recent years, the interest on providing better rehabilitation
interventions targeting motor function has been accompanied by an interesting
development on technologies for motor rehabilitation and assistance, as robotic
exoskeletons [1]. These are person-oriented devices that are worn on the body and that
provide controlled assistance to support or compensate pathological movement.
In the Latin-American context, neurological diseases are one of the leading causes of
disability, with a prevalence of 72 million of people (aprox. 11% of Ibero-American
population). In the framework of the Ibero-American Programme on Science and
Technology for Development (CYTED), created by the governments of Ibero-American
This work is supported by H2020 R&I EXTEND project (Ref. 779982), REASISTE of CYTED
programm (Ref. 216RT0504) and Fondo de Investigaciones Sanitarias del Instituto Carlos III
(Ref. PI15/01437).

https://doi.org/10.1007/978-3-030-01887-0_40
Lower Limb Exoskeletons in Latin-America 207
countries in order to promote cooperation in science, technology and innovation for the
harmonious development of Ibero-American countries, the Thematic Net-
work REASISTE (Ibero-American Network for Rehabilitation and Assistance of
patients with Neurological Diseases through Robotic Exoskeletons) has conducted a
research on the State-of-the-Art on robotic exoskeletons in these countries, with specific
emphasis on exoskeletons and associated applications developed in Ibero-American
countries, taking into consideration specific circumstances of these countries, such as:
technology access, economy or therapeutic approaches. This article provides an over-
view of the exoskeletons surveyed in the book [ref libro].
2 Lower Limb Exoskeletons in Latin-America
2.1 Allor
The Advanced Lower-Limb Orthosis for Rehabilitation (ALLOR) has been designed in
the Intelligent Automation Laboratory of the Universidade Federal do Espírito Santo,
based on the Exo-H2 (Technaid S.A.), aimed at providing ambulatory rehabilitation of
knee deficits in post-stroke population [2]. ALLOR features active actuation at knee
joint, passive actuation at both ankle and hip joints, and force and position sensors at
the knee joint. It targets diseases that affect knee movement such as SCI, stroke,
degenerative neuromuscular disorders among others. In any case, the users of ALLOR
should have preserved hip movement ability.
Control of ALLOR can range from active to passive depending on the therapeutic
needs of the patient. Furthermore, ALLOR features gravity compensation in the main
control strategy. Up to date, ALLOR has not being tested in pathological population.
2.2 CPWalker
CPWalker is one of the few exoskeletons specifically designed for children suffering
from cerebral palsy [Ref CPWAlker]. CPWalker was designed on the basis of the
NFWalker (Made For Movement), and comprises a robotic walker linked to a 6 d.o.f.
lower limb exoskeleton. CPWalker allows adjusting individually the joint(s) to provide
assistance, as well as the type of assistance to provide via admittance controller [3]. The
human-robot interface features a multi-modal approach, comprising EEG, one IMU
placed at the patient’s trunk, EMG for the main muscles of the lower limbs, and force
sensors distributed along the exoskeleton.
CPWalker was tested in three pediatric patients. One of the achievements of the
CPWalker was the introduction of trunk posture monitoring in the therapy by moni-
toring trunk and head inclination, generating an acoustic signal when the posture was
not the recommended by the therapists. This way CPWalker not only provided walking
assistance, but also posture monitoring and correction in real time.
208 A. J. del-Ama et al.
2.3 BioMot
BioMot exoskeleton was developed within a FP7 EU project. BioMot aimed at
improving transparency in human-robot interaction via symbiotic conception of the
integration of the sensory-motor control loops of the human within the control structure
of the robot [4]. The main characteristics that comprise the BioMot exoskeleton con-
cept are:
• Series-elastic actuators with variable stiffness based on the MACCEPA concept [5].
• A computational neuromuscular model for real-time analysis of human-robot
interaction during assisted walking.
• A bioinspired controller based on the Tacit Learning control approach, aiming at
adapting to the walking ability of the user.
• Novel methods for monitoring and evaluation of cognitive attention to motor task.
• Automated methods for gait events detection based on inertial sensors and adap-
tative algorithms.
BioMot features a hierarchical control approach in which the computational
biomechanical model of HR interaction is integrated within the sensor-actuator loop of
the exoskeleton. This controller is further modulated based on estimations of the user
attention to the walking task.
The application scenarios of BioMot exoskeleton ranges from support walking of
people with incomplete SCI, in which a residual motor activity of the lower limbs is
still present, to healthy conditions.
2.4 Kinesis
Kinesis exoskeleton features a combination of a lower limb exoskeleton and a neu-
roprosthesis for combined actuation at the knee joint [6] which was defined as Hybrid
Exoskeleton. Kinesis was designed by the National Hospital for Paraplegics in col-
laboration with the Neural Rehabilitation Group (former Bioengineering Group) of
CSIC to provide walking rehabilitation of incomplete SCI patients with lesion levels
below the 12th Thoracic vertebrae.
Kinesis exoskeleton features position and force sensor for control of the knee joint,
pressure sensors for gait event detection, and hand switches for control of left/right step
initiation. The control strategy was designed around the continuous monitoring of the
HR physical interaction at the knee joint. An impedance control strategy allowed the
knee muscles to move the joint, allowing certain trajectory error, promoting the
adaptation of the user to the kinematic pattern, and the contribution of the stimulated
muscles to the movement. An estimator of muscle fatigue allowed modifying both
exoskeleton stiffness and stimulation control.
Kinesis was tested in three incomplete SCI patients. These tests demonstrated that
Kinesis was able to adapt to the functional deficits of the patient, adapting exoskeleton
stiffness to the muscle contribution at the knee joint, while estimating muscle fatigue [7].
Lower Limb Exoskeletons in Latin-America 209
2.5 Chief
This exoskeleton was developed at the Tecnológico de Monterrey (Mexico) to par-
ticipate in the first edition of the Cybathlon competition. CHIEF targets complete
paraplegic population, following the guidelines of the Cybathlon. It features three d.o.f.
for each leg, being hip and knee actives and the ankle passive, and the fastening straps
are 3D printed in carbon fiber. The sensory system of CHIEF is comprised by
accelerometers and monitors of the current fed to the motors.
3 Conclusion
This article has surveyed the exoskeletons developed, or under development, in Latin-
America. While there are several groups working in this field, which approaches and
results are comparable to those reported by other groups in Europe or North America,
the overall activity in this field is limited within the Latin-American community.
Regarding clinical application of these exoskeletons, there is a major lack of clinical
experiences, which further prevents the advancement of the filed in Latin-America. The
specific conditions of the healthcare systems, as well as the differences among cultures
may yield valuable information towards the rethinking of the design of the
exoskeletons.
References
1. Pons, J.L.: Rehabilitation exoskeletal robotics. IEEE Eng. Med. Biol. Mag. 29(3), 57–63
(2010). https://doi.org/10.1109/MEMB.2010.936548
2. Villa-Parra, A., Delisle-Rodriguez, D., Souza Lima, J., Frizera-Neto, A., Bastos, T.: Knee
impedance modulation to control an active orthosis using insole sensors. Sensors 17(12), 2751
(2017). https://doi.org/10.3390/s17122751
3. Bayon, C., Ramírez, O., Del Castillo, M.D., Serrano, J.I., Raya, R., Belda-Lois, J.M., Poveda,
R., Mollà, F., Martin, T., Martínez, I. and Lara, S.L., Rocon, E.: CPWalker: robotic platform
for gait rehabilitation in patients with Cerebral Palsy. In: Proceedings - IEEE International
Conference on Robotics and Automation, 2016–June (2016). http://doi.org/10.1109/ICRA.
2016.7487561
4. Moreno, J.C., Asin, G., Pons, J.L., Cuypers, H., Vanderborght, B., Lefeber, D., Ceseracciu,
E., Reggiani, M., Thorsteinsson, F., del-Ama, A.J. and Gil-Agudo, A.: Symbiotic wearable
robotic exoskeletons: the concept of the BioMot project. In: Lecture Notes in Computer
Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in
Bioinformatics), vol. 8820, pp. 72–83 (2014). http://doi.org/10.1007/978-3-319-13500-7_6
5. Bacek, T., Unal, R., Moltedo, M., Junius, K., Cuypers, H., Vanderborght, B., Lefeber, D.: (n.d.).
Conceptual Design of a Novel Variable Stiffness Actuator for Use in Lower Limb Exoskeletons
6. del-Ama, A.J., Gil-Agudo, Á., Pons, J.L. Moreno, J.C.: Hybrid FES-Robot Cooperative
Control of Ambulatory Gait Rehabilitation Exoskeleton. J. Neuroeng. Rehabil. 11, 27 (2014).
http://doi.org/10.1186/1743-0003-11-27
7. Del-Ama, A.J., Gil-Agudo, Á., Pons, J.L., Moreno, J.C.: Hybrid gait training with an
overground robot for people with incomplete spinal cord injury: a pilot study. Frontiers in
Human Neuroscience 8(May), 298 (2014). https://doi.org/10.3389/fnhum.2014.00298
Development of a Visual-Inertial Motion
Tracking System for Muscular-Effort/
Angular Joint-Position Relation
to Obtain a Quantifiable Variable
of Spasticity
S. M. Orozco-Soto(B) , A. I. Pérez-Sanpablo, P. Vera-Bustamante,

and J. M. Ibarra-Zannatha
Automatic Control Department, CINVESTAV Mexico City,

07350 Mexico City, Mexico
sorozco@ctrl.cinvestav.mx
Abstract. In this work, the development of an electronic instrument to

assist in spasticity measurement is presented. The device measures joint
angular position of the elbow using inertial measurement units (IMUs)
and a monocular camera. A sensor fusion algorithm captures subject’s
motion using gyroscopes, accelerometers and camera information. The
developed instrument transmits data by wi-fi from a microcontroller to
PC, where IMU data is fused with camera data extracted from visual
markers. The developed motion tracking system data was compared with
a commercial device, showing better results, which encourage its further
development for spasticity evaluation.
1 Introduction
Spasticity is a motor disturbance related to pathological increase in the speed-
dependent stretch reflex of the spastic muscle [1], which is common for several
neurological disorders, such as cerebral Palsy (CP). Nowadays there is no a defini-
tive solution for spasticity. There are many research groups looking for robotic
based strategies to counteract spasticity [2]. But to counteract spasticity, detec-
tion is fundamental. Although several clinical and instrumented approaches to
assess spasticity have been proposed [3], only a few try to deal with spasticity
during active movement [4]. Analysis of muscle activation combined with joint
kinematics seems to be a useful strategy to detect spasticity during active move-
ment [5]. Joint kinematics can be measured by inertial sensors or video based
techniques in controlled environments during clinical assessments.
2 Hardware Description
The presented instrument uses two MPU-6050 IMUs for motion capture of the
upper limb, one is placed on the arm and one in the forearm. Such IMUs are com-
https://doi.org/10.1007/978-3-030-01887-0_41
Development of a Visual-Inertial Motion Tracking System 211
pound by a 3-axis accelerometer and a 3-axis gyroscope. Furthermore, the IMUs

use I2C data transfer protocol to send prefiltered data to the microcontroller.
The hardware that processes the acquired signals from IMUs is a Lua
NodeMCU. This board is based on the ESP8266 microcontroller, and it is devel-
oped mainly for Internet-of-Things (IoT) applications, since it works with Wi-Fi
technology. Additionally, it has 10 pins that may work as digital GPIOs, PWMs
or 1-Wire interfaces and it is also equipped with a 10-bit ADC pin.
For the vision-based motion tracking, a Microsoft Lifecam VX-2000 was
implemented to acquire real-time video using OpenCV Python libraries, how-
ever, almost any USB webcam can be used instead. A complete setup of the
motion tracking system is depicted in Fig. 1, where the markers, and IMU loca-
tions can be appreciated.
Fig. 1. Complete setup of the developed motion tracking system.
3 Data Processing and Sensor Fusion
3.1 IMUs Data Acquisition and Signal Conditioning
The acquired data from IMUs is processed within the Lua NodeMCU micro-
controller. The MPU-6050 casts low-pass filtered data from accelerometers and
gyroscopes and it is read by de Lua NodeMCU, distributed in 14 8-bit registers
to form 7 16-bit words containing measurements from the 3-axis accelerometer,
a Temperature sensor, and the 3-axis gyroscope. Since the IMUs are oriented
with their z axes parallel to the flexion/extension axis of the shoulder and elbow,
only the z axis angular velocity from gyroscope is considered. Nevertheless, the
obtained data is still noisy and affected by an offset, so the following expression
is used for signal conditioning:
vi (k) = (vir (k) − 2600)/32.8 (1)

212 S. M. Orozco-Soto et al.
Where vi (k), i = 0, 1 represent the angular velocity in deg/sec of shoulder and

elbow respectively at instant k and vir (k) ∈ [0, 16600] is the raw data. The used
constants are provided in the MPU-6050 datasheet. Then, the angular position
of each joint is estimated using:
qig (k) = sat(qig (k − 1) + hvi (k)) (2)
Where q0g is the shoulder joint flexion/extension angle, q1g is the elbow joint
angle and h is the integration step. Each joint values are saturated within its
corresponding range of motion. Eventhough (2) yields clean measurements, it
depends on motion velocity to estimate a correct angular position, so the data is
compared with the angle obtained from x and y axes accelerometers as follows:

180 Ay (k)
qia (k) = sat tan−1 − (3)
π Ax (k)
Where qia (k) represents each joint position estimated with accelerometers and
Ay (k), Ax (k) are instant accelerometer measurements. qia (k) and qig (k) are sent
using WiFi to a PC in order to be fusioned with the visual data.
3.2 Visual Data Acquisition
In order to capture the upper limb motion data, a USB monocular camera was
implemented. The motion capture algorithm consists in detecting colored mark-
ers from the video capture of the camera. It worth to be mentioned that camera
should be placed so that the plane of the upper limb of the patient is parallel to
the image plane. Hence, using the centroids of the colored markers, which also
hold the IMUs, the distances between such centroids are computed, so that, the
angular position of each joint can be computed respect to the vertical axis of the
image plane, using the following expression:

180 xi+1 (k) − xi (k)
qiv (k) = tan−1 (4)
π yi+1 (k) − yi (k)
Where i is the joint and xi , yi are the instant coordinates of each centroid.
3.3 Sensor Fusion
The implemented sensors have their corresponding advantages and drawbacks,

as can be read in Table 1. If only one sensor is used, and any of its correspond-
ing drawbacks is present, the measurement would result inaccurate or missing.
Hence, a Kalman filter was implemented in order to fusion the sensors to obtain
an accurate and robust computing of each joint of the patient upper limb [6].
Firstly, the averages of the measurements are computed as follows:
(·) (·) (·) (·)
μi (k) = μi (k − 1) + K(qi (k) − μi (k − 1)) (5)
Where i = 1, 2 for each joint and (·) = g, a or v: g for gyroscope, a for accelerom-
eter and v for vision. Then, the standard deviation from each measurement is
computed using:
2(·) 2(·)
σi (k) = (1 − K)σ̄i (k) (6)
2(·) 2(·) (·) (·)
Where σ̄i (k) = σi (k−1)+K(qi (k)−μi (k−1))2 . Finally, the joint positions
are estimated with the following expression:
(k)(qia (k) + qiv (k)) + σi (k)(qig (k) + qiv (k)) + σi (k)(qia (k) + qig (k))
2(g) 2(a) 2(v)
σi
q̂i = 2(g) 2(a) 2(v)
σi (k) + σi (k) + σi (k)
(7)
4 Results
In order to test the presented motion tracking system, a comparative experi-

ment using Shimmer commercial intertial motion tracking devices was put into
effect. Such experiment was carry out by wearing both sensors and measur-
ing flexion/extension elbow motion during 4800 samples. The goal is to detect
which device is capable to track motion without missing positions or with faster
response. In Fig. 2, the obtained measurementes of elbow angular position using
a commercial device and the presented system are presented. The commercial
device presents more noise than the developed one, and it can be appreciated
that its response is slower. Furthermore it misses some high speed motions, as
can be noticed during samples 4000 to 4500. The mean square error of the com-
mercial device is about 19◦ , against a 2.4◦ from the developed motion tracking
system. A video showing the operation of both systems can be found online
at https://youtu.be/QfzNEOPgLzc. Figure 3 shows a picture of the experiment
carried out to evaluate the performance of the developed system. Note how the
joint angular position values are displayed in the upper part of the screen that
shows the algorithm running.
Table 1. Sensors advantages and drawbacks
Sensor Advantages Drawbacks

Gyroscope Robust against translational Noisy, susceptible to motion speed,
motion offset, integration errors
Accelerometer Easy to compute angles with Noisy, susceptible to translational
raw data motion
Camera Accurate and robust Susceptible to illumination
changes, upper limb should move
always in the same plane
214 S. M. Orozco-Soto et al.
Fig. 2. Comparison between elbow position measurements using a commercial device

and the presented developed system.
Fig. 3. Experiment carry out with the developed and the commercial system. Video
available at https://youtu.be/QfzNEOPgLzc
5 Conclusions
In this work, the development of a low-cost visual-inertial motion tracking sys-

tem is presented. Both, implemented hardware and data processing algorithms
are detailed. A comparative experiment using an inertial commercial device
was carry out for performance evaluation. The developed system showed bet-
ter results while tracking elbow motion than the commercial one. Although the
requirement of perpendicularity between the camera and the motion plane of the
joint could be considered a major drawback, it is not. Subjects can be placed on
particular positions in order to test joint movements which agrees with evalua-
tion methods used during current clinical practice. Furthermore, system can be
improved by including image processing algorithms and calibration maneuvers to
deal with perspective changes. Results are motivating for further development
of the system for spasticity evaluation. Future work includes integration with
surface electromyography sensors and comparison with clinical gold standards
like Modified Asworth Scale or Modified Tardieu Scale obtained in subjects with
spasticity.
References
1. Lance, J.W.: What is spasticity? Lancet (London, England), vol. 335, no. 8689.
England, p. 606, March 1990
2. Pérez-sanpablo, A.I., Ibarra-zannatha, J.M., Cifuentes-garcı́a, C.A., Rodrı́guez,
L.E.: Implementation of a shoulder and elbow musculoskeletal model in muscu-
loskeletal modelling and simulation software (MSMS). In: IEEE Colombian Confer-
ence on Robotics and Automation 2016, pp. 2–6 (2016)
3. Bar-On, L., Aertbelien, E., Molenaers, G., Dan, B., Desloovere, K.: Manually con-
trolled instrumented spasticity assessments: a systematic review of psychometric
properties. Dev. Med. Child Neurol. 56(10), 932–950 (2014)
4. Bar-On, L., Molenaers, G., Aertbelien, E., Monari, D., Feys, H., Desloovere, K.:
The relation between spasticity and muscle behavior during the swing phase of gait
in children with cerebral palsy. Res. Dev. Disabil. 35(12), 3354–3364 (2014)
5. Becker, S., Von Werder, S.C.F.A., Lassek, A.-K., Disselhorst-klug, C.: Time-
frequency coherence of categorized sEMG data during dynamic contractions of
biceps, triceps and brachioradialis as an approach for spasticity detection. In: Inter-
national Society of Biomechanics, pp. 1–24 (2017)
6. Rojas, R.: The kalman filter, pp. 1–7 (2003). http://www.robocup.mi.fu-berlin.de/
buch/kalman.pdf
Wearable Robotic Solutions
for Factories of the Future
Towards Standard Specifications
for Back-Support Exoskeletons
Stefano Toxiri1(B) , Matteo Sposito1,2 , Maria Lazzaroni1,2 , Lorenza Mancini1 ,

Massimo Di Pardo3 , Darwin G. Caldwell1 , and Jesús Ortiz1
1
Department of Advanced Robotics, Istituto Italiano di Tecnologia, Genoa, Italy
stefano.toxiri@iit.it
2
Department of Electronics, Information and Bioengineering, Politecnico di Milano,
Milan, Italy
3
Centro Ricerche Fiat, Orbassano, Italy
Abstract. Back-support exoskeletons have shown the potential to

improve workplace ergonomics by reducing the risk of low-back injury. To
support the rapidly expanding landscape and to correspondingly promote
correct adoption, standard specifications for back-support exoskeletons
are desirable. We propose a list of properties and discuss their relevance
to industrial applications.
1 Introduction
The large prevalence of occupational low-back pain and injury highlights the
need for technical solutions to improve workplace ergonomics. A growing number
of exoskeletons for this application have been developed in the past few years,
including many research prototypes and, most recently, commercially available
products. These devices are known as “back support”, “lift assist”, “lumbar
support”, “hip orthosis”. They are wearable devices that produce forces between
the user’s torso and thighs. Their effect is to reduce compressive loading on
the lumbar spine, which is believed to reduce the ergonomic risk [1]. Lists of
prototypes and products in this category can be found in [2,3].
As exoskeletons are still a relatively new class of devices, a challenge associ-
ated to the rapidly expanding landscape is to keep track of the different types
of devices and their intended function in a standard way. The authors believe
that a standard description will positively contribute by facilitating the com-
munication between stakeholders, ultimately promoting adoption in industry.
A secondary but important impact of clearer communication will be to pro-
vide valuable feedback for developers to improve their devices based on real
needs. The interest in standard descriptions of exoskeletons is supported in the
recent literature. In [4], a framework to describe and compare different models of
lower-limb exoskeletons is proposed, as organized in categories. The study in [5]
This work has been funded by the Italian Workers’ Compensation Authority
(INAIL).
https://doi.org/10.1007/978-3-030-01887-0_42
220 S. Toxiri et al.
is specific to back-support exoskeletons. It describes an experimental setup and

procedure to estimate the physical effectiveness of a device. The setup consists
of a grounded articulated robot that replicates the shape and movements of the
human body during the target lifting tasks. The outcome of the experiment on
the HAL Lumbar Support is also reported.
This contribution proposes a set of properties to define back-support
exoskeletons. The goal is to promote a common, standardized language to
describe and compare devices belonging to this class.
2 Methods
The proposed properties of interest are grouped in different categories (see
Table 1). The physical properties include overall information on the fit of the
device on the wearer. The second category provides technical information such
as actuation technology, power autonomy and functionality. The usage proper-
ties describe the operator interface and related information that affect the use in
a factory scenario. For the sake of illustration and to foster discussion, Table 1
is populated with sample information describing different devices known to the
authors.
3 Discussion
A number of properties influence the success and fit for a given application.
Clearly, its total weight has a direct effect on user comfort and acceptance.
However, its weight should be considered in relation with the physical assistance
it provides. In particular, the physical effectiveness of a device is determined
by (a) the forces that it can provide and (b) how these are modulated during
operation. Since forces are typically generated on a joint by springs or motors,
we chose to represent them in Table 1 as joint torques. The larger the torque
capability, the greater the extent to which a device can contribute to the task
and thereby reduce the risk of injury. However, this value alone does not directly
determine the physical effectiveness of a device. Indeed, the actuation principle
(i.e. the physical component that generates torque) can make a substantial dif-
ference. The assistance provided by a passive device (i.e. only using mechanical
elements such as springs or dampers) is determined at design stage and cannot
be modulated during operation. On the other hand, active devices employing
powered actuators can potentially adjust automatically. The key to exploiting
this potential is the assistive function, or strategy. A strategy should reflect what
the user needs during the different phases of the target task. Different strategies
are possible on active exoskeletons, and they may be modulated in real time
to match different needs. Additionally, a strategy should operate automatically,
requiring no or minimal user intervention so as to reduce the cognitive burden to
a minimum. With the goal of promoting effectiveness while adapting to different
users, we refer to the operator interface as the way the function of a device may
be adjusted (e.g. tuning parameters such as thresholds). On an active device,
Towards Standard Specifications for Back-Support Exoskeletons 221
this may be possible by “digitally” interacting with the computer program that
determines the strategy. On a passive device, set screws or switches may for
instance determine the pretension or offset of a mechanical spring.
Table 1. Proposed list of specifications. For the sake of illustration, information is

provided for a few existing devices.
Device Robo-Mate Trunk HAL Lumbar Support Laevo [8]

Mk2b [6] [7]
Physical properties
Weight 10 kg 3 kg (incl. battery) 2.8 kg
Attachment Shoulders, abdomen, Abdomen, thighs Chest, waist,
points thighs thighs
Lateral footprint 62 cm 45 cm Adjustable
Accomodated 165–190 cm 140–180 cm 172–188 cm
user height
Technical properties
Assistive Assistive forces Combination of torso Elastic behavior
function increase with both (i) inclination and sEMG contrasting back
torso inclination and from spinal muscles and hip flexion
(ii) weight of held [9]
object (via forearm
sEMG) [6]
Max. assistive (2x) 20 Nm Not available (2x) 20 Nm
torque
Actuation (active) geared BLDC (active) electric (passive) gas
principle motors motors springs
Operating 24 V Not available Not applicable
voltage
Power autonomy Not applicable 3h Not applicable
(estimated) (external supply)
Joint range of Unrestricted Not available Unrestricted
motion
Usage properties
Time to <5 min Not available <3 min
don/setup/doff
Operator Console-based via Buttons to adjust Mechanical
interface Wi-Fi level of assistance switch to
disengage springs
Standards – IP54 Not available
Availability Research prototype Product (Japan only) Product
Accommodating different user sizes is a beneficial feature that promotes

adoption in real-life cases. In the same direction, the lateral footprint should
be kept to a minimum in order not to introduce space constraints that may
exclude application in tight spaces (e.g. inside a car frame). Furthermore, the
necessary time to don/setup/doff a device is also central to its adoption. Any
cumbersome or lengthy procedure may compromise the ability to integrate with
a specific working schedule. While long power autonomy is certainly a positive
feature, it may not have high priority in factory settings where frequent battery
replacement and recharge is facilitated. However, this may not be the case in
different scenarios such as outdoor construction sites.
4 Conclusion
The availability of standard specifications for back-support exoskeletons is

expected to support adoption in industry by helping to critically evaluate and
compare the fit of the different available options in a given application. In general
terms, it is important to weigh the assistance that a device can provide against
the limitations it imposes on a given operation.
Standard specifications will impact the different stakeholders. Potential
adopters will have a clearer picture of the different options and the advan-
tages and drawbacks associated to them. They will therefore be encouraged
to test available devices. On the other hand, exoskeleton developers including
researchers and manufacturers will benefit from extended valuable feedback as
more field tests are carried out and solutions are adopted.
References
tons for industrial application and their potential effects on physical work load.
Ergonomics 59(5), 671–681 (2016)
2. Sugar, T., Veneman, J., Hochberg, C., Shourijeh, M.S., Acosta, A., Vazquez-Torres,
R., Marinov, B., Nabeshima, C.: Hip exoskeleton market - review of lift assist wear-
ables. Technical report (2018)
3. Exoskeleton Report Catalog. https://exoskeletonreport.com/product-category/
exoskeleton-catalog/industrial/back-support/
4. Bryce, T.N., Dijkers, M.P., Kozlowski, A.J.: Framework for assessment of the usabil-
ity of lower-extremity robotic exoskeletal orthoses. Am. J. Phys. Med. Rehabil.
94(11), 1000–1014 (2015)
5. Nabeshima, C., Ayusawa, K., Hochberg, C., Yoshida, E.: Standard performance test
of wearable robots for lumbar support. IEEE Robot. Autom. Lett. 3(3), 2182–2189
(2018)
6. Toxiri, S., Koopman, A.S., Lazzaroni, M., Ortiz, J., Power, V., de Looze, M.P.,
O’Sullivan, L., Caldwell, D.G.: Rationale, implementation and evaluation of assistive
strategies for an active back-support exoskeleton. Front. Robot. AI 5(53) (2018)
Towards Standard Specifications for Back-Support Exoskeletons 223
7. HAL for Labor Support/Care Support (Lumbar Type). https://www.cyberdyne.jp/

english/products/Lumbar LaborSupport.html
8. Laevo. http://en.laevo.nl/
9. Hara, H., Sankai, Y.: Development of HAL for lumbar support. In: SCIS and ISIS
2010 - Joint 5th International Conference on Soft Computing and Intelligent Systems
and 11th International Symposium on Advanced Intelligent Systems, pp. 416–421
(2010)
Lift Movement Detection with a QDA Classifier
for an Active Hip Exoskeleton
Baojun Chen1(&), Lorenzo Grazi1, Francesco Lanotte1,

Nicola Vitiello1,2, and Simona Crea1,2
1
BioRobotics Institute, Scuola Superiore Sant’Anna, Pisa, Italy
baojun.chen@santannapisa.it
2
Abstract. To provide assistance with an active exoskeleton, the control system

of the device has to automatically detect the onset of the user’s movement and
provide timely assistance, according to the recognized movement. In this paper,
we present an algorithm designed to detect the lift movement with an active
pelvis exoskeleton, based on a quadratic-discriminant-analysis classifier com-
bined with a rule-based algorithm. The algorithm relies on sensory information
acquired from the sensory apparatus of the exoskeleton, without needing
additional sensors to be placed on the user’s body. The algorithm was validated
in experiments with seven healthy subjects. Participants were requested to
execute different actions, i.e. lift and lower a load, stand up, sit down and walk,
while wearing the exoskeleton. On average, the algorithm showed an accuracy
of 98.7 ± 0.6% in recognizing the lift task; such performance make it suitable
for use in real application scenarios.
1 Introduction
In recent years, the interest in exoskeletons for industrial applications has continuously
and significantly increased [1]. Among all, heavy material handling tasks represent one
of the most promising applications, in which exoskeletons can be used to provide
assistance to help low-back muscles, reducing the effort of the operator and improving
the workplace ergonomics conditions. In such application, compared to passive
exoskeletons, active exoskeletons have more versatile control systems and can deliver
more powerful and efficient assistance. However, to take full advantage of an active
exoskeleton, it is fundamental that the control system automatically detects the onset of
the lift movement in order to deliver proper assistance, i.e. synchronously with the
assisted movement.
For this specific application, only few studies proposed automatic algorithms to
detect the lift task [2, 3]. In most of them, the exoskeleton control systems are based on
electromyographic (EMG) signals to control the action of the robot. Despite the good
results, the main limitation of this approach is that it relies on extra sensors (i.e. EMG
This work was supported by Regione Toscana within the CENTAURO project (Bando FAR-FAS
2014) and EU within the HUMAN project (H2020-FOF-2016, GA 723737).

https://doi.org/10.1007/978-3-030-01887-0_43
Lift Movement Detection with a QDA Classifier 225
sensors) which should be placed on user’s body; such architecture increases the system
complexity and likely limits its usability in industrial scenarios.
In this paper, we present a lift detection algorithm based on signals collected by
sensors embedded in the exoskeleton. Compared to algorithms requiring extra sensors,
this approach is more suitable for industrial scenarios. The algorithm is based on a
quadratic-discriminant-analysis (QDA) classifier combined with a set of pre-defined
threshold-based rules. The algorithm was verified on seven subjects wearing an active
hip exoskeleton. For each subject, a training phase allowed to acquire movement data
and train the algorithm, while in the testing phase the output of the real-time algorithm
was used to control the assistive action of the device.
2 Methods
2.1 Experimental Setup and Protocol

The exoskeleton used in this study is a robotic hip exoskeleton designed to assist the
hip flexion/extension movement (APO [4, 5]). The system sensory apparatus includes
two joint encoders to measure left and right hip angles and one inertial measurement
unit (IMU) placed on the backpack to measure the trunk kinematics (Fig. 1(A)).
Fig. 1. (A) A subject wearing the APO. (B) Phase definition for lift detection. The dotted lines
represent in the order the start of the Grasp phase, the start of the Lift phase and the end of the Lift
phase.
Seven male subjects (age: 27.9 ± 2.3 years, height: 178.1 ± 8.1 cm, weight:
70.0 ± 6.4 kg) were recruited for this experiment. Subjects participated in two
experimental sessions: a training session and a testing session. In the training session,
subjects were asked to perform two types of tasks: (1) repetitive lifting and (2) a
movement sequence including standing up, walking, lifting and lowering a load, and
sitting down. Both tasks were performed with five different lift techniques (i.e. squat,
stoop, freestyle, left-asymmetric, and right-asymmetric lifting) and lift was performed
using a 5-kg box. In the testing session, subjects were asked to complete the same
226 B. Chen et al.
movement sequence of the training session, performing the lift action only with the
freestyle technique, at slow, normal and fast speeds. The APO was controlled in
transparent mode (i.e. zero-torque control) in the training session. Data collected in the
training session were used to train the detection model, which was used in the testing
session, in which the device was controlled to provide timely assistance, i.e. syn-
chronously with the detected movement onset.
2.2 Lift Detection Algorithm

Lift detection is performed in two steps. In the first step, a set of threshold-based rules
is used to detect possible lift movements. The movement can be segmented into three
phases: Grasp, Lift, and Other (Fig. 1(B)). The transition from Other to Grasp should
follow three rules: (1) hdiff \a1 , (2) hmean [ a2 and (3) hstd \a3 . The transition from
Grasp to Lift, i.e. the detection of possible lift, should satisfy three rules: (1) hstd [ a4 ,
(2) HasPeak ¼ 1, and (3) Tgrasp T0 . The transition from Lift to Other should meet two
rules: (1) hmean \a5 , and (2) hstd \a6 or HasValley ¼ 1. In the above rules: hdiff is
calculated as jhL hR j, where hL and hR denote hip joint angle of the left and right side,
respectively; hmean equals to ðhL þ hR Þ=2 and hstd is the standard deviation of ðhL þ hR Þ
computed over the last 100 ms; Tgrasp is the current duration of grasping; HasPeak
equals to 1 if a peak of ðhL þ hR Þ has occurred within the Grasp phase, otherwise it
equals to 0; HasValley equals to 1 if a valley of ðhL þ hR Þ is detected within the Lift
phase, otherwise it equals to 0; a1 to a6 , and T0 are predefined thresholds, empirically
determined and kept constant for all the participant subjects.
In the second step, when a possible lift movement is detected, a QDA classifier is
used to determine whether the movement corresponds to lift or it was a misclassifi-
cation. Two features are used for the recognition: f1 ¼ hmean and f2 ¼ bðiÞ b igrasp0 :
b is defined as hmean u, where u is roll value of the IMU on the backpack of the
exoskeleton. i denotes the current sample and igrasp0 denotes the sample corresponding
to the time instant when a grasp is detected.
3 Results
Table 1 reports the performance of the online lift detection algorithm, achieved in the
testing session, for the seven subjects. Note that true positives (TP) are the lifts which
were correctly detected as lifts; false positives (FP) denote non-lift movements which
were detected as lifts; true negatives (TN) are non-lift movements which were detected
as non-lifts; false negatives (FN) denote lifts which were not detected as lifts.
The average detection accuracy across seven subjects was 98.72 ± 0.58%
(mean ± SEM).
Lift Movement Detection with a QDA Classifier 227
Table 1. Lift detection performance in the testing session

Subject TP [#] FP [#] TN [#] FN [#] Accuracy
S1 174 0 29 0 100.0%
S2 177 0 30 3 98.6%
S3 183 8 22 1 95.8%
S4 180 1 29 0 99.5%
S5 182 1 29 0 99.5%
S6 180 5 25 0 97.6%
S7 180 0 30 0 100.0%
To provide assistance to a user, the accurate and timely detection of movement onset
plays a fundamental role in the design of exoskeletons control systems. In this paper,
we presented a lift detection algorithm for an active hip exoskeleton for assistance in
lift tasks. The algorithm used a two-step detection strategy. The first step was designed
to detect the onset of a possible lift movement and avoid false detections of non-lift
movements (e.g. stand up and level-ground walk) as lifts. However, some movements
(e.g. sit down) are similar to the lift task and brought to misclassifications.
The proposed algorithm has two main advantages. First, the algorithm only uses
signals measured from exoskeleton embedded sensors; this makes the whole system
compact and suitable to use in industrial application scenarios. Second, the algorithm is
able to detect the lift onset in a timely manner to provide efficient assistance. Subjective
feedback from all subjects showed that the APO controlled with the proposed algorithm
could provide helpful assistance, reducing the perceived physical effort of the users. In
addition, the application of a QDA classifier in the second step increases the potential
of the algorithm with respect to a simple rule-based strategy [4], because it could be
potentially used to solve multi-class recognition problems, e.g. to classify the lift
technique.
In future works, we will investigate other features to improve the performance of
lift detection, and explore the potential of this algorithm to recognize different lift
techniques and provide ad-hoc assistive profiles based on different techniques.
References
1. de Looze, M.P., Bosch, T., Krause, F., Stadler, K.S., O’Sullivan, L.W.: Exoskeletons for
industrial application and their potential effects on physical work load. Ergonomics 59, 671–
681 (2016)
2. Kawai, S., Yokoi, H., Naruse, K., Kakazu, Y.: Study for control of a power assist device.
Development of an EMG based controller considering a human model. In: 2004 IEEE/RSJ
International Conference on Intelligent Robots and Systems, pp. 2283–2288 (2004)
228 B. Chen et al.
3. Naruse, K., Kawai, S., Yokoi, H., Kakazu, Y.: Development of wearable exoskeleton power
assist system for lower back support. In: 2003 IEEE/RSJ International Conference on
Intelligent Robots and Systems, pp. 3630–3635 (2003)
4. Chen, B., Grazi, L., Lanotte, F., Vitiello, N., Crea, S.: A real-time lift detection strategy for a
hip exoskeleton. Front. Neurorobotics 12, 17 (2018)
5. Lanotte, F., Grazi, L., Chen, B., Vitiello, N., Crea, S.: A low-back exoskeleton can reduce the
erector spinae muscles activity during freestyle symmetrical load lifting tasks. In: 7th IEEE
RAS/EMBS International Conference on Biomedical Robotics and Biomechatronics,
Enschede, Netherlands (2018, in press)
The Effect of a Passive Trunk Exoskeleton
on Functional Performance
and Metabolic Costs
S. J. Baltrusch1(&), J. H. van Dieën2, S. M. Bruijn2, A. S. Koopman2,

C. A. M. van Bennekom1, and H. Houdijk2
1
Research and Development Department,
Rehabilitation Center Heliomare, Wijk aan Zee, The Netherlands
s.baltrusch@heliomare.nl
2
Faculty of Behavioural and Movement Sciences,
Vrije Universiteit Amsterdam, Amsterdam, The Netherlands
Abstract. The objective of this study was to assess the effect of a passive trunk
exoskeleton on functional performance and metabolic costs in healthy
individuals.
Functional performance of 12 work-related tasks was assessed based on
objective outcome measures and perceived task difficulty. In addition, we
measured energy expenditure during 5 min of repetitive lifting and walking,
with and without exoskeleton.
Wearing the exoskeleton tended to increase objective performance in static
forward bending. Performance in tasks that involved hip flexion decreased and
these were perceived as more difficult with the exoskeleton. Wearing the
exoskeleton during lifting decreased metabolic costs by as much as 17%, and
may reduce the development of fatigue and LBP risk. During walking, metabolic
costs increased by 17%. These results indicate the potential efficacy of the
exoskeleton to support trunk bending tasks, but also stress the need to allow
disengagement of support depending on activities performed.
1 Introduction
Mechanical work-related risk factors for low-back pain are difficult to efface from the
work environment [1]. Several studies have shown that body worn assistive devices
that passively support the user’s trunk, i.e. exoskeletons, can be used to decrease low-
back load at work [2–5].
Next to the mechanical risk factors for low back pain (LBP), physiological strain
needs to be considered when aiming to decrease workload. High physiological strain
can result in systemic or local fatigue which might increase LBP risk [6, 7]. Using an
The work presented in this paper was supported by the European Union’s Horizon 2020 research
and innovation program under grant agreement No 687662 – SPEXOR.
The authors would like to acknowledge the support of Laevo for unconditionally providing the
exoskeleton for this research.

https://doi.org/10.1007/978-3-030-01887-0_44
230 S. J. Baltrusch et al.
exoskeleton may reduce moments around the low back and hence muscle activity.
Therefore, it could be expected that it also reduces metabolic load and as such reduce
the risk of fatigue-related injuries.
Intervention studies have mostly been aimed at analyzing the effect on low back
load and metabolic costs in stereotypical lifting [8, 9]. However, many work envi-
ronments are characterized by a variety of tasks and trunk movement patterns, and
hence require a device that can be used across a range of different tasks, such as
walking, stair climbing or forward bending. An exoskeleton may not only support, but
also hamper performance of some of these tasks by increasing energy costs or affecting
task execution.
2 Objective
We aimed to assess the effect of a passive trunk exoskeleton on functional performance

and metabolic costs for a set of different work-related tasks including lifting and
walking. By using a currently commercially available device we aimed to point out
potential design problems and to create a benchmark for further developments.
3 Methods
3.1 Passive Trunk Exoskeleton

In this study, the passive trunk exoskeleton “Laevo” (Intespring, Delft, The Nether-
lands) was tested. It consists of four components: a pad at the anterior side of the chest,
leg pads at the anterior side of the thighs, a pelvis belt to keep the device in a fixed
position relative to the pelvis and a smart joint with spring-like characteristics. The
chest and thigh components are connected through semi-rigid bars running over this
joint, which generates a supporting extension moment at the level of the lower back
when bending forward.
3.2 Functional Performance

18 healthy men participated in the functional performance testing. Participants per-
formed a test battery of 12 functional tasks with and without the exoskeleton. The
selection of tasks was based on tasks from the functional capacity evaluation (FCE,
Isernhagen Work Systems) [10] and tasks derived from workplace observations.
Three different type of tasks were considered: (1) tasks in which the user potentially
benefits from the exoskeleton, (2) functional tasks in which the user is potentially
hindered by resistance against movement generated by the device and (3) basic tasks
requiring participants to use a large range of motion (ROM). Functional performance
was assessed both in objective outcomes (e.g. time to perform a task) and subjectively,
in terms of perceived task difficulty.
The Effect of a Passive Trunk Exoskeleton 231
3.3 Metabolic Costs

We measured energy expenditure in 11 healthy men during 5 min of repetitive lifting
and 5 min of walking, with and without an exoskeleton. Participants had to lift and
lower a 10 kg box from two heights at an auditory imposed pace (6 lifts/min) in three
different conditions: (1) Control (without the exoskeleton), (2) Low-cam Exoskeleton
(supports at bending angles >20°) and (3) High-cam Exoskeleton (supports at bending
angles 0–20°). In the walking protocol, participants walked with the Low-cam
Exoskeleton and without the exoskeleton at two different walking speeds: (a) preferred
walking speed determined without the exoskeleton and (b) preferred walking speed
determined with the exoskeleton.
4 Results
4.1 Functional Performance

Wearing the exoskeleton tended to increase objective performance in static forward
bending, but decreased performance in tasks, such as walking, carrying and ladder
climbing.
Fig. 1. Boxplots of perceived task difficulty. (The red line represents the sample median. The
distances between the tops and bottoms are the interquartile ranges. Whiskers show the min and
max values; outliers are represented as a +). The dotted lines represent the division between the
groups of tasks, in which the user is potentially assisted (left side), tasks, in which the user is
potentially hindered by resistance against movement generated by the device (middle) and tasks
requiring participants to use a large range of motion (right side). Brackets indicate significant
differences between the exoskeleton (with) and control condition (without). 0 = very easy,
10 = very difficult.
232 S. J. Baltrusch et al.
Subjectively, we found a significant decrease in perceived task difficulty and local

discomfort in the back during static forward bending, but a significant increase of
perceived difficulty in several other tasks, like walking, squatting and wide standing.
Especially non-load handling tasks that involved substantial trunk and hip flexion
without trunk inclination were perceived as more difficult with the exoskeleton (Fig. 1).
4.2 Metabolic Costs

Metabolic costs decreased by 17% and 16% when lifting with the low-cam exoskeleton
from knee and ankle height, respectively.
For walking, metabolic costs increased by 12% and 17% when wearing the
exoskeleton in the two different speed conditions (Fig. 2). Participants preferred to
walk faster without the exoskeleton.
Fig. 2. Left: Metabolic costs of lifting from knee and ankle height. Values are normalized for
bodyweight. Right: Metabolic costs of walking in preferred walking speed without exoskeleton
(PWS) and preferred walking speed with exoskeleton (PWSX). Values are normalized for
bodyweight and walking speed. Error bars indicate standard deviations. Brackets indicate
significant change in metabolic costs between control condition (without) and exoskeleton
condition (with exo/high cam).
5 Conclusion
Wearing an exoskeleton seems to effectively unload the back in static holding tasks and
decreases metabolic costs during lifting. It may hence reduce the development of
fatigue and LBP risk for these specific tasks.
However, it limits functional performance in several non-load-handling tasks.
Especially tasks that require hip flexion get hampered by the exoskeleton, such as
walking, which also showed increased metabolic costs when wearing an exoskeleton.
This stresses the need for a support system that can be disengaged depending on
activities performed. Design improvements should include provisions to allow full
range of motion of hips and trunk to increase versatility and user acceptance.
The Effect of a Passive Trunk Exoskeleton 233
References
1. Griffith, L.E., Shannon, H.S., Wells, R.P., Walter, S.D., Cole, D.C., Côté, P., et al.:
Individual participant data meta-analysis of mechanical workplace risk factors and low back
pain. Am. J. Public Health 102, 309–318 (2012)
2. Bosch, T., van Eck, J., Knitel, K., de Looze, M.: The effects of a passive exoskeleton on
muscle activity, discomfort and endurance time in forward bending work. Appl. Ergon. 54,
212–217 (2016)
3. Graham, R.B., Agnew, M.J., Stevenson, J.M.: Effectiveness of an on-body lifting aid at
reducing low back physical demands during an automotive assembly task: assessment of
EMG response and user acceptability. Appl. Ergon. 40, 936–942 (2009)
4. Ulrey, B.L., Fathallah, F.A.: Effect of a personal weight transfer device on muscle activities
and joint flexions in the stooped posture. J. Electromyogr. Kinesiol. 23, 195–205 (2013)
5. Wehner, M., Rempel, D., Kazerooni, H.: Lower extremity exoskeleton reduces back forces
in lifting. In: ASME 2009 Dynamic Systems and Control Conference, vol. 2, pp. 49–56
(2009)
6. Waters, T.R., Putz-Anderson, V., Garg, A., Fine, L.J.: Revised NIOSH equation for the
design and evaluation of manual lifting tasks. Ergonomics 36, 749–776 (1993)
7. Janssens, L., Brumagne, S., Polspoel, K., Troosters, T., McConnell, A.: The effect of
inspiratory muscles fatigue on postural control in people with and without recurrent low back
pain. Spine 35, 1088–1094 (2010)
8. Whitfield, B.H., Costigan, P.A., Stevenson, J.M., Smallman, C.L.: Effect of an on-body
ergonomic aid on oxygen consumption during a repetitive lifting task. Int. J. Ind. Ergon. 44,
39–44 (2014)
9. Abdoli-E, M., Stevenson, J.M.: The effect of on-body lift assistive device on the lumbar 3D
dynamic moments and EMG during asymmetric freestyle lifting. Clin. Biomech. 23, 372–
380 (2008)
10. Reneman, M.F., Brouwer, S., Meinema, A., Dijkstra, P.U., Geertzen, J.H., Groothoff, J.W.:
Test-retest reliability of the Isernhagen work systems functional capacity evaluation in
healthy adults. J. Ouccup. Rehabil. 14, 295–305 (2004)
Industrial Wearable Exoskeletons
and Exosuits Assessment Process
Jawad Masood1(B) , Angel Dacal-Nieto1 , Vı́ctor Alonso-Ramos1 ,

M. Isabel Fontano2 , Anthony Voilqué3 , and Julia Bou3
1
Processes and FoF Department, Processes and Materials Division, CTAG - Centro
Tecnológico de Automoción de Galicia, Pol. Ind. A Granxa, 36400 Porriño, Spain
jawad.masood@ctag.com
2
Centro de Vigo de Groupe PSA,
Avda. Citroën 3, 36210 Vigo, Spain
3
Groupe PSA, Centre Technique Vélizy,
Route de Gisy, 78140 Vélizy-Villacoublay, France
Abstract. Industrial wearable exoskeletons and exosuits represent a

vibrant technology with revolutionary potentials to enhance the oper-
ating conditions, health and safety of the worker. It brings forward the
important social and technological goal of helping the workers instead
of replacing them. An effective assessment process is a core for the sus-
tainability and deployment of these devices in the industry. We present a
process based on the evaluation criteria to validate the Impact on Worker,
Appropriation to the Task, Utility to the Task, Usability and Safety. We
test this criterion with the help of objective and subjective methods,
which depend upon assessment techniques, assessment devices, surveys
and subjective scales. In the end, we share our experience of implement-
ing this process, and we point out industrial needs which can help future
research and development directions.
1 Introduction
The Industrial Exoskeleton Technology (IET) is part of the Factories of the

Future and Industry 4.0 concepts [1]. Every year, manufacturing OEM’s spend
millions of euros on the musculoskeletal and tissue-based sick leaves of limited EU
workers. For example, in 2014–2015, the production lost to income in German
manufacturing industry due to such leaves was 600 euros and 950 euros per
worker. This figure is likely to increase in 2020 due to the delay in retirement age,
and grow of 50+ workers [2]. IET has the potential to cut this cost and improving
the working conditions. There are about 22 commercial industrial exoskeletons
OEM’s [3]. However, the usage of such devices is minimal due to the gap between
end-user requirements and their specification, lack of standardization and bench-
marking for industrial exoskeletons [4–6].
The authors thank the contribution of the F4.0 Automoción consortium.

https://doi.org/10.1007/978-3-030-01887-0_45
Industrial Wearable Exoskeletons and Exosuits Assessment Process 235
We believe that a generic testing process as shown in Fig. 1 can narrow this
gap. This process can provide flexibility of implementation of different exoskele-
tons and exosuits to industrial use-cases. On the one hand, it can give the end-
users confidence to invest in such innovative technologies. On the other hand,
it can offer useful feedback to exoskeleton OEM’s about end-user requirements.
This work is organized as follows: we start with the presentation of current state
of art IET. It is followed by the description of the testing process and evaluation
criteria. Finally, we conclude by sharing our experience in deploying this testing
process.
Fig. 1. The testing process designed for each use-case.
2 Wearable Technology Industrial - Assessment

2.1 State of the Art Technologies
We can divide IET into four main categories based on the support that these
devices can offer to the worker body: Hand, Shoulder, Trunk and Leg. Moreover,
each category can be further classified into four types, based on their actua-
tion principles: Passive, Active, Quasi-Passive and Quasi-Active. Today most of
these devices are passive as shown in Table 1. On perception end, these devices
are equipped with force sensors, inertial measurement units, strain gauges, dry
surface EMGs, or combination of two or more. On standardization end, their
evaluation is carried out with the help of protocols developed using bio-medicine
and bio-mechanics. In practice, we cannot find the standard process for testing
these devices. However, there are ongoing efforts in the areas of normalization,
terminology development and benchmarking1,2,3 of industrial exoskeletons and
exosuits.
1
https://www.iso.org/committee/5915511.html.
2
https://standards.globalspec.com/std/10203247/afnor-ac-z68-800.
3
https://www.astm.org/COMMITTEE/F48.htm.
236 J. Masood et al.
Table 1. Industrial wearable robotics technology
Product (Company) [3] Assistance Challenges

Hand support
Power Glove (BioServo) Active (electric) Cost and weight
Shoulder support
Airframe (Levitate) Passive (spring) Weight, cost, flexibility, user
acceptance, safety, and durability
shoulderX (suitX) Passive (spring)
eksoVest (ekso bionics) Passive (spring)
exhauss (exhauss) Passive (spring)
Fortis (Lockhead Martin) Passive (spring)
Trunk support
Laevo V2 (Laevo) No Weight, cost, flexibility, user
acceptance, safety, and durability
backX (suitX) Passive (Spring)
ErgoSkeleton (StrongArm) No
CrayX (German Bionics) Active (electric)
HAL (Cyberdyne) Active (electric)
Leg support
Chairless Chair (Noonee) Active (electric) Weight, cost, and safety
legX (suitX) Active (electric)
2.2 Testing Process

The proposed testing process can be divided into four phases: solution identifica-
tion, laboratory validation, pilot testing and in-line testing, which can be further
divided into four sub-groups: planning, preparation, data collection, and analy-
sis as shown in Fig. 1. We have devised the testing protocol based on the available
bio-mechanical standards and practices. We start the testing process by targeting
the problematic use-cases at assembly line. The tangible objectives are set based
on the historical health data of the use-case. An IET market survey is performed
to find the most relevant devices as shown in Sect. 2.1. All use-cases are differ-
ent from each other in terms of sub-tasks, complexity, range of movement and
flexibility. It is necessary to test the exoskeleton device for all use-cases to verify
and validate its benefits. During the controlled environment lab testing, we select
the objective and subjective methods, number of subjects, testing experts, place-
ment of the testing devices, device interference with the environment, and other
safety aspects. After the lab testing, an analysis report is prepared to validate if
the device is good enough for the next phase of testing. If so, the simulated testing
is performed on the real use-case but under controlled environmental conditions.
A skilled subject performs the use-case with all the measurement devices, and
data is collected by designing different testing scenarios. For example, one testing
Industrial Wearable Exoskeletons and Exosuits Assessment Process 237
scenario is to perform the same use-case with and without exoskeleton. The test-
ing scenarios are designed such that all the system features and specifications can
be tested. Finally, we can proceed to the in-line testing, once the analysis results
of the simulated task can meet the acceptance criteria.
2.3 Evaluation Criteria

The core of the testing process is the evaluation criteria, which can validate
the customer requirements with the device specifications. We have proposed an
evaluation criteria as shown in Fig. 2. The evaluation can be divided into two
main groups: Objective methods and Subjective methods. Objective methods
consist of techniques which are related to state or quality of being truth and it
is not pron to worker interpretation, feelings, bias and imagination. Subjective
methods consist of techniques related to the state or quality of the worker who
possess conscious experience (perspectives, feelings, beliefs and desires) of using
the device. The selection of the method depends on several factors such as num-
ber of subjects, testing objectives, time, cost, measurement devices, use-case and
subject morphology.
Fig. 2. The Evaluation Criteria with Objective and subjective Methods. The horizontal
row shows the completeness of the individual criteria and the vertical row shows the
completeness of the method.
3 Conclusion and Discussion
We have deployed the process at PSA Groupe for seven commercially avail-
able IET for the back, shoulder and hand support. We have tested three indus-
trial use-cases i.e. lowering-lifting and bending, overhead and low load high fre-
quency task. The process highlights important aspects such as the lack of tan-
gible testing objectives, few subjects, lack of portable evaluation methods, and
absence of acceptance criteria to switch from one phase to another. In addition,
it also point outs the need for collaboration between end-users and exoskeleton
OEM’s. Following are the main points: IMPACT: The testing process has shown
improvement in terms of muscle activity, range of motion and task efficiency by
238 J. Masood et al.
using IET. We think its impact can further improve by designing simple, cost-
effective and accessible solutions. APPROPRIATION: We observe that the IET
is up to the mark in facilitating training and familiarity of use. However, train-
ing tutorials, product documentation and labeling can be improved. UTILITY:
We observe that the IET is developed for one specific task. However, industrial
use-cases consist of various sub-tasks. We think that the rigorous testing and
feedback can help in improving the performance. USABILITY: We experience
that there is a gap between industrial user requirements and today products.
This gap can be reduced by constructive feedback from testers to IET OEM’s.
SAFETY: We emphasize to focus on the safety aspects of the worker. We pro-
pose a safety and risk analysis (Hazard Analysis and Risk Assessment) and
human-centric design approach to identify the IET risks.
References
1. Dahmen, C., Wöllecke, F., Constantinescu, C.: Challenges and possible solutions for
enhancing the workplaces of the future by integrating smart and adaptive exoskele-
tons. Procedia CIRP 67, 268–273 (2018)
2. Nöllenheidt, C., Brenscheidt, S.: Arbeitswelt im wandel: Zahlen - daten - fakten
(2016). Technical report, Federal Institute for Occupational Safety and Health
(2016)
3. Marinov, B., Dao, T.: Exoskeleton report. http://exoskeletonreport.com/. Accessed
09 May 2018
4. van der Vorm, J., Nugent, R., O’Sullivan, L.: Safety and risk management in design-
ing for the lifecycle of an exoskeleton: a novel process developed in the robo-mate
project. Procedia Manuf. 3, 1410–1417 (2015)
5. Van der Vorm, J., OSullivan, L., Nugent, R., de Looze, M.: Considerations for
developing safety standards for industrial exoskeletons (2015)
6. Masood, J., Mateos, L.A., Ortiz, J., Toxiri, S., O’Sullivan, L., Caldwell, D.: Active
safety functions for industrial lower body exoskeletons: concept and assessment. In:
Wearable Robotics: Challenges and Trends, pp. 299–303. Springer (2017)
Trunk Range of Motion in the Sagittal
Plane with and Without a Flexible Back
Support Exoskeleton
Matthias B. Näf1(B) , Axel S. Koopman2 , Carlos Rodriguez-Guerrero1 ,

Bram Vanderborght1 , and Dirk Lefeber1
1
Vrije Universiteit Brussel (VUB) and Flanders Make, Brussel, Belgium
matthias.naf@vub.be
2
Vrije Universiteit Amsterdam (VUA), Amsterdam, The Netherlands
Abstract. A large portion of the working population is affected by back

and shoulder pain. Lower back support exoskeletons were introduced as
a preventative measure, but they are not widely adopted by the industry
yet. Their adoption is hindered chiefly by discomfort, loss of range of
motion and kinematic incompatibility. In this work, we discuss the range
of motion of the trunk in the sagittal plane, once wearing a flexible
exoskeleton and once without wearing an exoskeleton (N = 2).
1 Introduction
More than 40% of the working population in the EU are affected by lower back
pain and shoulder pain [1]. One of the key risk factors for lower back pain are
compression forces on the spine [2].
Lower back support exoskeletons have been suggested as a preventative mea-
sure. Most of these devices [3] aim at decreasing these compression forces (See
Fig. 1) on the lumbar spine.
Despite significant advancements [3] in these exoskeletons, no design has
emerged, and is widely adopted by the industry yet. This is often attributed
to discomfort, loss of Range Of Motion (ROM) and kinematic incompatibility
among others [3].
Very few exoskeletons [4,5,9] account for the ROM of both the hip and the
lumbar spine (See Fig. 2). While few designs are published, the impact on the
ROM of the wearer has not been widely investigated.
In this paper we report the impact on the ROM of the lower back in the sagit-
tal plane of wearing a back support exoskeleton, which was specifically designed
to allow for a large ROM.
This work has been funded by the European Commissions as part of the project
SPEXOR under grant no. 687662. www.spexor.eu.
https://doi.org/10.1007/978-3-030-01887-0_46
240 M. B. Näf et al.

2.1 Requirements
The ROM of the human trunk in the sagittal plane relies on two joints; on the
one hand on the hip joint and on the other hand on the lumbar region, which is
often modelled as one single lumbo-sacral (L5S1) joint (See Fig. 2) at the base
of the spine. The contributions of these two joints are documented in Table 1.
Table 1. Range of motion of the trunk in the sagittal plane [6]
Flexion Extension
Hip 120◦ 15◦
◦
Lumbar 60 35◦
◦
Trunk 180 50◦
Estimates for the torque requirements range from approximately 20 Nm from

optimisation based models [7] up to peak torques of 254 Nm from bio-mechanical
measurements [8]. It is questionable, if a support torque of 100% of the bio-
mechanic estimates is desirable. Therefore, in practice, back support exoskeletons
opt towards support torques between 20–30 Nm [9].
2.2 Concept
Like for most back support exoskeletons, the goal of this one is also to reduce the
compression forces on the lumbar spine (See Fig. 1). However, special about this
purely passive back support exoskeleton is, that flexible carbon fiber beams are
placed parallel to the spine. Allowing for an increased range of motion, compared
to a completely rigid structure.
2.3 Experiments
The range of motion of the subjects (N = 2) is tested in two conditions, once

wearing an exoskeleton and once not wearing an exoskeleton. The subject were
asked to perform a ROM test in the sagittal plane while the positions and orienta-
tions of marker clusters (See Fig. 2) were recorded with motion capture cameras
(Certus Optotrak) at 50 Hz. The subjects were asked to put the same amount
of effort during the two test conditions.
The torque angle characteristic of the exoskeleton components were charac-
terised prior to the human experiments, allowing for torque estimation based on
kinematic data.
Trunk Range of Motion in the Sagittal Plane 241
Fig. 1. Forces acting on the spine (a): In order to balance the trunk against gravity, a
muscle force FMuscle is required. Since this force acts almost in parallel with the spine,
this force also compresses the spine, which is a risk factor for developing lower back
pain. The muscle force FMuscle can be reduced by introducing external forces with an
exoskeleton FExo Trunk , FExo Pelvis and FExo Thigh ; lower back support prototype (b):
The forces of the exo with a flexible back support are acting on the wearer.
Fig. 2. Angle definitions (a): hip angle (q), lumbar angle (s) and the sum of the two:
trunk angle (t); Experimental setup (b).
3 Results
Approximately 88% of the range of motion was conserved in the exoskeleton test
condition, compared to the no exoskeleton condition (See Fig. 3). Approximately
84% and 95% of the ROM in the sagittal plane of the lumbar spine and the hip
were preserved, respectively in the exoskeleton condition.
Torque estimates indicated a reduction of up to 25 Nm at the lumbo-sacral
joint while wearing the exoskeleton compared to not wearing the exoskeleton.
4 Discussion
Despite the efforts to preserve most of the trunk ROM in the sagittal plane, the
ROM was reduced to approximately 88% in the exoskeleton condition compared
to the no exoskeleton condition. The majority of the reduction (16%) originates
242 M. B. Näf et al.
Trunk angles with and without exo (N = 2)

160
143
140 ROM loss
127
120 70
Trunk flexion angle [deg]

59
100
Lumbar Lumbar
80
60
65 62
40
Hip Hip
20
0
No Exo Exo
Fig. 3. Trunk angles with and without the exoskeleton: despite the flexible design,
the overall trunk angle is reduced from 143◦ without the exoskeleton to 127◦ with
exoskeleton (ROM loss), which is approximately 88% of the original ROM.
from the lumbar spine. At the same time, a torque of approximately 25 Nm

was reducing the required torque around the lumbo-sacral joint, and thereby
decreasing the compression forces on the lower back.
This ROM might seem small, but compared to the design goal of the Robo-
mate exoskeleton, which aimed at a ROM of 60◦ in flexion extension [10], the
achieved 127◦ of the here tested exoskeleton are more than double. However, the
goal for the torque in the Robomate exoskeleton is also significantly higher with
above 75 Nm.
Most likely, there is a trade-off between the support provided by the exoskele-
ton and the ROM, that is attainable with the same effort. While less support
compared to this experiment would increase the ROM - with the limit case of
zero support, where the no exoskeleton ROM is preserved - would more support
further decrease the ROM. However, more experiments are required to verify
this hypothesis. Furthermore, factors such as body mass and height should be
investigated further in future experiments.
5 Conclusion
The majority of the trunk ROM (88%) in the sagittal plane was preserved wear-
ing the exoskeleton, while at the same time a torque of approximately 25 Nm
was provided. For industrial workers, a fine tuning between support provided
and attainable ROM, based on body mass and height might be necessary.
References
1. Eurofound: Fifth European Working Conditions Survey. Publications Office of the
European Union (2012)
2. Waters, T.R., Putz-Anderson, V., Garg, A., Fine, L.J.: Revised NIOSH equation
for the design and evaluation of manual lifting tasks. Ergonomics 36, 749–776
(1993)
Trunk Range of Motion in the Sagittal Plane 243
Ergonomics 59, 671–681 (2015)
4. Näf, M.B., Koopman, A.S., Baltrusch, S., Rodriguez-Guerrero, C., Vanderborght,
B., Lefeber, D.: Passive back support exoskeleton improves range of motion using
flexible beams. Front. Robot. AI 5, 72 (2018)
5. Muramatsu, Y., Umehara, H., Kobayashi, H.: Improvement and quantitative per-
formance estimation of the back support muscle suit. In: 2013 35th Annual Inter-
national Conference of the IEEE Engineering in Medicine and Biology Society
(EMBC) (2013)
6. Magee, D.J.: Orthopedic Physical Assessment. Elsevier Health Sciences, San Fran-
cisco (2006)
7. Millard, M., Sreenivasa, M., Mombaur, K.: Predicting the motions and forces of
wearable robotic systems using optimal control. Front. Robot. AI 4, 41 (2017)
8. Kingma, I., Baten, C.T.M., Dolan, P., Toussaint, H.M., Van Dieën, J.H., De Looze,
M.P., Adams, M.A.: Lumbar loading during lifting: a comparative study of three
measurement techniques. J. Electromyogr. Kinesiol. 11, 337–345 (2001)
9. Abdoli-Eramaki, M., Stevenson, J.M., Reid, S.A., Bryant, T.J.: Mathematical
and empirical proof of principle for an on-body personal lift augmentation device
(PLAD). J. Biomech. 40, 1694–1700 (2007)
10. Toxiri, S., Ortiz, J., Masood, J., Fernandez, J., Mateos, L.A., Caldwell, D.G.:
A wearable device for reducing spinal loads during lifting tasks: biomechanics
and design concepts. In: 2015 IEEE International Conference on Robotics and
Biomimetics. IEEE-ROBIO 2015 (2016)
Real-Time Control of Quasi-Active Hip
Exoskeleton Based on Gaussian Mixture
Model Approach
Mišel Cevzar1,2(B) , Tadej Petrič1 , Marko Jamšek1 , and Jan Babič1

1
Laboratory of Neuromechanics and Biorobotics, Department for Automation,
Biocybernetics and Robotics, Jožef Stefan Institute, Ljubljana, Slovenia
michel.cevzar@ijs.si
2
Jožef Stefan International Postgraduate School,
Jamova Cesta 39, 1000 Ljubljana, Slovenia
Abstract. Lower back pain is a major cause of disability and sick day
absences. As lower back pain can result in decreased life quality as well
as lower industrial productivity, multiple research groups and compa-
nies are looking into possible solutions. One of such solutions could
be exoskeletons, that engage and disengage the actuators depending on
the movements performed by the user. Otherwise we risk hindering the
users movements and increasing his metabolic costs. We implemented an
exoskeleton control using finite state machine combined with a Gaussian
mixture model movement classifier. By conducting a test battery with a
subject wearing the exoskeleton we were able to engage the exoskeleton
actuators when appropriate and keep them disengaged to allow a full
and unhindered range of motion. The results show our exoskeleton con-
trol correctly engages and disengages actuators based on the movements
being performed by the user.
1 Introduction
Most commercially available back exoskeletons are targeting blue collar workers.
Studies have shown correlation between physically demanding jobs and preva-
lence of lower back pain (LBP) [1]. It is important for the exoskeleton user, that
the exoskeleton does not hinder his range of motion while providing support
when needed [2]. This becomes especially important with passive exoskeletons
that use a preset inclination threshold to determine when to provide support
but also generate unwanted resistance for the user, for example: when walking
or squatting [3]. Exoskeleton should be engaged when the person is bending for-
ward and disengaged when the person is walking, chair sitting, squatting... We
suggest a clutch like system, that would engage exoskeleton support when needed
and keep it disengaged when deemed appropriate. We used pneumatic cylinders
This work was supported by the European Union’s Horizon 2020 through the
SPEXOR project (contract no. 687662).
https://doi.org/10.1007/978-3-030-01887-0_47
Real-Time Control of Quasi-Active Hip Exoskeleton 245
for actuation to simulate such a clutch. This paper presents the implementa-
tion of a control scheme on a bilateral exoskeleton, equipped with angle sensors.
Our approach is based on a finite state machine with Gaussian mixture models
(GMM) used for movement classification and a preset forward inclination thresh-
old. We performed an experiment with a male subject wearing the exoskeleton
while executing five different movements, representative for an industrial working
environment.
For the experiment we used a bilateral powered exoskeleton (Fig. 1), comprised
of a: belt, chest and leg straps, and pneumatic actuation mechanism. The con-
trol schematic is presented in Fig. 2. The inputs are absolute trunk angle in the
sagittal plane of the human and encoder angles. First, the input angles were
checked if they fall under a preset threshold of 120◦ as the exoskeleton frame
only allows movements up to 120◦ in range. A value bigger than that would
mean a sensor malfunction, in which case the actuators should disengage. Then
the GMM classified the current movement depending on the input angles. The
GMMs were calculated using data from the subject performing 20 repetitions
of the following movements: forward bending, squatting, walking, stair climb-
ing and chair sitting. A combination of classified movement and a preset trunk
angle threshold of thr = 15◦ , were conditionals for changing states in our finite
state machine. Depending on the active state, the output is an ON-OFF signal
that controls the pneumatic valves trough the pneumatics controller. We set the
Fig. 1. Photograph of exoskeleton setup.

246 M. Cevzar et al.
Fig. 2. Schematic representation of exoskeleton control.
controller to provide a maximum of 15 Nm of torque at 90◦ forward trunk incli-

nation under the condition that the user is standing. The amount of torque was
chosen by the user, based on his comfort level while wearing the exoskeleton.
During all other movements there should be no provided torque and actuators
should remain disengaged to allow unhindered movement.
At the beginning of the experiment, the subject was instructed to stand
still while wearing the exoskeleton. The sensor angles were then set to 0◦ which
corresponds to a normal upright stance. This ensured the sensors were calibrated
to the users standing posture. The subject was instructed to perform one of the
following movements in random order: forward bending, squatting, chair sitting
and stair climbing. Each movement was followed by walking over a distance of
3 m and then performing the next movement. A total of 80 movements were
carried out.
3 Results
To evaluate the effects of the exoskeleton providing back support, we measured

the provided torque and actuator activation time. We can observe that the
exoskeleton provided torque in real time and only when the subject was stand-
ing and bending forward (Fig. 3). During all other movements the actuators
remained disengaged. We calculated the human theoretical back torque, using
the data from the head, arm, torso (HAT) model as provided by Yamaguchi [4]
using the mass m and lever length lcom to calculate the torque τ :
τ = m g lcom sin(φ). (1)
g is the gravity acceleration and φ is the trunk angle we acquired with the
IMU. As shown on Fig. 4, we demonstrated a significant reduction of back torque
that needs to be generated by a human when bending forward.
Real-Time Control of Quasi-Active Hip Exoskeleton 247
Fig. 3. Test battery sample demonstrating exoskeleton provided torque during forward
bending.
Fig. 4. Theoretical human torque and exoskeleton generated total torque. The grayed
our areas indicate when the exoskeleton actuators were disengaged.
4 Discussion
Recorded data indicates that the exoskeleton is capable of producing torque
when appropriate, in real time to facilitate back support for the user. We can
observe some difference in the torque generated by the left and right actuator
respectively in Fig. 3. This happens because we compute the torque separately for
248 M. Cevzar et al.
each actuator and actuator straps are moving slightly when the user is moving.
In conclusion, we have seen that our control setup solves some of the problems
affecting other exoskeletons, namely generating unwanted resistance when the
user is walking, squatting or chair sitting. The main limitation of our work is that
we only measured one subject and acquired only kinematic data. We would like to
extend our work by using also angular velocity data for movement classification.
This would enable us to classify more movements and provide more option for
fine tuning of the classifier. We would also like to activate the left and right hip
actuator separately, depending on the movement being performed. In cases such
as stair climbing, lounging or crouch walking, a separate hip support could be
beneficial to the user.
References
1. Waddell, G., Burton, A.K.: Occupational health guidelines for the management of
low back pain at work: evidence review. Occup. Med. 51, 124–135 (2001)
2. de Rossi, S.M.M., Vitiello, N., Lenzi, T., Ronsse, R., Koopman, B., Persichetti,
A., Vecchi, F., Ijspeert, A.J., van der Kooij, H., Carrozza, M.C.: Sensing pressure
distribution on a lower-limb exoskeleton physical human-machine interface. Sensors
11(1), 207–227 (2011). http://www.mdpi.com/1424-8220/11/1/207/htm
Ergonomics, 1–11 (2015)
4. Yamaguchi, G.T.: Overview of Dynamic Musculoskeletal Modeling, pp. 3–22.
Springer, Boston (2001). https://doi.org/10.1007/978-0-387-28750-8 1
Optimizing Design Characteristics
of Passive and Active Spinal Exoskeletons
for Challenging Work Tasks
Monika Harant1(B) , Manish Sreenivasa1 , Matthew Millard1 , Nejc Šarabon2,3 ,

and Katja Mombaur1
1
Institute of Computer Engineering, Optimization, Robotics and Biomechanics
Group (ORB), 69120 Heidelberg, Germany
monika.harant@ziti.uni-heidelberg.de
2
Department for Kinesiology and Physiotherapy, Faculty of Health Sciences,
University of Primorska, 3610 Izola, Slovenia
3
Laboratory for Motor Control and Motor Behaviour, Techonology Park, S2P,
Science to Practice, Ltd., 1000 Ljubljana, Slovenia
Abstract. Spinal exoskeletons can reduce the cumulative back load of

physically demanding working tasks and, thus, have the potential to
reduce the risk of low-back injuries. In this work, we perform a compar-
ative design study of active and passive spinal exoskeletons to support
stoop-lifts of a 10 kg box. We recorded various healthy subjects perform-
ing this motion and created mathematical models of the subjects and
of active spinal exoskeletons. The spring characteristics as well as the
torque profiles are optimized to reduce the load on the subjects while
they are tracking the recorded stoop-lifts. In addition, it is ensured that
the exoskeletons remain comfortable to wear during the motion. The
obtained results are compared to results from a similar setup using a
passive spinal exoskeleton.
1 Introduction
Low-back pain is a common disorder in our society. It is estimated that 50% to

80% of all people will experience it in their lifetime [1] and, hence, has a big
impact on the industry as well. About 149 million workdays in the US are lost
per year due to low-back pain alone [2]. Wearable robots show high potential in
assisting industrial workers in physically demanding jobs, e.g. the exoskeletons
“Robo-Mate” [3] and “Laevo” (Intespring, Delft, The Netherlands).
Designing such an exoskeleton is challenging because it must be adaptable to
a wide range of industrial working environments with the following requirements:
Financial support by the European Commission within the H2020 project SPEXOR
(GA 687662) is gratefully acknowledged.
https://doi.org/10.1007/978-3-030-01887-0_48
250 M. Harant et al.
• high user support during a various set of arduous tasks

• little restriction of basic motions like sitting and walking
• high comfort so that it is wearable for several hours.
Even supporting a single task may already require a certain adaptability. As

observed in [5,6], optimal spring characteristics of a passive spinal exoskeleton
vary with the upright standing posture and lifting style of the user and the
amount of support is limited by the need to be comfortable to wear.
Here, we evaluate two exoskeleton designs during a stoop-lift using optimal
control techniques. Furthermore, we compare different objective functions for
computing spring and actuator characteristics to determine the most efficient
support for the user since the cost functions used in [5,6] did not clearly serve
the purpose to reduce the cumulative back load (CBL), which is associated with
reducing the risk of low-back injuries [4], as all joints were treated similarly.

We process recorded lifting motions of several subjects to create subject-specific
human models and reference motions for the optimization process. Then optimal
control problems with varying objective functions are solved so that the human
models combined with parameterized models of active spinal exoskeletons are
tracking the reference motions while design characteristics are optimized. The
results are compared to the ones from [6] where a purely passive exoskeleton
was used.
Fig. 1. Degrees of freedom and attachment/contact points of the combined model as

well as passive/active actuators of the exoskeletons
Optimizing Design Characteristics of Passive and Active Spinal Exoskeletons 251
2.1 Experiments
Full body kinematics, ground reaction forces and electromyography of 12 male

subjects performing stoop-lifts were recorded using an Optotrack system, two
force plates and 12 EMG sensors. The subjects lift a 10 kg box of a 0.3 m high
pedestal placed directly in front. The subjects were instructed on the general
description of a stoop-lift but were advised to lift in a way that is comfortable
for them, so bending the knees was allowed. For further description, please refer
to [5,6]. Using inverse kinematics, we reconstruct human joint kinematics from
the recorded marker data of three subjects for the use in the optimal control
context.
2.2 Human and Exoskeleton Model
The human, exoskeleton, and the box are modeled as rigid multibody systems
in the sagittal plane (Fig. 1). The properties of the modeled box match the one
used in the experiments. Mass, center of mass and inertia of each segment of
the human are derived according to the regression equations by de Leva [7].
The different muscle systems are modeled as Muscle Torque Generators (MTG)
[8]. We adjusted the strength and flexibility of the MTGs so that the activa-
tions are within reasonable limits [0, amax ] using the muscle fitting routine of
[8]. The dynamics of the system consist of the equation of motion and the acti-
vation dynamics of the MTGs. Two types of exoskeletons are considered in the
comparison between active and passive version:
P a passive exoskeleton with one spring attached between pelvis module and
torso bar and one spring attached between pelvis module and thigh bar
was used in [6]
AP the exoskeleton of point P with two additional actuators placed at the same
location as the springs (Fig. 1).
2.3 Optimal Control Problem Solution

Several least squares fitting problems of human models wearing the active
exoskeleton while tracking recorded lifting motions are solved. Additional to
the spring characteristics, torque profiles of the actuators are optimized as well.
In order to determine the best possible distribution of efforts between
exoskeleton and human, we investigate several objectives: min. active and pas-
sive muscle forces (MF), min. active muscle forces (AF), min. net lumbar and
hip torques (T). All objective functions include a term for tracking the recorded
motion and a term for minimizing the controls.
In [6] a similar LSQ-problem was solved using the passive exoskeleton P. The
cost function applied there was the same as MF. As the exoskeletons should
be comfortable to wear, limits on the interaction forces between human and
exoskeleton (based on [9]) are considered as constraints in both optimal control
settings.
252 M. Harant et al.
3 Results
The AF cost function was most effective at reducing the CBL. Comparing the
objectives, MF favors decreasing the moments at the hip joints whereas AF
favors decreasing the moments at the lumbar joint (Table 1). The effect of T is
strongly influenced by the motion (Table 1). For subject 2 with high hip flexion
and relatively straight back reducing the hip moments was preferred whereas
for subject 3 with less hip flexion and more bent back reducing the CBL was
preferred. The different cost functions also led to differing levels of tracking
accuracy (0.15–0.27◦ avg. joint angle error). The best fit was achieved with
objective T because it focuses mainly on reducing hip and lumbar torques but
this also led to undesirable peaks in the torques at the other joints.
While the spring was sufficient to reduce the CBL, adding actuators further
reduced the effort of the models’ hip extensors. The additional support provided
by the active exoskeleton resulted in higher contact forces, especially at the
pelvis. Tracking accuracy was unaffected by the type of exoskeleton used.
Table 1. Reduction (–) or increase (+) of lumbar and hip moments of 3 subjects with
respect to the results from [6]
Cost func. Reduction of CBL (%) Red. of peak lumbar Reduction of hip
mom. (%) moment (%)
MF −2.3 −0.2 2.4 −0.5 1.5 3.5 −3.1 −4.3 −4.5
AF −19.5 −11.2 −18.1 −13.3 −9.4 −14.0 5.8 3.9 0.0
T −5.9 −3.0 −21.0 −3.2 −0.5 −16.4 −3.1 −7.6 −1.0
4 Discussion
This study is motivated by the need to reduce the risk of low-back injuries
during demanding work tasks. Wearable robots can meet this demand, but it is
challenging to develop a design that is applicable for a wide range of users and
working environments.
Our results show that the objective function highly influences the computed
exoskeleton design characteristics and thus, also the provided support by the
exoskeleton. The cost function applied in [5,6] was not ideal because it favors
reducing the hip moments although reducing CBL is more effective at reducing
the risk of injury.
Note that the preferred cost function also depends on the design of the
exoskeleton. If the design changes, e.g. if a part of the force generated at the hip
joint is transmitted to the back, a different distribution of the force generation
might be desired.
The reference motions are taken from experiments where no exoskeleton was
used. Thus, in the computations it was assumed that the users will lift the box
Optimizing Design Characteristics of Passive and Active Spinal Exoskeletons 253
the same way even if they are wearing an exoskeleton but it is likely that they
will adapt to it to a certain extent. For future work, experiments with subjects
wearing an exoskeleton would give an insight on how much the users comply
with it and adjust their motions.
5 Conclusion
Our results indicate that a reduction in CBL provided by the exoskeleton

depends strongly on how the support of the hip and back are coordinated during
the lift - even when the kinematics of the lift are nearly identical.
References
1. Rubin, D.I.: Epidemiology and risk factors for spine pain. Neurol. Clin. 25(2), 353–
371 (2007)
2. Guo, H.R., Tanaka, S., Halperin, W.E., Cameron, L.L.: Back pain prevalence in us
industry and estimates of lost workdays. Am. J. Public Health 89(7), 1029–1035
(1999)
3. Stadler, K., Elspass, W., van de Venn, H.: Robo-mate: exoskeleton to enhance indus-
trial production. In: Mobile Service Robotics, pp. 53–60. WORLD SCIENTIFIC
(2014)
4. Coenen, P., Kingma, I., Boot, C.R., Bongers, P.M., van Dieën, J.H.: Cumulative
mechanical low-back load at work is a determinant of low-back pain. Occup. Environ.
Med. 71(5), 332–337 (2014)
5. Harant, M., Sreenivasa, M., Millard, M., Šarabon, N., Mombaur, K.: Parameter
optimization for passive spinal exoskeletons based on experimental data and optimal
control. In: IEEE/RAS International Conference on Humanoid Robots (Humanoids
2017) (2017)
6. Harant, M., Sreenivasa, M., Millard, M., Šarabon, N., Mombaur, K.: Identifica-
tion of optimal compliance in passive spinal exoskeletons based on motion capture
experiments and subject-specific musculoskeletal models, unpublished
7. De Leva, P.: Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters. J.
BIOMECH 29(9), 1223–1230 (1996)
8. Millard, M., Kleesattel, A.L., Harant, M., Mombaur, K.: A reduced muscle model
and planar musculoskeletal model fit for synthesis of whole body movements. J.
BIOMECH, submitted for publication
9. Huysamen, K.: Assessment of an active industrial exoskeleton to aid dynamic lifting
and lowering manual handling tasks. Appl. Ergon. 68, 125–131 (2018)
Human Modeling and Simulation
for Neurorehabilitation Engineering
Calibration and Validation of a Skeletal
Multibody Model for Leg-Orthosis Contact
Force Estimation
Francisco Mouzo1, Urbano Lugris1, Javier Cuadrado1(&),

Josep M. Font-Llagunes2, and Francisco J. Alonso3
1
Laboratory of Mechanical Engineering, University of La Coruña, Ferrol, Spain
javier.cuadrado@udc.es
2
Department of Mechanical Engineering and the Biomedical Engineering
Research Centre, Technical University of Catalonia, Barcelona, Spain
3
Department of Mechanical, Energetics and Materials Engineering,
University of Extremadura, Badajoz, Spain
Abstract. Estimation of contact forces between lower limb and orthosis during
gait is useful to prevent skin issues in subjects wearing this type of assistive
devices. While inverse-dynamics based gait analysis of multibody models is
difficult to apply due to the limited accuracy of motion capture systems, a
forward-dynamics based analysis in which leg and orthosis are considered as
independent entities is shown to provide acceptable results. Contact model
parameters are calibrated through comparison of measured and calculated
bending torque at the orthosis location where a load cell is installed, and the
attained correlation allows to validate the model.
1 Introduction
The authors developed an active knee-ankle foot orthosis (KAFO) as an assistive

device for the gait of spinal cord injured (SCI) subjects [1]. The prototype (Fig. 1a)
features a brushless DC motor at knee level to provide knee motion during the swing
phase and an inertial sensor at shank level to detect motion intention and trigger the
swing cycle.
For bilateral patients, the control algorithm that launches the orthosis swing cycle is
based on the orientation of both shanks obtained after processing data coming from the
two inertial sensors. To have the maximum time to complete the cycle before the foot
touches down again, the swing cycle must be launched as soon as motion intention is
detected, but false positives cannot be allowed due to fall risk.
For unilateral patients (Fig. 1b), one single inertial sensor is available, which means
less information to detect motion intention. Therefore, a load cell is included in the
This work was funded by the Spanish MINECO under project DPI2015-65959-C3-1-R,
cofinanced by the EU through the EFRD program, and by the Galician Government under grant
ED431B2016/031.

https://doi.org/10.1007/978-3-030-01887-0_49
258 F. Mouzo et al.
Fig. 1. Prototype of active knee-ankle-foot orthosis: (a) actuator and sensors; (b) unilateral
patient wearing the orthosis.
orthosis structure to alleviate this problem (while avoiding the many issues raised by
pressure sensors): it is formed by two strain gauges located in the front and rear faces of
the external orthosis upright at thigh level, just above the DC motor. The load cell
detects the increment in load suffered by the orthosis during stance, as the knee bends
under the subject’s weight and the leg presses against the orthosis. This serves to
reliably detect stance and, hence, provides the necessary information to guarantee a
safe orthosis operation.
In this work, the load cell is used to calibrate and validate a computational
multibody model of a subject wearing the orthosis which includes a leg-orthosis contact
model.
An experiment was conducted with a 49-year-old adult spinal-cord-injured male, mass

82 kg and height 1.90 m, requiring, in order to walk, a KAFO on his left leg and an
ankle-foot orthosis on his right leg, along with a pair of crutches (Fig. 1b). In the
experiment, he was wearing our active KAFO in his left leg, and walked over two
embedded force plates (AMTI, AccuGait sampling at 100 Hz) with the help of two
instrumented crutches that measured the ground contact forces at their tips. Motion was
captured by 12 optical infrared cameras (Natural Point, OptiTrack FLEX:V100 also
sampling at 100 Hz) that computed the position of 43 optical markers. The study was
approved by the institutional ethical committee and the subject gave his informed
consent.
A 59 degree-of-freedom multibody model of the subject wearing orthoses and
crutches was developed as in [2] but, this time, links of active orthosis and left leg were
modeled as independent entities (Fig. 2a). Hence, relative motion and contact forces
Calibration and Validation of a Skeletal Multibody Model 259
between leg and orthosis could be estimated when running a forward dynamic simu-
lation of the model that tracked the subject’s captured motion through a CTC control
scheme [3]. Figure 2b illustrates the modeling of leg and orthosis in the multibody
model, along with the contact spring-damper elements at hip and knee levels.
Fig. 2. Model of unilateral patient with active orthosis: (a) graphical output; (b) modeling of leg
and orthosis as independent entities.
In a previous work [4], the variations of leg-orthosis contact forces and misalign-
ments for bilateral subjects were studied as functions of the stiffness/damping
parameters of the contact elements. Figure 3 shows the interaction forces at right hip
level for three different values of the parameters for the case of a bilateral SCI female.
The blue and pink areas correspond to the swing phase of the right and left leg,
Fig. 3. Contact forces between leg and orthosis at hip connection for three different values of the
parameters of the contact elements.
260 F. Mouzo et al.
respectively. The curves show a similar shape for the three values of the parameters,
just differing in a vertical scaling.
In this work, to calibrate the stiffness/damping parameters the load cell was used. It
measures the strain due to axial stresses caused by the bending effects of the forces
acting on the KAFO above the location of the gauges, i.e. leg-orthosis contact force at
the thigh strap and weight plus inertia forces of the orthosis thigh body (Fig. 2b). The
load cell was calibrated with known loads, so as to make it a transducer of the bending
torque. Then, the history of such magnitude measured in the experiment was compared
with that calculated for different values of the contact model parameters, looking for a
good correlation, so as to calibrate the stiffness/damping of the contact elements.
3 Results
Figure 4 shows the history of the bending torque at the load cell location obtained by
measurement and calculation using the best set of the contact parameters selected in
several manual iterations. The RMS error is 2.83 Nm.
Fig. 4. Comparison of the torque measured by the load cell and calculated from the model.
4 Discussion
Results shown in the previous section confirm the validity of the proposed model and
analysis method to estimate the contact forces between leg and orthosis during gait. It
must be pointed out that, while contact forces could be hardly obtained through an
inverse-dynamics based approach, as it is not easy to discriminate the motions of leg
and orthosis with the accuracy provided by current motion capture systems, application
of a forward-dynamics based approach, in which only the motion of leg or orthosis is
measured and the counterpart is left to its own dynamics, provides acceptable results.
Calibration and Validation of a Skeletal Multibody Model 261
5 Conclusions
In this work, it has been demonstrated that, for subjects wearing knee-ankle-foot
orthoses and crutches, the contact forces between leg and orthosis during gait can be
reasonably estimated through a forward-dynamics based analysis of a full multibody
model in which legs and orthoses are considered as independent entities.
References
1. Font-Llagunes, J.M., Clos, D., Lugris, U., Alonso, F.J., Cuadrado, J.: Design and
experimental evaluation of a low-cost robotic orthosis for gait assistance in subjects with
spinal cord injury. In: Gonzalez Vargas, J., et al. (eds.) Wearable Robotics: Challenges and
Trends, pp. 281–286. Springer (2016)
2. Lugris, U., Carlin, J., Luaces, A., Cuadrado, J.: Gait analysis system for spinal cord injured
subjects assisted by active orthoses and crutches. J. Multi-body Dyn. 227, 363–374 (2013)
3. Mouzo, F., Lugris, U., Pamies-Vila, R., Cuadrado, J.: Skeletal-level control-based forward
dynamic analysis of acquired healthy and assisted gait motion. Multibody Syst. Dyn. 44(1),
1–29 (2018)
4. Mouzo, F., Lugris, U., Cuadrado, J., Font-Llagunes, J.M., Alonso, F.J.: Evaluation of
motion/force transmission between passive/active orthosis and subject through forward
dynamic analysis. In: Ibañez, J., et al. (eds.) Converging Clinical and Engineering Research
on Neurorehabilitation II, pp. 815–819. Springer (2016)
A Continuous and Differentiable
Mechanical Model of Muscle Force
and Impedance
Matthew Millard1(B) , David Franklin2 , and Walter Herzog3

1
Optimization, Robotics and Biomechanics Group, Heidelberg University,
Heidelberg, Germany
matthew.millard@iwr.uni-heidelberg.de
2
Neuromuscular Diagnostics Group, Technical University of Munich,
Munich, Germany
david.franklin@tum.de
3
Human Performance Laboratory, University of Calgary, Calgary, Canada
wherzog@ucalgary.ca
Abstract. No single muscle model exists that has the same mechanical
impedance and force development properties as biological muscle. It is
essential to develop a muscle model with the same force limitations and
impedance as biological muscle, especially for predictive simulations, as
these properties are taken into account when choosing a posture for a
specific task. We propose a mechanics-based muscle model that has the
same impedance and force development properties as biological muscle by
making a small topology change that turns titin, an enormous viscoelastic
protein, from acting in parallel to the cross-bridges to acting in series
with the cross-bridges.
1 Introduction
The stiffness and damping properties of muscle affect not only how the body
responds to perturbations, but also the postures that people choose to adopt
for a particular task [1]. Unfortunately, no single muscle model captures all of
the mechanical properties of muscle: Hill-type muscle models can have a region
of negative stiffness on the descending limb of the active-force-length curve [2],
Huxley-type models over-estimate the forces developed during rapid eccentric
contractions [3], and the active spring-damper muscle models used in motor
control simulations [4] have not been extended to include the well-known varia-
tion of muscle force with length and velocity. While curve fitting has been used
to artificially add the missing stiffness and damping forces to Hill models [5], it
is unlikely that this modification will generalize outside of the data used for the
curve fit.
Financial support from Deutsche Forschungs Gemeinschaft grant no. MI 2109/1-1 is

gratefully acknowledged.
https://doi.org/10.1007/978-3-030-01887-0_50
A Continuous and Differentiable Mechanical Model 263
The forces developed by muscle during ramp-stretch experiments provide

insight into the mechanisms that underlie stiffness and damping in muscle.
When active muscle is stretched, the force increase follows a stereotypical pattern
(Fig. 1): a rapid increase in force over a short-range, followed by a more gradual
increase over a longer range. As the rate of lengthening is increased, the peak
force developed in the short-range increases [6] consistent with a damping ele-
ment. In contrast, the long-range tension profiles appear to be strictly a function
of length: the long-range profiles of Figs. 7A–C of [6] are nearly identical when
plotted against muscle length, consistent with an elastic element. Furthermore,
the stiffness of the long-range force profile does not appear to change with muscle
length nor active force, and therefore is not dependent on the number of attached
cross-bridges: Fig. 5A of [7] shows the active force profiles of 8 ramp-stretches
of a rat soleus done at lengths ranging from ˜M = 0.51 − 0.88, each with nearly
identical long-range force profile slopes. Taken together, the ramp-stretch exper-
iments of [6,7] suggest that the mechanism of force development in lengthening
muscle is not due to cross-bridge cycling, but instead due to the lengthening of
a viscoelastic element capable of sustaining large strains.
Fig. 1. The stereotypical tension profile of an activated muscle being stretched at a

constant velocity. Note that ˜M is the length of the muscle fiber divided by M
o the
length at which the fiber develops its maximum active isometric force, and f˜M is the
force of the muscle divided by its maximum active isometric force foM .
There is only one known viscoelastic element in a sarcomere capable of sus-

taining large strains: titin. Titin is enormous, spanning half the length of a
sarcomere, is 100× more compliant than either myosin or actin, and develops
significant damping forces [8]. The fantastic properties of titin have been pro-
posed to affect active muscle force by attaching to actin, as detailed in the
winding-filament theory and realized in a model [9]. Here we present an alter-
native mechanism which can explain the short and long-range force profiles
observed in eccentrically contracting muscle. While it is clear that titin plays
a central role in the development of active and passive muscle force, the precise
mechanism by which this occurs is not clear.
264 M. Millard et al.
2 Model
A slight modification to the topology of the sarcomere can change titin from
acting in parallel to the cross-bridges (Fig. 2A) to acting in series (Fig. 2B), which
provides the active muscle model with titin’s stiffness and damping properties.
It is known that titin connects to the actin filament of the adjacent sarcomere
[10]. Thus if the actin filament can transfer a large fraction of its tension to the
titin molecule in the neighboring sarcomere, and is free to move with respect
to the Z-line, then it is possible that titin could function in series with myosin
(Fig. 2B). We develop a muscle model under the assumption that titin acts in
series with myosin (Fig. 2C) as the force response of lengthening muscle bears a
striking similarity to that of a spring-damper.
Fig. 2. Although titin is normally treated as a parallel element (A), it may function
as a series element (B) since it is connected to the actin filament of the neighboring
sarcomere. Assuming symmetry, we model a muscle as a scaled sarcomere that has a
series element for titin (C).
To model this new topology we introduce two small masses: one between the
titin element and its neighboring actin filament, mN1 , and another, mM , between
the myosin filament and the applied external force. Since the damping forces of
titin scale with its strain [8], we model titin as
f N = foM (f K (N /N

s ) + f ( /s )(βv ))
D N N N
(1)
a nonlinear spring-damper where f N is the tension developed by titin, foM is

the maximum active isometric force of the muscle, N is the length of titin, N s
is the slack-length of titin, v N is the lengthening rate of titin, β is a damping
coefficient, f K is the normalized passive-force-length curve, and f D is a smooth
step function that increases from 0 to 1 as a function of N . The force developed
by the cross-bridge cycling between the myosin and actin filaments is modeled
as
f CE = foM af L (N1 , M )f V (v M − v N1 ) (2)
A Continuous and Differentiable Mechanical Model 265
being proportional to the activation a of the muscle and scaled by the number
of available attachment sites f L and a modified force-velocity curve f V . The
force-velocity curve f V follows Hill’s hyperbola for concentric contractions, with
a linear extrapolation on the eccentric side: when stretched this modified model
will cause the myosin and actin filaments to move together, stretching titin.
Since the actin and myosin elements are now independent we have derived an
active-force surface f L which evaluates how many active sites for cross-bridge
attachment are available given the actin-myosin overlap and interference. Finally,
the acceleration of the titin and myosin elements are given by
1
v̇ M = (f T cos α − f CE + f Z (N1 , M )) (3)
mM
1
v̇ N1 = N1 (f CE − f N1 + f N2 ) (4)
m
where f Z is the force developed when two neighboring myosin filaments come
into contact (pinching the Z-line), and f T cos α is the force of the tendon along
the fiber. We use conventional models of activation dynamics, tendon elasticity,
and pennation [2] for this model. To test this model we reproduce, in simulation,
the ramp lengthening experiments of [6] using the proposed two-element fiber
model with titin as a series viscoelastic element.
Fig. 3. The simulation results of a maximum-activation ramp-stretch test compared

to experimental data [6].
3 Results and Conclusion

The simulation results (Fig. 3) of the ramp-stretch experiments of [6] show that
the model is a good candidate: it has a similar short-range and long-range force
profile as biological muscle. If titin does function as a series element this will
change the way that muscle force is understood to develop, affecting many
researchers who study musculoskeletal systems. However, much work remains
to be done: these ideas need to be experimentally tested, and many details need
to be developed before this model is complete.
266 M. Millard et al.
References
1. Trumbower, R.D., Krutky, M.A., Yang, B.S., Perreault, E.J.: Use of self-selected
postures to regulate multi-joint stiffness during unconstrained tasks. PloS One
4(5), e5411 (2009)
2. Millard, M., Uchida, T., Seth, A., Delp, S.L.: Flexing computational muscle: mod-
eling and simulation of musculotendon dynamics. J. Biomech. Eng. 135(2), 021005
(2013)
3. van den Bogert, A.J., Gerritsen, K.G.M., Cole, G.K.: Human muscle modelling
from a user’s perspective. J. Electromyogr. Kinesiol. 8(2), 119–124 (1998)
4. Tee, K.P., Burdet, E., Chew, C.M., Milner, T.E.: A model of force and impedance
in human arm movements. Biol. Cybern. 90(5), 368–375 (2004)
5. McGowan, C.P., Neptune, R.R., Herzog, W.: A phenomenological model and
validation of shortening-induced force depression during muscle contractions. J.
Biomech. 43(3), 449–454 (2010)
6. Herzog, W., Leonard, T.R.: Force enhancement following stretching of skeletal
muscle: a new mechanism. J. Exp. Biol. 205(9), 1275–1283 (2002)
7. Krylow, A.M., Sandercock, T.G.: Dynamic force responses of muscle involving
eccentric contraction. J. Biomech. 30(1), 27–33 (1997)
8. Herzog, J.A., Leonard, T.R., Jinha, A., Herzog, W.: Are titin properties reflected
in single myofibrils? J. Biomech. 45(11), 1893–1899 (2012)
9. Schappacher-Tilp, G., Leonard, T.R., Desch, G., Herzog, W.: A novel three-
filament model of force generation in eccentric contraction of skeletal muscles.
PLoS One 10(3), e0117634 (2015)
10. Gregorio, C.C., et al.: The NH2 terminus of titin spans the Z-disc: its interaction
with a novel 19-kD ligand (T-cap) is required for sarcomeric integrity. J. Cell Biol.
143(4), 1013–1027 (1998)
SimCP: A Simulation Platform to Predict Gait
Performance Following Orthopedic
Intervention in Children with Cerebral Palsy
Friedl De Groote(&), Lorenzo Pitto, Hans Kainz, Antoine Falisse,

Eirini Papageorgiou, Mariska Wesseling, Sam Van Rossom,
Kaat Desloovere, and Ilse Jonkers
Faculty of Movement and Rehabilitation Sciences,

KU Leuven, Leuven, Belgium
{friedl.degroote,lorenzo.pitto,hans.kainz,
antoine.falisse,eirini.papageorgiou,
mariska.wesseling,sam.vanrossom,kaat.desloovere,
ilse.jonkers}@kuleuven.be
Abstract. We present a simulation platform that will enable clinicians to

evaluate the effect of different treatment options on gait performance in children
with cerebral palsy (CP) in order to select the treatment with the highest
potential to normalize the patient’s gait pattern. We present a case study to
demonstrate the use of the platform. We created a neuro-musculoskeletal model
of a 10-year old female child with mild spastic triplegic CP (GMFCS II) who
was treated with single-event multilevel surgery based on medical imaging and
motion capture data collected before the surgery. Based on this model, we
predicted that the treatment would reduce the capability gap, i.e. the torque
deficit of the patient with respect to the joint torques needed for normal walking.
This prediction was in accordance with the closer-to-normal post-treatment gait
kinetics of the child.
1 Introduction
Three-dimensional gait analysis is typically used in the clinical decision-making pro-

cess in children with cerebral palsy (CP). However, decision-making is complicated
because gait impairments result from the complex interaction between changes in
musculoskeletal geometry, muscle weakness and impaired motor control. Since mus-
culoskeletal modeling in combination with simulations of motion has the potential to
non-invasively predict the effect of an intervention on gait performance, it might be a
useful tool to establish a more data-informed decision on treatment selection compared
to experimental gait analysis alone.
This work was supported by IWT-TBM Grant 140184. Antoine Falisse is funded by a Ph.D. grant
strategic basic research (SB) from the Research Foundation Flanders (FWO). Hans Kainz is
funded by a H2020-MSCA individual fellowship (796120).

https://doi.org/10.1007/978-3-030-01887-0_51
268 F. De Groote et al.
First, we present a simulation platform that will enable clinicians to compare the
effect of different treatment modalities on gait performance in order to identify the
treatment that has the highest potential to result in a closer-to-normal gait pattern, using
a musculoskeletal model containing personalized musculoskeletal geometry and con-
trol. Second, we present a case study in which the simulation platform was used to
investigate if pre-intervention motion capture data in combination with a muscu-
loskeletal model containing personalized musculoskeletal geometry and control could
be used to predict gait performance after a single-event multilevel surgery in a child
with CP.
The simulation framework uses individualized musculoskeletal models as well as the

joint kinematics from a typically developing (TD) child to calculate the ability of the
patient to adopt a normal gait pattern. We quantify the ability of the patient by the
capability gap, which is the difference between the joint moments the model repre-
senting the CP child can produce, and the joint moments required for a TD gait pattern
[1]. Bony geometries and muscle paths are personalized based on magnetic resonance
imaging (MRI) [2]. The patient-specific lack of selective motor control is modeled
based on muscle synergies computed from EMG measured during the pre-operative
gait analysis [3]. We assume that the intervention does not affect the number of syn-
ergies and the relative weighting of different muscles within a synergy. We developed a
GUI to create the post-operative model by modifying the musculoskeletal geometry of
the pre-operative model to mimic the effect of dedicated surgical interventions (e.g.
extension osteotomy).
The pre- and post-operative model are subsequently used to compute the capability
gap. To this aim, we performed synergy-constrained static optimization based on the
patient’s model while imposing joint kinematics and kinetics from a TD child. Static
optimization computes the contribution of individual muscles and reserve actuators to
the movement. Reserve actuators are not physiological and therefore the reserve
actuator torques represent the gap between the required joint torques and the joint
torque the muscles can produce, i.e. the capability gap according to our definition. The
capability gap is a measure of the inability of the patient to perform a typical gait
pattern. By comparing the capability gap pre- and post-treatment, as well as between
different treatment options, the effect of the surgery on the patient’s capability to
perform a normal gait pattern can be assessed.
This workflow was applied for a 10-year old female child with mild spastic triplegic
CP (GMFCS II). The patient underwent a single-event multilevel surgery including
bilateral psoas lengthening, rectus femoris transfer, tibia derotation, femur derotation
and extension, and patella distalisation. Clinical pre-operative assessment of the patient
showed a triple flexion gait pattern with increased femoral anteversion and increased
muscle tone in the bilateral hip flexors, hip adductors, rectus femoris, hamstrings,
soleus and gastrocnemius.
During the pre- and post-surgery gait analysis (12 months post-surgery), marker
trajectories, EMG (eight muscles/leg) and ground reaction forces were measured.
SimCP: A Simulation Platform to Predict Gait Performance 269
Pre- and post-surgery musculoskeletal models were created as described above, and
were used to evaluate the patient’s pre- and post-surgery capability gap, the latter with
and without the corrective effect of the rectus femoris transfer.
The predicted capability gap was validated against the post-surgery gait analysis
data. To this aim, we compared the pre- and post- surgery model-based capability gap
(root mean square value) to the root mean square difference (RMSD) between mea-
sured TD kinetics, and the patient’s pre- and post-surgery kinetics.
Post-operative joint kinetics during gait were more similar to TD kinetics than pre-
operative joint kinetics. The post-surgery RMSD between patient and TD kinetics were
smaller for six out of eight analyzed joints (average over all joints 0.28 ± 0.18 Nm/kg
and 0.17 ± 0.10 Nm/kg for pre- and post-surgery, respectively). The pre- and post-
surgery joint kinetics for left hip adduction are shown in Fig. 1.
Fig. 1. Experimental and model-based hip adduction joint moments and the capability gap pre-
(A) and post-surgery (B).
270 F. De Groote et al.
Our model-based simulations predicted a reduction of the root mean square

capability gap post-surgery, indicating that the ability of the patient to produce normal
gait kinetics increased. The RMS post-surgery capability gap was smaller for seven out
of eight analyzed joints (average over all joints 0.14 ± 0.06 Nm/kg and
0.10 ± 0.03 Nm/kg for pre- and post-surgery, respectively), with the rectus femoris
transfer further reducing the capability gap for hip and knee in the sagittal plane. The
pre- and post-surgery capability gap for left hip adduction are shown in Fig. 1.
4 Conclusion
Our simulations predicted that the ability of the patient to produce healthy gait kinetics
improved. This prediction is in accordance with the measured kinetics that was closer-
to-normal after the intervention. At this stage, the simulation platform computes the
theoretical capability and does not account for compensation strategies the patient
adopts to overcome this capability gap. This might explain the difference between the
capability gap and the RMSD between TD and CP kinetics. We are in the process of
validating our framework by performing additional case studies. In addition, we will
further improve model personalization by including changes in muscle-tendon prop-
erties and spasticity, and we are developing predictive simulations of movement
kinematics. This first case study demonstrates the potential of our simulation platform
to predict the effect of a surgical intervention on gait performance.
References
1. Afschrift, M., De Groote, F., De Schutter, J., Jonkers, I.: The effect of muscle weakness on the
capability gap during gross motor function: a simulation study supporting design criteria for
exoskeletons of the lower limb. Biomed. Eng. OnLine 13, 111 (2014)
2. Bosmans, L., Wesseling, M., Desloovere, K., Molenaers, G., Scheys, L., Jonkers, I.: Hip
contact force in presence of aberrant bone geometry during normal and pathological gait.
J. Orthop. Res. 32(11), 1406–1415 (2014)
3. Meyer, A., Eskinazi, I., Jackson, J.N., Rao, A.V., Patten, C., Fregly, B.J.: Muscle synergies
facilitate computational prediction of subject-specific walking motions. Front. Bioeng.
Biotechnol. 4, 77 (2016)
Bio-inspired Walking: From Humanoids
to Assistive Devices
Renaud Ronsse(B)
Institute of Mechanics, Materials, and Civil Engineering, The Institute of

Neuroscience, and Louvain Bionics, Université catholique de Louvain,
Louvain-la-Neuve, Belgium
renaud.ronsse@uclouvain.be
Abstract. In this document, a general framework for generating bio-

inspired walking is outlined. This framework relies on the combination
of a musculoskeletal model of the leg and different bio-inspired neural
principles for providing activation signals to these virtual muscles. We
explored this framework both for humanoid walking – achieving both
versatile and human-like gaits – and for human walking assistance.
1 Introduction
Observing animal and human locomotion reveals the large gap that is still exist-
ing between biological motor skills and their robotic counterparts. Indeed, the
most advanced robots are still poor competitors as compared to biological agents
regarding versatility, agility, and energy efficiency. To cite one recent example
among the many, [1] reported that a nonathletic human could complete all the
tasks of the DARPA Robotics Challenge about 20 times faster than the win-
ning robot; although this contest likely gathered the best humanoid robots to
date. Consequently, bio-inspiration has became a fantastic incentive for designing
highly skilled robots, particularly regarding agile locomotion [2].
In this document, a survey of our recent contributions to bio-inspired con-
trol of bipedal walking is provided. These contributions all relied on a common
ground, namely the modeling of the human neuro-musculo-skeletal apparatus.
This approach postulates that emulating human-like neuromuscular impedance
in artificial leg control is crucial for copying or restoring proper interactions with
the environment [3], or reaction to perturbations [4]. First, an overview of this
neuromuscular framework is provided. Next, the most significant results obtained
in generating humanoid walking or human walking assistance are reported.
This work was supported by the EU within the WALK-MAN (FP7-ICT-2013-10,

Grant Agreement #16744574) and the CYBERLEGsPlusPlus (H2020-ICT-2016-1,
Grant Agreement #731931) projects, and the Belgian F.R.S.-FNRS.
https://doi.org/10.1007/978-3-030-01887-0_52
272 R. Ronsse
2 General Framework
Our framework relies first on the modeling of the skeletal chain of the human
leg, i.e. through the degrees of freedom of the hip, knee, and ankle. Then, these
joints are actuated by virtual muscles whose dynamics obey the ones of the well-
know Hill model [6]. In short, this model captures muscle force production as
the product of a low-pass filtered version of their level of activation, and non-
linear bio-relevant force-length and force-velocity relationships. Figure 1 shows
this musculoskelal framework printed over the COMAN humanoid robot [7].
The walking controller has then to produce appropriate activation signals to
these virtual muscles. We explored two different approaches for generating such
activation signals, that we validated with humanoid walking and human walking
assistance, respectively.
Fig. 1. The biped’s degrees of freedom are actuated by virtual Hill muscles (in red).
These muscles can be commanded by a combination of feedback-driven reflexes (in
blue), a feed-forward CPG (in black), or feed-forward descending primitives (not dis-
played). See [5] for details.
3 Humanoid Walking by Combining Reflexes and a CPG
Since 2010, Geyer and colleagues have been developing a similar framework,
where muscle activations are exclusively produced by feedback-driven reflexes.
This provided walking gaits strikingly similar to human ones [8,9]. However,
this approach requires to re-tune the many parameters of the resulting model
for converging to different gait features, like forward speed. In [5], we extended
this model with a Central Pattern Generator (CPG), i.e. a descending neural
circuit that is known to play a central role in rhythmic motor behaviors of
vertebrates, possibly including humans [10]. The influence of this CPG was found
Bio-inspired Walking: From Humanoids to Assistive Devices 273
to be more prominent for proximal muscles (see Fig. 1), guiding the legs with
its intrinsic frequency, that can be used to modulate the forward speed. We
obtained stable and robust walking patterns from 0.45 m/s to 0.9 m/s with the
simulated COMAN robot [5]. Scaled to the actual robot size, this is comparable
to a comfortable walking range of a healthy human adult.
Our most recent results extended this framework to running, stair climbing,
and, last but not least, to steering control, so that the robot can be controlled
in real-time with only two high-level inputs: one providing a forward speed ref-
erence, and the other providing the desired heading angle.
.
Fig. 2. Muscle activations are generated through combination of 6 phase-dependent

primitives. The combination weights depend on the detected task (blue arrow) and
walking frequency (red arrow). Adapted from [11].
4 Human Walking Assistance by Artificial Primitives

We also explored a similar approach for walking assistance, this time through
an exoskeleton being connected in parallel to the human user’s legs. However,
in order to avoid complex user- and task-dependent tuning of the many param-
eters of the humanoid neuromuscular model, we adopted another bio-inspired
formalism, namely the one of motor primitives. Motor primitives are networks
of spinal neurons that form one basic module activating a determined set of
muscles. Through proper recombination, they can thus generate a large set of
muscle stimulations for different locomotion tasks, like walking and stair ascend-
ing/descending. Such primitives were identified in animals [12] and humans [13]
through dedicated signal decomposition techniques.
In [11,14], we combined the musculoskeletal model outlined in Sect. 2 with
precomputed activation primitives that were generated from kinematic and walk-
ing data of different locomotion tasks (walking at different cadences, and stair
climbing/descending), and with a higher-level layer detecting the performed loco-
motion task in real-time [15], see Fig. 2. Delivering this assistance through the
exoskeleton series developed in [16,17], we indeed proved the assistance effective-
ness in decreasing the metabolic cost of walking on a treadmill [11], or decreasing
the completion time of a route combining stairs and overground walking [14].
274 R. Ronsse
5 Conclusion
In this paper, we reviewed recent developments in designing bio-inspired neuro-
musculo-skeletal models of the leg to generate humanoid walking, or walking
assistance. The key ingredient of this approach is to include a virtual muscular
layer in the control architecture, so that the leg joints are governed by human-like
impedance and dynamics. Our research in humanoid walking led to locomotion
patterns whose features (cadence, heading direction) can be changed on the fly,
while displaying key characteristics of human walking: stretched knee during
stance, heel strike, etc. This was obtained by combining feedback-driven reflexes
and a feed-forward CPG. Regarding human locomotion assistance, the controller
combined the musculoskeletal model with motor primitives and proved to be
efficient in providing assistance to different locomotion tasks.
References
1. Krotkov, E., Hackett, D., Jackel, L., Perschbacher, M., Pippine, J., Strauss, J.,
Pratt, G., Orlowski, C.: The DARPA robotics challenge finals: results and per-
spectives. J. Field Robot. 34(2), 229–240 (2017)
2. Ijspeert, A.J.: Biorobotics: using robots to emulate and investigate agile loco-
motion. Science 346(6206), 196–203 (2014). http://dx.doi.org/10.1126/science.
1254486
3. Lee, H., Rouse, E.J., Krebs, H.I.: Summary of human ankle mechanical impedance
during walking. IEEE J. Transl. Eng. Health Med. 4, 1–7 (2016)
4. Eilenberg, M., Geyer, H., Herr, H.: Control of a powered ankle-foot prosthesis
based on a neuromuscular model. IEEE Trans. Neural Syst. Rehabil. Eng. 18(2),
164–173 (2010)
5. Van der Noot, N., Ijspeert, A.J., Ronsse, R.: Bio-inspired controller achieving for-
ward speed modulation with a 3D bipedal walker. Int. J. Robot. Res. 37(1), 168–
196 (2018)
6. Hill, A.V.: The heat of shortening and the dynamic constants of muscle. Proc. R.
Soc. London B Biol. Sci. 126(843), 136–195 (1938)
7. Tsagarakis, N.G., Morfey, S., Cerda, G.M., Zhibin, L., Caldwell, D.G.: COMpliant
huMANoid COMAN: optimal joint stiffness tuning for modal frequency control. In:
2013 IEEE International Conference on Robotics and Automation, pp. 673–678.
IEEE, May 2013
8. Geyer, H., Herr, H.: A muscle-reflex model that encodes principles of legged
mechanics produces human walking dynamics and muscle activities. IEEE Trans.
Neural Syst. Rehabil. Eng. 18(3), 263–273 (2010)
9. Song, S., Geyer, H.: A neural circuitry that emphasizes spinal feedback generates
diverse behaviours of human locomotion. J. Physiol. 593(16), 3493–3511 (2015)
10. Minassian, K., Hofstoetter, U.S., Dzeladini, F., Guertin, P.A., Ijspeert, A.: The
human central pattern generator for locomotion: does it exist and contribute to
walking? Neuroscientist 23(6), 649–663 (2017)
11. Garate, V.R., Parri, A., Yan, T., Munih, M., Lova, R.M., Vitiello, N., Ronsse,
R.: Walking assistance using artificial primitives: a novel bioinspired framework
using motor primitives for locomotion assistance through a wearable cooperative
exoskeleton. IEEE Robot. Autom. Mag. 23(1), 83–95 (2016). http://dx.doi.org/
10.1109/MRA.2015.2510778
Bio-inspired Walking: From Humanoids to Assistive Devices 275
12. Giszter, S.F.: Motor primitives-new data and future questions. Curr. Opin. Neu-
robiol. 33, 156–165 (2015). http://dx.doi.org/10.1016/j.conb.2015.04.004
13. Cappellini, G., Ivanenko, Y.P., Poppele, R.E., Lacquaniti, F.: Motor patterns in
human walking and running. J. Neurophysiol. 95(6), 3426–3437 (2006). http://dx.
doi.org/10.1152/jn.00081.2006
14. Garate, V.R., Parri, A., Yan, T., Munih, M., Lova, R.M., Vitiello, N., Ronsse,
R.: Experimental validation of motor primitive-based control for leg exoskeletons
during continuous multi-locomotion tasks. Front. Neurorobot. 11, 15 (2017)
15. Ambrozic, L., Gorsic, M., Geeroms, J., Flynn, L., Molino Lova, R., Kamnik, R.,
Munih, M., Vitiello, N.: CYBERLEGs: a user-oriented robotic transfemoral pros-
thesis with whole-body awareness control. IEEE Robot. Autom. Mag. 21(4), 82–93
(2014)
16. Giovacchini, F., Vannetti, F., Fantozzi, M., Cempini, M., Cortese, M., Parri, A.,
Yan, T., Lefeber, D., Vitiello, N.: A light-weight active orthosis for hip movement
assistance. Robot. Autonom. Syst. 73, 123–134 (2015)
17. Yan, T., Parri, A., Fantozzi, M., Cortese, M., Muscolo, M., Cempini, M., Giovacchini,
F., Pasquini, G., Munih, M., Vitiello, N.: A novel adaptive oscillators-based control
for a powered multi-joint lower-limb orthosis. In: Proceedings of the IEEE Interna-
tional Conference on Rehabilitation Robotics (ICORR) (2015)
Design of a Hand Exoskeleton for Use
with Upper Limb Exoskeletons
Peter Walker Ferguson(B) , Brando Dimapasoc, Yang Shen, and Jacob Rosen
Bionics Lab at the University of California Los Angeles, Los Angeles, CA 90095, USA
PWFerguson@ucla.edu
Abstract. Due to high degree of freedom and different mechanism foci,

hand and arm exoskeletons are usually developed separately and sel-
dom combined together. Hand exoskeletons are typically more complex
mechanisms than arm or leg exoskeletons due to the numerous degrees
of freedom encapsulated in the hand and the small anatomical structure
involved. This study presents the design of a 12 DOF (6 active) recon-
figurable hand exoskeleton for rehabilitation that will be installed on
the upper limb exoskeletons, EXO-UL8 and BLUE SABINO. Given the
mechanism architecture, a nonlinear optimization framework minimizes
physical footprint while maximizing mechanism isotropy and device func-
tionality.
1 Introduction
For decades, stroke has been one of the leading diseases that causes long-
term disabilities [1]. Researchers and physical therapists have been working on
exoskeleton-like robots to help patients regain capabilities post stroke. The upper
limb is an area of focus, but research has focused mainly on individual parts,
either arm or hand [2–4]. Although there is a need for rehabilitating the capa-
bilities in reach and grasp activities of daily living, only a handful of systems
have a working combination of both arm and hand [5,6]. Even fewer combination
systems actively actuate multiple DOFs across multiple fingers [7]. Based on the
upper limb exoskeletons EXO-UL8 [8] and BLUE SABINO [9], a reconfigurable
hand exoskeleton is designed.
2 Methods
2.1 Design Requirements
The following requirements were formulated for a rehabilitation hand exoskeleton

that attaches to an arm exoskeleton:
This work was funded in part by the National Science Foundation through Award
#1532239.
P. W. Ferguson and B. Dimapasoc—These authors contributed equally to this work.
https://doi.org/10.1007/978-3-030-01887-0_53
Design of a Hand Exoskeleton for Use with Upper Limb Exoskeletons 277
(1) Low Mass: Mass at the hand must be minimized to reduce required torque of
the upper limb exoskeleton.
(2) Torque: The torque capabilities of the exoskeleton must be sufficiently large
to actuate the hand.
(3) Workspace: The workspace of the exoskeleton must contain the workspace of
the human hand.
(4) Grasp: It must be able to actuate a variety of grasps.
(5) Open Palm: It must leave the palm and fingers unoccupied to permit inter-
action with physical objects.
(6) Unisize: It must fit 95% of the general population.
2.2 Actuation Method
For the low mass and torque requirements, a Bowden cable transmission system
with brushed DC motors was chosen. The cable transmission enabled remote
location of the motor pack, reducing mass at the hand. It also allowed use of
over-sized actuators with sufficient torque for hand rehabilitation.
2.3 Basic Topology
Workspace, grasp, open palm, and unisize requirements are satisfied by a recon-
figurable design topology of three 3R planar serial linkages that attach on the
dorsal side of the hand to the distal phalanges. Three linkages are used as 95%
of human grasps are achievable with a thumb and two fingers [10]. The topology
allows a one-size-fits-all design that neither requires adjustment for different fin-
ger lengths nor impedes grasping physical objects. The third joint is made passive
to decrease complexity and inertia compared to an active joint. Due to the link
lengths, this joint mainly relates to orientation. A passive rotational joint added
at the end-effector of each finger linkage permits slight adduction/abduction to
improve comfort and allow more natural movement. Bending beam load cells are
used as the structure of the first link (L1 ) of the thumb linkage and second link
(L2 ) of the finger linkages, enabling admittance control.
Fig. 1. 1-1-3 Configuration shown with (A): Open Hand, (B): Closed Fist, (C):
Pointing, (D): Pincer Grasp.
278 P. W. Ferguson et al.
The linkages are reconfigurable to enable a variety of grasps. The first linkage
attaches from above the CPC joint to the distal phalanx of the thumb. The plane
of the workspace of this linkage is adjustable via rotation around the CPC. The
second and third linkages connect from above the MCP joints to the distal
phalanges of the fingers. The origin of these linkages is adjustable for different
hand widths or to place them in plane with different fingers. The distal end of
the second and third linkage feature interchangeable customizable 3D-printed
finger attachments that enable different sets of fingers to be actuated by each
linkage. Notable configurations include 1-1-3 (thumb, index, middle+ring+little)
and 1-2-2 (thumb, index+middle, ring+little). The 1-1-3 configuration is shown
for a set of representative hand positions in Fig. 1. To account for motion of the
little finger relative to the ring finger, a passive slider mechanism connects the
finger attachment for the little finger to the third linkage.
2.4 Link Length Optimization

To satisfy the unisize requirement, the link lengths were chosen via a brute force
optimization algorithm considering fingers in the 95th percentile for length.
The lengths of L1 of the thumb linkage and L2 of the finger linkages were
set to 8.9 cm due to the length of the bending beam load cells. For each linkage,
the remaining link lengths were varied across a reasonable range. Each combi-
nation, L, of potential link lengths L1 , L2 , and L3 , was checked for kinematic
feasibility. Forward and inverse kinematics were used to verify that the linkage
could correctly attach to the tip of the distal phalanx of the appropriate finger
at all combinations of joint angles (θ1 , θ2 , θ3 ) within the workspace with 3◦ res-
olution. To correctly attach, L3 must be capable of connecting perpendicularly
to the dorsal side of the distal phalanx, and the joints of the linkage must not
physically touch or cross through the finger.
A design score, J, was calculated for each L based on mechanism isotropy
and link length. Mechanism isotropy (ISO), a function of the joint angles is a
measure of kinematic performance. It is defined in (1) as the ratio of the min
(λmin ) and max (λmax ) eigenvalue of the Jacobian matrix.
λmin
ISO(θ1 , θ2 , θ3 ) = ∈ (0, 1) (1)
λmax
A value of 0 indicates singularity while a value of 1 means the end effector can
move equally well in all directions.
Mechanism isotropy is calculated for each set of joint angles previously men-
tioned. To account for varying densities of the end effector location in these
sets, the finger workspace area is discretized into a grid of cells, K, and the
isotropy is averaged for each cell. Summing the average isotropy of the cells pro-
vides an indication of the kinematic capabilities of the mechanism across the
entire workspace. It is desirable for the mechanism to avoid singular or near-
singular configuration within the workspace of the finger. Therefore, J of each
L is proportional to both overall performance (sum of ISO) and to worst-case
performance (minimum ISO value calculated).
Design of a Hand Exoskeleton for Use with Upper Limb Exoskeletons 279
As mechanical isotropy tends to reward longer link lengths, but it is desirable

to keep size and mass of the mechanism low, an additional term is included in J
score to reward shorter designs. This was accomplished by making J inversely
proportional to the sum of the link lengths raised to a hyperparameter A, as
shown in (2). A prototype with adjustable link lengths was used to experimen-
tally verify the design produced by a variety of A. Based on this verification,
link lengths were chosen for each linkage.
ΣK ISO(θ1 , θ2 , θ3 ) ∗ M INK (ISO(θ1 , θ2 , θ3 ))

J= (2)
(L1 + L2 + L3 )A
The results of the optimization are illustrated for the linkage that connects
to the index finger in Fig. 2.
Fig. 2. Optimization results for the linkage connecting to the index finger for A = 5.
Dots represent kinematically valid combinations of L1 and L3 for L2 = 8.9 cm. Set
L1 ≥ 4.4 cm due to minimum axes size.
3 Conclusion
The hand exoskeleton presented is multi-fingered, multi-DOF, reconfigurable,

and designed to attach to a full-arm exoskeleton. The link lengths were deter-
mined by optimization to maximize mechanism isotropy and minimize footprint.
280 P. W. Ferguson et al.
References
1. Benjamin, E.J., et al.: Heart disease and stroke statistics–2017 update: a report
from the American heart association. Circulation 135(10), e146 (2017)
2. Wege, A., Zimmermann, A.: Electromyography sensor based control for a hand
exoskeleton. In: 2007 IEEE International Conference on Robotics and Biomimetics
(ROBIO), pp. 1470–1475, December 2007
3. Ho, N.S.K., et al.: An EMG-driven exoskeleton hand robotic training device on
chronic stroke subjects: task training system for stroke rehabilitation. In: 2011
IEEE International Conference on Rehabilitation Robotics, pp. 1–5, June 2011
4. Schabowsky, C.N., Godfrey, S.B., Holley, R.J., Lum, P.S.: Development and pilot
testing of hexorr: hand exoskeleton rehabilitation robot. J. NeuroEngineering Reha-
bil. 7(1), 36 (2010)
5. Ren, Y., Park, H.S., Zhang, L.Q.: Developing a whole-arm exoskeleton robot with
hand opening and closing mechanism for upper limb stroke rehabilitation. In: 2009
IEEE International Conference on Rehabilitation Robotics, pp. 761–765, June 2009
6. Frisoli, A., et al.: A new force-feedback arm exoskeleton for haptic interaction
in virtual environments. In: First Joint Eurohaptics Conference and Symposium
on Haptic Interfaces for Virtual Environment and Teleoperator Systems. World
Haptics Conference, pp. 195–201, March 2005
7. Lauretti, C., et al.: Learning by demonstration for motion planning of upper-limb
exoskeletons. Front. Neurorobotics 12, 5 (2018)
8. Shen, Y., Ma, J., Dobkin, B., Rosen, J.: Asymmetric dual arm approach for post
stroke recovery of motor function utilizing the EXO-UL8 exoskeleton system: a
pilot study. In: Engineering in Medicine and Biology Society (EMBC), June 2018
9. Perry, J.C., Maura, R., Bitikofer, C.K., Wolbrecht, E.T.: Blue sabino: develop-
ment of a bilateral exoskeleton instrument for comprehensive upper-extremity neu-
romuscular assessment. In: 2018 International Conference on Neurorehabilitation
(ICNR), October 2018, to be published
10. Kapandji, I., Honoré, L., Poilleux, F.: Upper Limb, ser. Physiology of the joints:
annotated diagrams of the mechanics of the human joints/I.A. Kapandji. L.H.
Honoré Übers. Churchill Livingstone (1982)
A Real-time Graphic Interface for the
Monitoring of the Human Joint
Overloadings with Application
to Assistive Exoskeletons
Marta Lorenzini1,2(B) , Wansoo Kim1 , Elena De Momi2 , and Arash Ajoudani1

1
HRI2 Lab, Department of Advanced Robotics, Istituto Italiano di Tecnologia,
Genoa, Italy
{marta.lorenzini,wan-soo.kim}@iit.it
2
Department of Electronics, Information and Bioengineering, Politecnico di Milano,
Milano, Italy
Abstract. This work presents an intuitive graphic interface to make its

users aware of potentially risky body configurations while being exposed
to external loads. Employing an algorithm we proposed in a recent work,
we estimate the human joint torque overloading caused by an external
force. This information is used as an input for the graphical interface to
provide the user with an intuitive feedback about the strain on each joint.
Hence, the users can be aware of the loading states, react to them accord-
ingly, and minimise the risk of injuries or chronic pain. This graphical
interface can help the users learn and achieve more ergonomic configu-
rations during industrial job duties.
1 Introduction
Musculoskeletal disorders (MSD) represents the single largest category of work-
related diseases in several industrial countries [1], and impose extremely large
costs each year in lost productivity and absence due to sickness. One of the
major contributors to MSDs in every industry division is performing repetitive or
heavy manufacturing tasks (e.g. lifting, pushing, or pulling on objects), resulted
by an excessive and continuous mechanical overloading of body joints [2]. Within
this context, an intelligent framework which ensures workers’ well-being in the
execution of their daily work is highly needed. To encourage the workers not to
assume improper body postures and minimise the risk of work-related injuries,
two keys factors must be considered: a method to the real-time monitoring of
excessive physical loading on body joints, and intuitive feedback and guidance
interfaces to drive them to more ergonomic postures and working conditions.
Extensive literature exists on the assessment of human physical loads asso-
ciated with lifting or carrying heavy objects in different postures [3], but only
off-line procedures are proposed. Alternative research works develop accurate
biomechanical models to evaluate the human dynamic behaviour. However, these
https://doi.org/10.1007/978-3-030-01887-0_54
282 M. Lorenzini et al.
models may be defined relatively precise with the identification of a large num-
ber of parameters [4] or differently, they can be obtained with anthropomet-
ric tables, hence, they are not subject-specific and can introduce a large level
of uncertainty. These conditions can limit the real-time applicability of such
approaches in industrial use-cases. On the other hand, regarding the feedback
technologies, different modalities (e.g. visual, auditory and haptic) have been
proposed to improve workers’ risk-awareness, but the feedback provided to the
users only contains kinematic information [5].
Accordingly, the main objective of this work is to present an intuitive graphic
interface which makes humans aware of whole-body kinematic and dynamic
information. The dynamic information is expressed in terms of overloadings on
the joints caused by an external force, that are estimated by means of a reduced-
complexity approach we recently proposed in [6]. Such a feedback system can
be combined with assistive technologies like industrial exoskeletons to serve as
a support tool in reducing physical loadings while performing heavy tasks. In
particular, when using lower-body or upper-body exoskeletons, the loading con-
dition of the rest of the joints can be evaluated by using the proposed interface.
2 Monitoring Method
We recently proposed an algorithm for the real-time estimation of human over-
loading joint torques1 [6], defined as Δτi , where i denote the i-th joint. Our
method is based on the estimation of the translational displacement of the whole
body Centre of Pressure (CoP) in the presence of external forces, calculated from
the difference between an estimated one (using an off-line calibrated model) and
the measured one (using wearable sensors). If no external interactions of the
human with the environment (or object) are in place, the estimation of the CoP
vector ĈPwo achieved by the human body model, which consider no external
load except the body weight, is similar to the measured one CPwt . If an exter-
nal force is applied to the human, the estimated and the measured CoP vectors
differ. Accordingly, the overloading joint torque vector is estimated using the
contact Jacobians of this CoP displacement along with the difference between
the measured vertical ground reaction force (vGRF), which takes into account
the effect of external forces, and the estimated one (simply equal to the body
weight). A reduce-complexity human model is developed to ensure the real-time
applicability of the method and to take specifically into account the body joints
which are mainly at risk of injuries. The needed measurements on the human are
collected by means of a wearable motion-capture system and foot insole sensors.
A complete and detailed explanation of the method can be found in [6].
3 Visual Feedback Interface

The method we briefly present in Sect. 2 enables the on-line estimation of human
joint overloadings due to external forces. To make the human aware of this
1
The overloading joint torque refers to the torque induced into the human joint by
an external load.
A Real-time Graphic Interface for the Monitoring of the Human Joint 283
Fig. 1. An overview of the graphic interface: in each row the human is represented
in three different body configurations, without any tool in the first row and with two
different tools in the second and third row, respectively. The level of overloading joint
torque is color-coded to denote a high, medium or low value, and illustrated in the
main joints of the human body.
meaningful information along with his/her current body configuration we take

advantage of a graphic interface, shown to the users by means of a dedicated
screen. The ROS 3D visualizer RViz is employed for the displaying purpose.
Figure 1 illustrates an example of the information that can be provided in real-
time to the human: the current body configuration, the level of the computed
overloading joint torques, which are color-coded to denote a high (red), medium
(orange) or low (green) value and finally, the positions of the measured and the
estimated CoP. The three levels of overloading are determined as explained in
Table 1. Different tools/objects are represented with a different color/shape in
the graphic interface. Depending on their weight, the external force experienced
by the human changes and thus the parameters of the overloading joint torques
estimation technique can be tuned accordingly. In each row of Fig. 1, three dif-
ferent body configurations are represented, passing from a very risky condition,
denoted by high values of overloadings on the joints, to a safe and comfortable
one in which the physical effort is minimised, as suggested by the lower values of
284 M. Lorenzini et al.
joints overloadings. Nevertheless, the first row shows the human operating with-
out any tool and thus without the effect of any external forces, therefore, the
overloading on the joints is always low. The second and the third row, instead,
depict the human operating with two different types of tool, respectively.
Considering that industrial exoskeletons in most cases provide power or sup-
port only to the lower limbs or to the upper extremities and in some cases even
just to one single-joint, this visual feedback can take into account the body parts
which are not covered by the exoskeleton. As long as the human is assisted by the
exoskeleton, he/she can view the information provided by the graphic interface,
displayed on a screen, and monitor the level of the overloading torques on all
his/her own joints. As a result, the human is supported and guided towards more
ergonomic and comfortable whole-body configurations by means of a combined
contribution of the exoskeleton with the visual feedback.
Table 1. Stepwise scheme for joint torque overloading level
Overloading level Control threshold

Green 0< Δτi ≤ 0.3τmaxi
Orange 0.3τmaxi < Δτi ≤ 0.6τmaxi
Red 0.6τmaxi < Δτi ≤ τmaxi
4 Conclusion
In this work we presented a graphic interface to provide human workers with a
visual feedback about the overloadings on their joints while performing a heavy
task with an external tool or object. This system can be integrated to assistive
technologies such as industrial exoskeletons, making the human aware of the
physical loadings exerted on all his/her joints, along with meaningful information
regarding the current body configuration, CoP and the object/tool properties.
Through this synergistic combination, the workers are assisted and guided while
performing heavy or repetitive tasks and the risk of injuries or chronic pain can be
reduced. In future works, the proposed visual feedback interface will be combined
with robotic assistive systems or considered for rehabilitation purposes.
References
1. Bevan, S.: Fit for work? Musculoskeletal disorders in the European workforce. The
Work Foundation (2009)
2. Kumar, S.: Theories of musculoskeletal injury causation. Ergonomics 44(1), 17–47
(2001)
3. Van Den Bogert, A.J.: Analysis and simulation of mechanical loads on the human
musculoskeletal system: a methodological overview. Exerc. Sport. Sci. Rev. 22, 23–
23 (1994)
A Real-time Graphic Interface for the Monitoring of the Human Joint 285
4. Ayusawa, K.: Identifiability and identification of inertial parameters using the under-
actuated base-link dynamics for legged multibody systems. Int. J. Robot. Res. 33(3),
446–468 (2013)
5. Azuma, R.: Recent advances in augmented reality. IEEE Comput. Graph. Appl.
21(6), 34–47 (2001)
6. Kim, W.: A real-time and reduced-complexity approach to the detection and moni-
toring of static joint overloading in humans. In: International Conference on Reha-
bilitation Robotics (ICORR), pp. 828–834. IEEE (2017)
Smart Human-Machine Systems for
Lower-Limb Assistance and
Rehabilitation After Paralysis
Study of Algorithms Classifiers for an Offline
BMI Based on Motor Imagery of Pedaling
Mario Ortiz(&), Marisol Rodríguez-Ugarte, Eduardo Iáñez,

and José M. Azorín
Brain-Machine Interface Systems Lab, Miguel Hernández University of Elche,

Elche, Spain
mortiz@umh.es
Abstract. The paper compares different signal processing algorithms and

classifiers to evaluate the accuracy of a BMI based on lower-limb motor ima-
gery. The methods were based on the analysis of the peaks of the different
processing epochs for the alpha, beta and gamma EEG bands through the
Marginal Hilbert Spectrum, Power Spectral Density and Fourier harmonic
components. Data were classified and analyzed by three classifiers: Support
Vector Machine, Self-Organizing Maps and Linear Discriminator analysis.
Results show accuracy is dependent on the subject, but there is not dependency
between the subjects and the methods, and classifiers. Best accuracy results were
achieved by using the value of the peak of the Hilbert Marginal Spectrum,
obtaining the analytical signal with the Stockwell transform. Regarding the
classifiers SOM presented lower accuracy values than SVM and LDA.
1 Introduction
Patients that have suffered a stroke or traumatic brain injuries can have their motion
capability reduced. The use of motion assistant devices controlled by a brain-machine
interface (BMI) can improve the rehabilitation process through the cognitive
involvement of the subject [1].
One of the most used BMI control approach involves the use of electroen-
cephalographic (EEG) signals due to its non-invasive nature. A BMI collects the EEG
signals of the subject’s brain through several electrodes. The channel information is
then analyzed by a computer in order to extract the representative features of the waves.
Once the classification model is created, it can be used to identify similar patterns in the
real-time performance, controlling the assistant device depending on the identified
mental will.
This research has been carried out in the framework of the project Associate - Decoding and
stimulation of motor and sensory brain activity to support long term potentiation through Hebbian
and paired associative stimulation during rehabilitation of gait (DPI2014-58431-C4-2-R), funded
by the Spanish Ministry of Economy and Competitiveness and by the European Union through
the European Regional Development Fund (ERDF) A way to build Europe.

https://doi.org/10.1007/978-3-030-01887-0_55
290 M. Ortiz et al.
Motor imagery (MI) is one of the common used control methods. It consists of the
mental imagination of the movement action without real movement. Literature indi-
cates that there is an event related (de)synchronization (ERD/ERS) [2] mainly at the mu
frequency band (10–14 Hz), as a part of alpha band (8–13 Hz), consisting of a fluc-
tuation of the power due to the MI actions in comparison to the relaxed state. This is
more noticed on the zone around the CZ electrode based on the 10/10 international
system.
In this paper, besides the previously mentioned alpha band (8–13 Hz), beta
(13–32 Hz) and gamma bands (32–50 Hz) [3] were also considered in order to assess
the attention focus [4] of the subject, and to improve the results of the MI identification.
In addition, three different processing algorithms and classifiers were studied.
2.1 Subjects
Three healthy subjects (1 male and 2 women) participated voluntary in the experi-
mental sessions. The users were previously informed about the procedure and signed an
informed consent according to the Helsinki declaration. The whole experimental pro-
cedure was approved by the ethics committee of the University.
2.2 Equipment
EEG acquisition was performed by a Starstim 32 cap of Neuroelectrics™. The sam-
pling frequency was 500 Hz and the data were transmitted by wire to the computer
where they were analyzed with the help of the custom algorithms developed in Mat-
lab™. Although, all the electrodes were used for spatial filtering, only the electrodes
around the sensory-motor cortex zone were considered for the data characterization:
CZ, CP1, CP2, C1, C2, C3, C4, FC1 and FC2.

Every subject performed 5 trials. Each of them consisted of 15 consecutive events of
relax/imagination, while the subject was comfortably seated. Each event lasted around
5 s and required the subject not to move or to blink. During the event, the subjects had
to focus on the mental action of pedaling during imagination periods or to leave the
mind in a blanked state during the relaxed periods. Between events, a little time of 2 s
was allowed to the subject in order to blink or to move during the trial. Event infor-
mation was communicated to the subject by a screen interface during the whole trial.
Study of Algorithms and Classifiers for an Offline BMI 291
2.4 Signal Processing

EEG signals were processed in 1 s epochs shifted every 0.2 s. Data were preprocessed
by a high pass filter (0.05 Hz), a low pass filter (45 Hz) and a Laplacian spatial filter.
For the pattern characterization, three methods were considered to extract one
feature per alpha, beta and gamma bands:
– ST: Value of the peak of the Hilbert Marginal Spectrum [5], obtaining the analytical
signal with the Stockwell transform [6]
– PSD: Value of the peak of the power spectral density by Welch’s method
– H: Value of the amplitude of the most representative harmonics per band obtained
by Fast Fourier Transform.
Accuracy evaluation was done by leave-one-out cross-validation. Three different
classifiers were also considered: Support Vector Machine (SVM) [7], Linear Dis-
criminator Analysis (LDA) [8] and Self-Organizing Maps (SOM) [9]. Accuracy rep-
resents in percentage the number of events correctly detected.
3 Results
Table 1 shows the accuracy performance of the classifiers. While SVM and LDA
present a similar result, SOM shows a lower accuracy. Table 2 shows the results by
subject, classifier and type of event neglecting SOM classifier, due to its lower per-
formance, to limit the volume of data.
Table 1. Accuracy performance of the classifiers.

Classifier Accuracy (% ± standard deviation)
SVM 74.5 ± 15.0
LDA 75.1 ± 14.5
SOM 63.8 ± 11.1
Data is shown averaged for the 3 subjects and
relax and imagine events.
Statistical Manova analysis by SPSS revealed that accuracy had a high dependency
on the subject’s expertise (p < 0.001). The test of within-subjects effects indicated that
there was not a dependency on the subjects regarding the methods (p > 0.05), or the
classifier (p > 0.05). However, the performance of the type of event (relax/imagination)
showed a clear dependency on the subject (p < 0.001).
292 M. Ortiz et al.
Table 2. Accuracy results for the 3 subjects, best 2 classifiers and relax and imagine events.
Subject Method Classifier Event Accuracy (% ± standard deviation)
S1 ST SVM Relax 93.9 – 2.7
Imagine 96.6 ± 1.4
LDA Relax 88.2 ± 3.3
Imagine 98.9 – 0.6
PSD SVM Relax 86.0 ± 4.6
Imagine 86.5 ± 3.4
LDA Relax 78.3 ± 11.0
Imagine 94.6 ± 3.4
H SVM Relax 88.0 ± 4.7
Imagine 88.2 ± 3.6
LDA Relax 78.3 ± 11.0
Imagine 95.3 ± 2.5
Imagine 60.4 ± 6.0
Imagine 68.1 – 6.5
Imagine 57.6 ± 7.4
Imagine 66.2 ± 8.1
H SVM Relax 68.7 ± 7.5
Imagine 54.2 ± 5.5
Imagine 66.3 ± 6.5
Imagine 74.4 ± 15.8
LDA Relax 74.4 ± 10.9
Imagine 79.5 – 12.2
Imagine 74.5 ± 17.9
LDA Relax 73.4 ± 14.1
Imagine 77.2 ± 18.9
H SVM Relax 76.4 ± 10.2
Imagine 75.5 ± 14.7
LDA Relax 74.4 ± 15.8
Imagine 79.0 ± 17.4
Best results per subject are shown in bold text
Study of Algorithms and Classifiers for an Offline BMI 293
4 Discussion
Results indicated that SOM performance was lower than SVM or LDA classifiers.
Time of computing was also higher. The SOM performance could be improved by an
algorithm tuning and number of trainings. However, the simplicity and faster com-
puting of SVM or LDA makes them more suited for a BMI. BMI performance is very
dependent on the subject’s expertise as S1 results indicate. ST obtained the best results
for all the subjects with a better result of SVM during relax and LDA during imagine.
However, as in an online application, the type of event is not known at first hand, it is
not possible to use both.
5 Conclusion
The paper has introduced several MI algorithms and classifiers. A combination of ST

+SVM or LDA would be the best for a future real-time application.
References
1. Gharabaghi, A.: What turns assistive into restorative brain-machine interfaces? Front.
Neurosci. 10, 456 (2016)
2. Pfurtscheller, G., Brunner, C., Schlögl, A., Lopes da Silva, F.H.: Mu rhythm (de)
synchronization and EEG single-trial classification of different motor imagery tasks.
Neuroimage 31(1), 153–159 (2006)
3. Rao, R.P.N.: Brain-Computer Interfacing: An Introduction. Cambridge University Press,
Cambridge (2013)
4. Costa, Á., et al.: Attention level measurement during exoskeleton rehabilitation through a
BMI system. In: González-Vargas, J., Ibáñez, J., Contreras-Vidal, J.L., van der Kooij, H.,
Pons, J.L. (eds.) Wearable Robotics: Challenges and Trends, vol. 16, pp. 243–247. Springer,
Cham (2016)
5. Huang, N.E., et al.: The empirical mode decomposition and the Hilbert spectrum for nonlinear
and non-stationary time series analysis. Proc. R. Soc. London. Ser. A Math. Phys. Eng. Sci.
454(1971), 903LP–995 (1998)
6. Stockwell, R.G., Mansinha, L., Lowe, R.P.: Localization of the complex spectrum: the S
transform. IEEE Trans. Signal Process. 44(4), 998–1001 (1996)
7. Steinwart, I., Christmann, A.: Support Vector Machines. Springer, New York (2008)
8. Izenman, A.J.: Linear Discriminant Analysis, pp. 237–280. Springer, New York (2013)
9. Kohonen, T.: The self-organizing map. Proc. IEEE 78(9), 1464–1480 (1990)
Exoskeleton Control for Post-stoke Gait
Training of a Paretic Limb
Based on Extraction of the Contralateral
Gait Phase
Gabriel Aguirre-Ollinger, Ashwin Narayan, Hsiao-Ju Cheng,

and Haoyong Yu(B)
Department of Biomedical Engineering, National University of Singapore,

Singapore, Singapore
bieyhy@nus.edu.sg
Abstract. We developed a lightweight lower-limb exoskeleton to assist

the paretic leg of stroke patients during gait training. The device features
compliant actuators separated from the patient’s limb, thus avoiding any
gait disruption caused by the actuators’ inertia. The exoskeleton control
uses motion data from the healthy leg to extract a reference gait phase.
In this context, phase is a continuous variable that tracks the progress
of the gait over one cycle and wraps around at the end of the cycle. The
extracted phase information is used to time the assistive torque acting
on the impaired leg. Control of the assistive torque is implemented as
a force control acting on a time-varying linear system representing the
actuator and exoskeleton. Results from one experiment show how the
exoskeleton helps improve knee flexion during the swing phase of the
gait cycle.
1 Introduction
Stroke results in muscle stiffness which reduces the paretic limb’s hip flexion,
knee flexion and ankle dorsiflexion. Insufficient flexion of the joints compromises
toe clearance during the swing phase of walking. To facilitate toe clearance
post stroke, patients resort to compensatory movements such as hip hiking and
circumduction which result in asymmetric gait. We present here a method for
controlling a lower-limb exoskeleton to assist the paretic leg during post-stroke
gait training, with the aim of compensating deficiencies in joint flexion. The
control concept, based on tracking the contralateral (healthy side) gait phase,
has been previously introduced in [1].
This project is supported by the FRC Tier 1 grant with WBS: R-397-000-302-114
from National University of Singapore.
https://doi.org/10.1007/978-3-030-01887-0_56
Exoskeleton Control for Post-stoke Gait Training 295

2.1 Lower-Limb Exoskeleton and Actuators
We developed a lightweight unilateral exoskeleton to aid the gait training of
stroke patients. The device provides torque to assist the flexion and extension of
the knee and ankle joints of the impaired leg. Users can walk either on a treadmill
or over ground. Our exoskeleton features a detached, dual actuator package that
connects to the joints via Bowden cables (Fig. 1(a)). Thus our system avoids any
potential disruption of the gait cycle due to the actuators’ inertia. The actuator
package features a novel series elastic actuator (SEA) design. The SEA transmits
force to a Bowden cable via two springs mounted in series (Fig. 1(b)). In turn,
the Bowden cable exerts torque on the lower-limb joint. Torque is controlled by
regulating the amount of spring deflection.
Fig. 1. Lower-limb wearable exoskeleton. (a) Complete assembly: dual series-elastic

actuator package, Bowden cables and lightweight leg brace. (b) Dual-spring series elas-
tic actuator: mechanism (top) and linear time-varying model (bottom).
2.2 Gait Phase Extraction

Our exoskeleton control employs an inertial measurement unit (IMU) to measure
the instantaneous orientation of the healthy leg, specifically the angle of the thigh
on the sagittal angle, identified as θh . This angle provides the feedback required
to compute the assistive torques which are applied to the paretic leg. The IMU
extracts orientation as a quaternion, the value of which is relayed wirelessly to
the exoskeleton control. θh is calculated by converting the quaternion to Euler
angles.
The control uses θh (t) to extract the phase of the gait cycle of the healthy
leg [2]. Phase in this context is a continuous variable φ(t) that increases mono-
tonically over time and wraps around modulo 2π every time the user completes
296 G. Aguirre-Ollinger et al.
one gait cycle. The timing of the assistive torques is indexed to the theoretical
phase of the paretic leg, given by φ (t) = φ(t) + π. A virtual dynamical system
driven by an adaptive frequency oscillator (henceforth simply the “oscillator”)
extracts the phase and frequency of the measured thigh angle in real time. The
dynamical system is given by
φ̇ = ω − e(t) sin φ, ω̇ = − e(t) sin φ (1)

α̇k = η cos(kφ) e(t), β̇k = η sin(kφ) e(t) (2)
Nf

θrec = αk cos(kφ) + βk sin(kφ) (3)
k=0
where φ is the oscillator’s phase, ω the oscillator’s intrinsic frequency and is the
coupling strength. The oscillator input is the error signal e(t) = θh,m (t)−θrec (t),
where θh,m (t) is the measured thigh angle of the healthy leg, and θrec (t) is the
reconstructed angle, given the a finite-term Fourier decomposition (3), which is
performed on-line.
2.3 Generating the Assistive Torque
Using φ (t) to time the assistive torque has two advantages: φ (t) is linked to
the healthy leg and is therefore closer to a normal gait cycle; since θh is a
relatively smooth function, it is easier for the oscillator to synchronize with
it. We use the example of generating an exoskeleton torque to assist flexion
of the impaired knee during the swing phase. Our selected torque profile is
a stretched bell shape defined by hyperbolic tangents (Fig. 2(a)). The desired
assistive torque is converted to a reference force profile Fref (t) to be delivered
by the SEA. Figure 1(b) shows an equivalent linear motion model of the SEA and
the exoskeleton components that assist a single joint (the knee in our example).
The soft translational spring k23 handles the low force range with high tracking
fidelity; the high-stiffness spring k12 handles the high force range when k23 is
fully compressed, thereby increasing the control bandwidth.
The force controller for tracking of Fref (t) represents the SEA-exoskeleton
system as a time-varying system in order to account for the constraint imposed
by the periodic full compression of k23 . Given the system’s positions x =
[x1 , x2 , x3 ]T and velocities v = [v1 , v2 , v3 ]T , the unconstrained system is v̇ =
Fx x + Fv v + Gu. Full compression of k23 is represented by the constraint
Av = 0 where A = [0 1 −1]. The constrained SEA-exoskeleton is mod-
eled using the Udwadia-Kalaba formulation [3]. Given the system’s mass matrix
M, we define the constraint term D(h) = I − hC, where h is a switching vari-
able equal to 1 when the compression constraint is active and to 0 when it is
inactive, and C = M−1 AT (AM−1 AT )−1 . The constrained dynamical system is
ẇ = Fw (h)w + Gw (h)u, where

0 I 0
Fw = , Gw = (4)
D(h)Fx D(h)Fv D(h)G
Exoskeleton Control for Post-stoke Gait Training 297
and w = [xT , vT ]T . Control of the switched system (4) is performed in discrete

time using a forward-propagating Riccati equation.
3 Results
The target patient group for our exoskeleton system is patients with a Functional
Ambulatory Category (FAC) score between 2 and 3, and a time after stroke of
6 to 24 months. In order for the oscillator-based control to function, kinematic
data of the healthy leg must reveal an identifiable step frequency. As part of
an ongoing study (approval: Domain Specific Review Board of the National
Healthcare Group, Singapore), a stroke patient (male, age 49, height 1.68 m,
body mass 75.6 kg) walked along a straight path, first unassisted, then with the
exoskeleton providing assistance only to the knee joint. In Fig. 2(b), the top
(a) (b) θ(t), rad (healthy) f

AFO
(t), Hz φ(t) (scaled)
1
Δφrise Δφwid
0.5
0
Fref(t), N (paretic) Fest(d), N (paretic) φ(t) (scaled)
300
φc 200
0 1
100
φ’
0
0 1 2 3 4 5 6 t (sec)
Fig. 2. Generating the assistive torque. (a) Reference torque indexed to theoretical
phase φ of the impaired leg. Timing of the torque is controlled by the midpoint phase
value φc , its duration by Δφrise and Δφwid . (b) Top: healthy thigh angle measured
from IMU data. Superimposed are the phase φ(t) and frequency fosc (t) = ω(t)/(2π)
of the thigh angle as computed by the oscillator-based dynamical system. Bottom:
reference assistive force Fref (t) and actual assistive force Fest (t) acting on the knee
joint.
Fig. 3. Ensemble averages of the knee angle for several steps (mean ± std. dev.) in two
conditions: walking freely, and walking with exoskeleton assistance.
298 G. Aguirre-Ollinger et al.
plot shows the phase and frequency of the healthy leg’s motion, computed form
the measured thigh angle. The bottom plot shows the generated assistive force,
which acted during the swing phase of the paretic leg’s motion. Figure 3 shows
the ensemble averages of the knee angle for the two different conditions.
4 Discussion
The experimental data in Fig. 3 shows how the patient achieved consistently
larger knee flexion when assisted by the exoskeleton, which was the primary
objective of the experiment. There was also a noticeable reduction in the vari-
ability of the knee joint trajectory, as evidenced by the standard deviation values;
the primary cause is probably an indirect increase in joint stiffness caused by the
actuator’s springs. Our current research aims to determine whether exoskeleton
assistance contributes to an improvement in gait symmetry, as well as retention
of the improved gait function.
References
1. Chen, G., Qi, P., Guo, Z., Yu, H.: Gait-event-based synchronization method for gait
rehabilitation robots via a bioinspired adaptive oscillator. IEEE Trans. Biomed.
Eng. 64(6), 1345–1356 (2017)
2. Aguirre-Ollinger, G.: Exoskeleton control for lower-extremity assistance based on
adaptive frequency oscillators: adaptation of muscle activation and movement fre-
quency. Proc. Inst. Mech. Eng. Part H 229(1), 52–68 (2015)
3. Kalaba, R.E., Udwadia, F.E.: Equations of motion for nonholonomic, constrained
dynamical systems via Gauss’s principle. J. Appl. Mech. 60(3), 662–668 (1993)
Design of a Passive Exoskeleton to Support
Sit-to-Stand Movement: A 2D Model
for the Dynamic Analysis of Motion
Luís P. Quinto1,2(&), Sérgio B. Gonçalves2, and Miguel T. Silva2

1
CINAMIL of Academia Militar of the Instituto Universitário Militar, Lisbon,
Portugal
luis.quinto@academiamilitar.pt
2
LAETA, IDMEC of Instituto Superior Técnico of the Universidade de Lisboa,
Lisbon, Portugal
{sergiogoncalves,miguelsilva}@tecnico.ulisboa.pt
Abstract. A significant number of people suffer from musculoskeletal

pathologies, which result in limitations in sit-to-stand (STS) movement or
during locomotion. Allowing disabled people to stand, can reduce secondary
conditions, increasing their life expectancy and reducing healthcare costs.
Exoskeletons can be used to support human motion, helping to solve these
problems.
This work regards a preliminary study to develop a passive exoskeleton to
support sit-to-stand movement. For that purpose, a biomechanical model was
implemented in a computational multibody dynamics software, to estimate
reaction forces and moments at the joints.
Data concerning STS movement with arm support and STS without arm
support was collected. Outcomes include reaction forces and moments calcu-
lated at the ankle, knee and hip joints, giving insights about the torque and
power requirements for the exoskeleton design.
Preliminary studies revealed that 10% of the force required to perform the
standing motion can be granted through the user’s arms action force.
1 Introduction
Every year between 250.000 to 500.000 people suffer a spinal cord injury (SCI), caused
by traumatic events and due to degeneration or diseases. Most trauma causes are
preventable, such as violence, falls or road traffic accidents. People with SCI are two to
five times more likely to die prematurely [1]. Cerebrovascular accidents (stroke) are the
third cause of disability, leading in many cases to impairment or loss of the ability to
stand or walk. Moreover, soldiers in the battlefield, as do workers in industry are asked
to do more, to carry more equipment or accomplish their role faster, with an inherent
increase of the probability of lesion/accident. Population aging is also an important
Authors would like to thank the Portuguese Army and LAETA for supporting this investigation.

https://doi.org/10.1007/978-3-030-01887-0_57
300 L. P. Quinto et al.
factor for mobility loss, being stated as one of the major global challenges for mankind.
All the cited examples result in substantial costs to social and health care structures
worldwide.
Mobility is one of most important human capacities, but to do so, first the person
needs to stand. There are significant similarities between sit-to-stand and heavy object
lifting, suggesting similar approaches for both problems.
Exoskeletons can be used in some cases to enhance its user’s capacity to perform
certain tasks, augmenting their overall performance and minimizing risks. In some
cases they can also restore mobility to its user, reducing secondary conditions like
musculoskeletal pathologies, obesity, cardiac, respiratory and urinary conditions, and
depression and neuropathic pain [2].
Recent developments in this field led to solutions that lack portability due to its
actuated joints and rigid bulky structure [3, 4]. Results point out the need for lighter
solutions, easy to use and with significant increase of energy autonomy.
Passive solutions may, in some cases, be the answer for a simpler but efficient
result. Complacent applications can also lead to interesting results, due to its light-
weight and adaptability.
Based on literature it is reasonable to suggest that STS movement has a major
influence in daily life activities and is a key aspect for a person to be independent. The
selection of the seat, namely its height and the use of armrests, influences the subject’s
effort performing STS.
This work is a preliminary study to develop a passive exoskeleton to support sit-to-
stand movement, aiming also to support human locomotion, decreasing users’ meta-
bolic costs. The study was developed based on the assumption that the user can
maintain balance, either by himself of using specific equipment (ex. crutches) and is
able to use arm support while performing STS. Using a conservative approach,
experimental apparatus considered a chair without armrests, so that the subject uses the
seats platform for arm support.
To design and develop this solution, a study of its kinematics and dynamics is
required, so that reaction forces and moments at joints can be estimated. Kinematic and
kinetic data were acquired in a biomechanics laboratory, using force plates and a
motion analysis system, as suggested in literature [5], to access moment joint forces
and reaction forces in each body segment, to estimate actuation forces at the hip, knee
and ankle.

Experimental data was acquired in the Lisbon Biomechanics Laboratory, using a 3D
motion system (MOCAP) with 14 infrared cameras Qualisys ProReflex MCU
500/1000. Ground reaction forces and forces applied in the chair, were acquired using 2
force plates AMTI OR 6-7-1000 (AMTI).
One volunteer [male, 34 years old, 73 kg and 1,80 m] was selected as control
subject, with no musculoskeletal pathologies, or previous fractures or low back pain.
Design of a Passive Exoskeleton to Support Sit-to-Stand Movement 301
A marker set composed by 47 reflective passive markers was used to collect

kinematic data at 100 Hz, following recommendations presented in [6].
Two different sets of experiments were conducted: (i) STS with arm support;
(ii) STS without arm support. The procedure was performed with two different appa-
ratus for each case, so that data acquisition with the force plates could register different
parameters, namely: (i) subject with both feet on two different force plates, registering
ground reaction forces for each foot; (ii) supporting the chair on one force plate and
subject’s both feet on a different force plate, registering ground reaction forces for the
chair and both feet. This apparatus enabled the caption of forces regarding the chair and
the subject’s both feet, using only two force plates, with minimum deviation from ideal
apparatus, using three force plates, one for each ground support.
2.2 Biomechanical Model

STS analysis was conducted based on a simplified model regarding motion in the
sagittal plane. The biomechanical model used in the inverse dynamics analysis was
formed by 14 segments, according to Fig. 1.
Fig. 1. Biomechanical model used in the inverse dynamics analysis (different segment lengths
used for graphical representation only)
Each segment was considered as a rigid body, constrained by ideal joints. The
developed model was implemented in a multibody dynamics software with natural
coordinates, APOLLO [7], developed in Instituto Superior Técnico. The computational
apparatus focused on obtaining joint moment forces on the hip, knee and ankle, so that
the exoskeleton actuators could be dimensioned.
302 L. P. Quinto et al.
Experimental apparatus regard two different objectives: (i) confirming subjects balance
during STS motion; (ii) analyzing STS motion;
Regarding balance, results show similar ground reaction forces during data
acquisition for both feet.
Analyzing STS, data was compared separately, for sit motion, with and without
support, and for stand motion also with and without support.
For sitting motion, results show a minimum deviation of ground reaction forces,
suggesting arm support while sitting has no significant influence.
During stand motion, three different effects can be identified: maximum ground
reaction force registered for both feet has a 10% reduction; reduction in force transition
on both force plates, using arm support; when performing motion with arm support,
ground reaction forces inflect during tight lift off, in both force plates.
Maximum ground force reduction on both feet reaction forces during motion with
arm support can be associated to a smoother transition between both supports, due to
arm action. Reaction force inflection during tight lift off can also be associated to the
arm support motion. Comparing results with and without arm support for the same
motion and phase, enables reaction force related to arm support to be estimated. Pre-
liminary results revealed that 10% of the force required to perform the standing motion
can be granted through the user’s arms action force.
4 Conclusions
A preliminary study on STS movement, with and without arm support was conducted,
so torque and power requirements for a passive exoskeleton design could be accessed.
Results show that 10% of the effort while performing STS can be associated to arm
support.
Future work will focus on the development of a contact model so that the effect of
foot contact with the ground during STS can be evaluated. Results will be used for the
optimization of the model, concerning its structural design, actuation and control
methodologies. Furthermore, rigid and complacent solutions will also be addressed.
References
1. WHO | Spinal Cord Injury. WHO. http://www.who.int/mediacentre/factsheets/fs384/en/.
Accessed 16 Nov 2016
2. Chen, B., et al.: Recent developments and challenges of lower extremity exoskeletons.
J. Orthop. Transl. 5, 26–37 (2016)
3. Quinto, L., Gonçalves, S.B., Silva, M.T.: Exoesqueletos para membros inferiores: Estado da
arte. Presented at the Congresso Nacional de Biomecânica, Guimarães, Portugal (2017)
4. Dollar, A.M., Herr, H.: Lower extremity exoskeletons and active orthoses: challenges and
state-of-the-art. IEEE Trans. Robot. 24(1), 144–158 (2008)
Design of a Passive Exoskeleton to Support Sit-to-Stand Movement 303
5. Janssen, W.G., Bussmann, H.B., Stam, H.J.: Determinants of the sit-to-stand movement: a
review. Phys. Ther. 82(9), 866–879 (2002)
6. Winter, D.A.: Biomechanics and Motor Control of Human Movement, 4th edn. Wiley,
Hoboken (2009)
7. Silva, M.T.: Human Motion Analysis Using Multibody Dynamics and Optimization Tools.
Instituto Superior Técnico - Universidade Técnica de Lisboa, Lisboa, Portugal (2003)
Walking Assistance of Subjects with Spinal
Cord Injury with an Ankle Exoskeleton
and Neuromuscular Controller
M. Arquilla1, I. Pisotta1, F. Tamburella1, N. L. Tagliamonte1(&),

M. Masciullo1, A. R. Wu2, C. Meijneke3, H. van der Kooij4,5,
A. J. Ijspeert2, and M. Molinari1
1
IRCCS Fondazione Santa Lucia, NeuroRobot Lab, Rome, Italy
n.tagliamonte@hsantalucia.it
2
École Polytechnique Fédérale de Lausanne, BIOROB Lab, Lausanne,
Switzerland
3
Electronic and Mechanical Support Division, Delft University of Technology,
Delft, The Netherlands
4
Biomechatronics and Rehabilitation Technology, Department of
Biomechanical Engineering, University of Twente, Enschede, The Netherlands
5
Delft University of Technology, Delft, The Netherlands
Abstract. This work was devoted to preliminary test the Achilles ankle
exoskeleton and its NeuroMuscular Controller (NMC) with a test pilot affected
by incomplete spinal cord injury. The customization of the robot controller, i.e. a
subject-specific tailoring of the assistance level, was performed and a 10-session
training to optimize human-robot interaction was finalized. Results demon-
strated that controller tuning was in line with the functional clinical assessment.
NMC adapted to the variable walking speed during the training and the test pilot
was successfully trained in exploiting robotic support and also improved his
performance in terms of walking speed and stability. After the training, a higher
speed could also be achieved during free walking and hence a slight unexpected
rehabilitation effect was evidenced.
1 Introduction
Human-robot mutual adaptation is a key mechanism to be considered in the control of

assistive exoskeletons. Robotic support should be not provided by setting predefined
motion patterns but rather by properly adapting assistance to the residual motor
functions of the user, who in turn should be able to exploit robotic aid while preserving
his personal active control over the machine. This aspect is particularly crucial in the
case of exoskeleton-aided walking of subjects with incomplete Spinal Cord Injury
(SCI), who might demonstrate preserved functions and need only partial robotic
Financial support for this work was provided by the European Union research program FP7-ICT
(SYMBITRON grant #611626).

https://doi.org/10.1007/978-3-030-01887-0_58
Walking Assistance of Subjects with Spinal Cord Injury 305
support. This paper presents the results of the preliminary experimental validation of
the Achilles exoskeleton [1] controlled through a novel NeuroMuscular Controller
(NMC) [2] and used by a test pilot with incomplete SCI. NMC is capable of modu-
lating the level of assistance (by scaling commanded assistive torques) to account for
subject-specific conditions (e.g. to accommodate for left/right leg or flexor/extensor
muscles asymmetries), without imparting fixed joint kinematic trajectories. The NMC
is capable of automatic and online adaptation to a variety of walking speeds without
any specific commands. In [2] the NMC was tested in pilot experiments on a treadmill-
based gait trainer (LOPES III). In this work, the NMC was tested for the first time with
a test pilot with a spinal cord injury on an overground exoskeleton and with a multiple-
session training.
2.1 Achilles Ankle Exoskeleton

Achilles is an exoskeleton designed to assist ankle plantar/dorsiflexion during walking
(Fig. 1a). The system has compliant linear actuators (motor, spindle drive and leaf
spring) for ankle joint torque control: the measurement of the spring deflection,
together with the knowledge of its stiffness and of the kinematic structure, allows the
estimation of the delivered torque to be used as a feedback in a low-level torque control
loop. The robot is connected to the lower leg through a carbon fiber shank shell and to
the foot through a carbon fiber foot shell inserted within a standard gym shoe. Sen-
sorized insoles in the feet shells are used to detect ground foot contact and estimate gait
phases.
motor encoder
motor
spindle
shank shell
leaf spring
ankle encoder
foot shell
pressure sensor
(a) (b)
Fig. 1. (a) Drawing of the Achilles exoskeleton with its components. (b) Test pilot with SCI
wearing the Achilles exoskeleton.
306 M. Arquilla et al.
2.2 Neuromuscular Controller

The NMC, implemented on top of the Achilles torque controller, is based on a reflex-
based neuromechanical model of walking developed in [3]. The controller model
consists of two virtual Hill-type muscle units per leg: Tibialis Anterior (TA), and
Soleus (SO). It uses joint angles and ground contact information as inputs to produce
desired joint torques to be delivered by the Achilles. NMC can produce human-like
kinematics, kinetics, and muscle activations. Joint angles yield muscle length and
velocity sensory information, which are used to activate muscles through reflex loops.
During stance the SO is driven by positive force feedback to increase tension, while
negative force feedback decreases the tension on the TA. During swing and throughout
the gait cycle, a positive length feedback on TA prevents overextension. A gain
multiplying the nominal torque output modulates the assistance level (100% corre-
sponds to the muscles contribution required for the simulated model to walk at
1.3 m/s).
2.3 Enrolled Test Pilot

The test pilot (Fig. 1b) is a male, 48 y.o., with incomplete lesion at C7 due to a
traumatic event in November 2013. AIS and WISCI levels are D and 20, respectively.
2.4 Controller Customization and Walking Training

NMC assistance was initially customized based on systematic tests taking into account
test pilot’s quantitative performance (walking speed) and qualitative feedback (per-
ceived usability). During the preliminary tuning phase, three levels of assistance for
each leg were tested: 0%, 50% and 100% of the torque deliverable by the nominal
simulated model. Ten days of training were performed (three times per week) with the
main goal of improving the maximum comfortable walking speed (primary outcome).
During each session, after a preliminary familiarization phase and balance exercises,
preparatory for handling antero-posterior and medio-lateral weight shifting, the test
pilot was asked to walk for 20 min with his optimized NMC setting and to try to
progressively increase his comfortable speed.
2.5 Assessment
Before the training (T0), after 5 training sessions (T1) and at the end of the training
(T2) the test pilot was asked to walk at his maximum comfortable speed five times on a
5-m walking path including four forces plates (P6000, BTS Bioengineering, Italy) to
record speed, step length/width and stance percentage. Moreover, the 10 Meter
Walking Test (10MWT) was performed with and without the Achilles before and after
each training session. At T0 and T2 the Manual Muscle Test (MMT) was also
administered.
Walking Assistance of Subjects with Spinal Cord Injury 307
3 Results
The subject-specific customization phase led to a final preferred NMC setting con-
sisting in a 100% assistance on the left leg and 50% assistance on the right leg. This
setting was consistent with the clinical functional assessment based on the MMT,
which evidenced a higher motor deficit on the left leg. No clinical differences were
assessed at T2. Results of the force plates and of 10MWT assessments are reported in
Figs. 2 and 3, respectively.
Fig. 2. Results of the quantitative assessment (step length and width are averaged between left
and right sides).
Fig. 3. Results of the 10MWT.
4 Discussion
An increase in walking speed and in step length was found after the training (T2 with
respect to T0), demonstrating an improvement of the performance in line with the
training primary outcome, as well as a decrease in step width and stance duration (with
a slight right/leg asymmetry), demonstrating an improvement in walking stability.
10MWT highlighted the achievement of a performance plateau after around 5 training
sessions and a slight improvement of the performance even without the use of the
robot, thus suggesting an un expected rehabilitation effect due to the training. No major
differences were highlighted in the 10MWT before and after each training session.
308 M. Arquilla et al.
5 Conclusion
The Achilles assistive exoskeleton, controlled through the NMC was validated on one
subject with incomplete SCI during a 10-session training. The tuning of controller
setting was properly optimized and appeared to be in line with functional clinical
needs. NMC adapted to test pilot’s variable walking speed during the training and the
test pilot was successfully trained in exploiting Achilles support. He improved his
performance in terms of walking speed and stability while still preserving the control
over the device. Despite the Achilles is an assistive exoskeleton, a slight unexpected
rehabilitation effect was demonstrated by an increase in walking speed without the use
of the robot after the training. Future work will be devoted to extending tests to
additional subjects with SCI.
References
1. Meijneke, C., van Dijk, W., van der Kooij, H.: Achilles: an autonomous lightweight ankle
exoskeleton to provide push-off power. In: 5th IEEE RAS/EMBS BIOROB, Sao Paulo,
pp. 918–923 (2014)
2. Wu, A.R., Dzeladini, F., Brug, T.J.H., Tamburella, F., Tagliamonte, N.L., van Asseldonk,
E.H.F., van der Kooij, H., Ijspeert, A.J.: An adaptive neuromuscular controller for assistive
lower-limb exoskeletons: a preliminary study on subjects with spinal cord injury, Front.
Neurorob. 11, 30 (2017)
3. Geyer, H., Herr, H.: A muscle-reflex model that encodes principles of legged mechanics
produces human walking dynamics and muscle activities. IEEE Trans. Neural Syst. Rehabil.
Eng. 18, 263–273 (2010)
Center of Mass and Postural Adaptations
During Robotic Exoskeleton-Assisted Walking
for Individuals with Spinal Cord Injury
Arvind Ramanujam1, Kamyar Momeni1,2, Syed R. Husain1,

Jonathan Augustine1, Erica Garbarini1, Peter Barrance1,2,
Ann M. Spungen3, Pierre K. Asselin3, Steven Knezevic3,
and Gail F. Forrest1(&)
1
Kessler Foundation, West Orange, NJ, USA
gforrest@kesslerfoundation.org
2
Rutgers, New Jersey Medical School, Newark, NJ, USA
3
James J. Peters Veterans Affairs Medical Center, Bronx, NY, USA
Abstract. The goal of this study is to understand the postural adaptations

characterized by the whole body center of mass (COM) for individuals with SCI
while walking with powered robotic exoskeletons, EksoGTTM and ReWalkTM.
COM excursions showed a greater medial-lateral weight shift approach while
walking in the EksoGTTM compared to a more forward-lean approach in the
ReWalk™, however, postural trunk lean was significantly (p < 0.05) higher in
the ReWalkTM. Understanding the effects of exoskeleton designs on posture and
sway is crucial towards developing effective and efficient training protocols for
rehabilitation and recovery post SCI.
1 Introduction
Spinal cord injury (SCI) is characterized by loss of motor and/or sensory function
below the site of the injury limiting independent overground (OG) ambulation and
participation within the community, leading to reduced quality of life [1]. Restoring the
ability to walk independently and other functions related to activities of daily living are
paramount for individuals with SCI [2]. Powered robotic exoskeletons have been used
for rehabilitation, mobility, and walking OG in those with SCI to potentially improve
mobility and independence [3]. Currently, three available devices have marketing
clearance by the U.S. FDA for personal use [4–6], and five for clinical use. Under-
standing the differences in gait stability is crucial towards improving walking efficiency
and fall prevention in these powered exoskeletons for both clinical and community use.
Dynamic gait stability has been assessed over the years using COM position and
momentum as it relates to center of pressure and base of support [7, 8]. In this study, as
Research supported by New Jersey Commission on Spinal Cord Research (Grant# CSCR13IR
G013).

https://doi.org/10.1007/978-3-030-01887-0_59
310 A. Ramanujam et al.
a first step towards assessing gait stability, we present a preliminary postural analysis of
the whole body COM during robotic-assisted gait for individuals with SCI.
2.1 Participant Demographics and Training Protocol

Fourteen individuals (9 males, 5 females) with chronic incomplete motor SCI (Age:
41.71 ± 11.72 years old, Weight: 75.93 ± 13.06 kg, Height: 1.75 ± 0.09 m, Time
post injury: 9.50 ± 7.44 years, Level of injury: C4-T11) and six male able-bodied
volunteers (Age: 29.50 ± 4.97 years old, Weight: 82.57 ± 13.23 kg, Height:
1.80 ± 0.07 m) were recruited and completed an informed consent approved by the
Kessler Foundation and James J. Peters Veterans Affairs Medical Center Institutional
Review Boards. Eight individuals with SCI trained using the EksoGT™ (Ekso Bionics,
Richmond, CA) while six individuals trained using ReWalk™ (ReWalk Robotics, Inc.,
Marlborough, MA) for 3–4 days/week (*5 h/week), accumulating 100 h of walking.
Able-bodied (AB) individuals received two sessions of training whereby they could
walk using the robotic exoskeleton with close supervision. Training sessions were
conducted under the guidance of a physical therapist, biomedical engineer, and an
exercise physiologist. At baseline, EksoGT™ was configured to offer the highest level
of assistance (“Max-Assist”, [9, 10]) and was progressively reduced (“Fixed-Assist”)
based on the individual’s ability in subsequent training sessions. ReWalk™ was
configured progressively to facilitate the training mode of the participant [3].
2.2 Data Collection and Statistical Analysis

Kinematic data (Motion Analysis Corporation, Santa Rosa, CA) were collected at
60 Hz, filtered (2nd order, low pass Butterworth filter, 6 Hz cut-off frequency) and
normalized to percent gait cycle at baseline and post-training (>100 h) for the SCI
group with their trained devices and AB group, with both devices and overground
(OG) walking at self-selected (SS) speed. Whole body COM excursions (medial-
lateral: COMML and anterior-posterior: COMAP), inclination angles (IA) and step-
width (StepW, distance between both feet) were calculated using the kinematic data
[8]. An analysis to compare devices (EksoGT™ vs ReWalk™) was performed for all
outcome variables. Analysis of covariance (ANCOVA) was performed, with treatment
group as the between-participants main effect in the model, and time since injury at
baseline and baseline measures as covariate. Paired-samples T-tests were used to
compare the means within each device across baseline and post-training time points
(p < 0.05).
Center of Mass and Postural Adaptations 311
3 Results
3.1 COM Excursions
Post-training, mean COMML and COMAP excursions were greater with the EksoGTTM,
especially for the “Fixed-Assist” condition; however, the step-widths (StepW) were
narrower. For ReWalk™, despite significantly greater StepW compared to EksoGTTM
(n = 14, p < 0.001), the COMML excursions were similar with marginally greater
COMAP excursions (Fig. 1A).
3.2 COM Inclination Angles

Lead & Trail Leg IAs: Within each device, lead leg IAs at heel-strike increased post-
training while no changes were observed in the trail leg IAs. Trunk Lean: Postural trunk
lean increased post-training in both devices and was significantly (n = 14, p < 0.05)
higher with the ReWalk™ compared to EksoGTTM. Included angle: COM included
angle was greater post-training for both devices and strongly correlated to stride length
(r2 = 0.925). Included angles and stride lengths were greater with the ReWalk™
compared to EksoGTTM (Fig. 1C).
4 Discussion
The COM excursions observed are indicative of the specific training paradigms for
EksoGTTM and ReWalk™. The EksoGTTM relies on the individual’s ability to transfer
their weight laterally onto one foot to complete the stepping motion on the contralateral
foot combined with moving the contralateral crutch forward. Repetitive training to
achieve the lateral weight shift targets in addition to the increased resistance offered by
the EksoGTTM in the progressive training modes (i.e. “Fixed-Assist”) translated into
greater COMML excursions post-training for a given step-width. Comparatively, the
ReWalk™ operates using a “tilt” action that initiates a step by tilting the trunk ante-
riorly and moving both crutches forward simultaneously. This is shown by greater
COM excursion in the AP direction compared to ML, a more forward position of the
COM at heel-strike and greater trunk lean in the ReWalk™ compared to EksoGTTM
(Fig. 1). Based on the definitions of the lead leg IA and the included angle, the greater
the lead leg IA, the greater the step length with the lead foot. Post > 100 h of training,
the study group was able to increase their lead leg IA to increase stride lengths
bilaterally in both devices.
312 A. Ramanujam et al.
Fig. 1. (A) Mean Medial-Lateral (X-axis) and Anterior-Posterior (Y-axis) COM excursions (in
meters) with respect to the mid-point between feet (shown by horizontal and vertical black lines)
during a gait cycle (left). Circular makers represent point at left heel-strike. Shaded area is the
step-width. (B) Diagram to show COM inclination angle (IA) calculations with respect to lead
leg, trail leg, trunk and included angle between both legs. (C) Jitter box-plots with mean, 95%
confidence intervals and standard deviation for lead & trail leg IAs, included angel and trunk lean
(degrees) showing the spread of individual data points at heel-strike for conditions 1–8
(*p < 0.05).
5 Conclusion
COM and postural adaptations after longitudinal exoskeleton training vary based on the
devices’ training paradigms. Exoskeleton users adopted a medial-lateral weight shift
approach while walking in the EksoGTTM compared to a more forward-lean approach
in the ReWalk™. Understanding the variations in exoskeleton designs and their effect
on posture and sway is crucial towards developing effective and efficient training
protocols for rehabilitation and recovery post SCI. Future research will further inves-
tigate the dynamic stability measures derived from COM and COP excursions taking
into consideration location of bilateral crutches relative to base of support, during
robotic-assisted gait, as well as independent OG walking without the robot.
Center of Mass and Postural Adaptations 313
References
1. Anderson, K.D.: Targeting recovery: priorities of the spinal cord-injured population.
J. Neurotrauma 21, 1371–1383 (2004)
2. Ditunno, P.L., Patrick, M., Stineman, M., Ditunno, J.F.: Who wants to walk? Preferences for
recovery after SCI: a longitudinal and cross-sectional study. Spinal Cord. 46(7), 500 (2008)
3. Asselin, P., Knezevic, S., Kornfeld, S., Cirnigliaro, C., Agranova-Breyter, I., Bauman, W.A.,
et al.: Heart rate and oxygen demand of powered exoskeleton-assisted walking in persons
with paraplegia. J. Rehabil. Res. Dev. 52(2), 147–158 (2015)
4. FDA: Evaluation Of Automatic Class III Designation (De Novo) For Argore Walk. 15 A.D.
Ref Type: Internet Communication
5. FDA. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfrl/ldetails.cfm?lid=482440. FDA.
16 A.D. Ref Type: Internet Communication
6. HAL For Medical Use (Lower Limb Type) 510(k) Premarket Notification, Accessdata.
fda.gov (2018). https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?ID=
K171909. Accessed 23 Jan 2018
7. Kaya, B.K., Krebs, D.E., Riley, P.O.: Dynamic stability in elders: momentum control in
locomotor ADL. J. Gerontol. A Biol. Sci. Med. Sci. 53, M126–M134 (1998)
8. Lee, H.J., Chou, L.S.: Detection of gait instability using the center of mass and center of
pressure inclination angles. Arch. Phys. Med. Rehabil. 87(4), 569–575 (2006)
9. Ramanujam, A., Cirnigliaro, C.M., Garbarini, E., Asselin, P., Pilkar, R., Forrest, G.F.:
Neuromechanical adaptations during a robotic powered exoskeleton assisted walking
session. J. Spinal Cord Med. 41(5), 518–528 (2017)
10. Ramanujam, A., Spungen, A., Asselin, P., Garbarini, E., Augustine, J., Canton, S., Barrance,
P., Forrest, G.F.: Training response to longitudinal powered exoskeleton training for SCI. In:
Wearable Robotics: Challenges and Trends 2017, pp. 361–366. Springer (2017)
Exoskeleton Controller and Design
Considerations: Effect on Training Response
for Persons with SCI
Gail F. Forrest1(&), Arvind Ramanujam1, Ann M. Spungen3,

Christopher Cirnigliaro3, Kamyar Momeni1,2, Syed R. Husain1,
Jonathan Augustine1, Erica Garbarini1, Pierre K. Asselin3,
and Steven Knezevic3
1
Kessler Foundation, West Orange, NJ, USA
gforrest@kesslerfoundation.org
2
New Jersey Medical School, Rutgers, Newark, NJ, USA
3
James J. Peters Veterans Affairs Medical Center, Bronx, NY, USA
Abstract. The objective of this research is to identify the demographic,

physiological, kinematic and biomechanical determinants of exoskeleton assis-
ted gait speed for individuals with a spinal cord injury (SCI). High number
(300) of gait cycles across multiple time-points were analyzed to identify the
parameter estimates from mixed model for dependent variable walk speed. Step
length, step width, single stance time did not contribute to walk speed whereas
trunk lean mass, stride length, step frequency were the most significant con-
tributors. These variables were more significant than any of the spatial temporal
parameters that are associated with human gait. Future research should deter-
mine the relative contributions of each independent variable to walk speed for
different devices. Understanding the effects of exoskeleton/human interface for
different devices is crucial for developing effective/efficient training protocols
for community ambulation, rehabilitation and recovery post SCI.
1 Introduction
The determinants controlling exoskeleton/human interface assisted gait speed are

especially relevant to rehabilitation, function and community ambulation using pow-
ered robotic devices [1]. For individuals with impaired motor and sensory function, a
natural interface between user and device is critical for community functioning as well
as facilitating recovery [1]. Further, a compromised musculoskeletal and neuromus-
cular system will inhibit the neural interaction between device and user [2]. Recent
research has shown that exoskeleton training can decrease upper/lower limb muscle
atrophy [3] and increase neuromuscular input to facilitate neural interaction between
device and user. Currently there are five US Food and Drug Administration
Research supported by New Jersey Commission on Spinal Cord Research (Grant# CSCR13IR
G013).

https://doi.org/10.1007/978-3-030-01887-0_60
Exoskeleton Controller and Design Considerations 315
(FDA) approved devices for clinical rehabilitation use and two for home use [4, 5].
Several studies have evaluated the biomechanical and neural effects of exoskeleton
training on kinematic and biomechanical training variables among different devices [6–
8]. However, there is minimal research evaluating the exoskeleton/human interface
effect on recovery outcomes or how differences in exoskeleton/human interface con-
troller can influence outcomes. The objective of this study is to identify, for a given
device controller, the demographic, physiological, kinematic and biomechanical
determinants of exoskeleton assisted gait speed for individuals with SCI.
2.1 Participant Demographics and Training Protocol

Fourteen individuals (9 males, 5 females) with chronic incomplete motor SCI (Age:
41.71 ± 11.72 years old, Weight: 75.93 ± 13.06 kg, Height: 1.75 ± 0.09 m, Time
post injury: 9.50 ± 7.44 years, Level of injury: C4-T11) were recruited and completed
an informed consent approved by the Kessler Foundation and James J. Peters Veterans
Affairs Medical Center Institutional Review Boards. Gait data were collected using
EksoGT™ (Ekso Bionics, Richmond, CA), configured to “Max-Assist” [7, 9] to
facilitate the training mode of the participant [3]. All individuals independently walked
using the exoskeleton without therapist assistance, albeit close supervision of physical
therapist was provided. Training details have been previously described [8].
2.2 Data Collection and Statistical Analysis

Kinematic data (Motion Analysis Corporation, Santa Rosa, CA) at multiple time-points
(baseline and post-intervention) were collected at 60 Hz, filtered (2nd order, low pass
Butterworth filter, 6 Hz cut-off frequency) and normalized to percent gait cycle for each
device condition. Dual energy x-ray absorptiometry (DXA, GE Lunar iDXA, enCore
and CoreScan, platform version 14.0, Madison, WI) was employed to measure total
body weight (TBW), left & right leg lean mass (LLegLM & RLegLM) and trunk lean
mass (TrunkLM). Three hundred bilateral gait cycles determined independent variables
for gender, age, body composition (LLegLM, RLegLM, TrunkLM), temporal measures
(time in secs and %) during Initial Double Stance (IDS), Single Stance (SS), Terminal
Double Stance (TDS), Swing (SW) and Toe off (ToeOff), spatial measures comprised
of Step Frequency (StepFreq), Step Length (StepL), Step Width (StepW) and Stride
Length (StrideL), and kinematic variables associated with center of mass (COM) in-
clination angles (IA) (Lead & Trail Leg IAs, included angle, trunk lean angle) [9].
Mixed effect model is applied to account for possible strong correlations between speed
measurements (Table 1) within each subject. No pattern is suggested by residual plot
(Fig. 1), indicating no violation to the assumption of mixed model.
316 G. F. Forrest et al.
Table 1. Summary statistics of walk speed (m/s)

Mean Std. Dev Min Median Max
Pre 0.154 0.072 0.040 0.170 0.308
Post 0.196 0.061 0.116 0.186 0.331
Female 0.127 0.061 0.040 0.117 0.252
Male 0.204 0.059 0.116 0.197 0.331
Table 2. Mixed model parameter estimate for dependent variable walk speed com (m/s) for
EksoGTTM device, left gait cycle.
Effect Estimate SE DF t Val P > |t|
Intercept −0.3541 0.07582 14 −4.67 0.0004
Pre/Post 0.002269 0.002188 187 1.04 0.3009
Sex (M/F) 0.004390 0.003810 187 1.15 0.2507
Age (yr) 0.000639 0.000177 187 3.61 0.0004
RLegLM (gm) 9.635E−6 3.397E−6 187 2.84 0.0051
LLegLM (gm) −0.00001 3.267E−6 187 −3.65 0.0003
TrunkLM (gm) −1.39E−6 0 187 −Infty <.0001
IDS (secs) 0.002921 0.000930 187 3.14 0.0020
SS (secs) −0.00773 0.005334 187 −1.45 0.1488
TDS (secs) 0.003776 0.000962 187 3.92 0.0001
Swing (s) 0.04505 0.01335 187 3.38 0.0009
StepFreq (s) 0.3957 0.02059 187 19.22 <.0001
ToeOff (%) 0.002331 0.000639 187 3.65 0.0003
StepL (mm) −1.67E−6 0.000032 187 −0.05 0.9590
StepW (mm) 0.000041 0.000035 187 1.16 0.2484
StrideL (mm) 0.000194 0.000027 187 7.28 <.0001
LeadLegIA 0.000940 0.000446 187 2.11 0.0365
TrailLegIA −0.00234 0.000453 187 −5.18 <.0001
Included Ang. 0.001598 0.000489 187 3.27 0.0013
TrunkLean 0.000813 0.000352 187 2.31 0.0221
**** Significance level P < 0.05.
Exoskeleton Controller and Design Considerations 317
3 Results
3.1 Gender and Velocity Comparison
Parameter estimates in Table 2 indicate that gender did
not contribute to determine walk speed. Difference
between pre and post time-point for walking speed was
also not statistically significant.
3.2 Body Composition Variables

LLegLM, RLegLM and TrunkLM had significant effect
on walk speed with TrunkLM having the greater effect.
3.3 Spatial Temporal Variables

and Inclination Angles Fig. 1. Studentized residuals
for walk speed (m/s). actual
SS, StepL, and StepW were not associated with walk mean – predicted mean: walk
speed while StrideL, StepFreq, ToeOff, IDS, TDS, and speed.
Swing had significant effect on walk speed for the Ekso
device. StrideL, StepFreq, ToeOff, and TDS had greater
effect than IDS and Swing. Trail Leg IA, Lead Leg IA, COM Included Angle, and
Trunk Lean angle all influence the dependent variable, walk speed for the EksoGTTM
device. The greatest significant independent variable to negatively affect walk speed
was Trail Leg IA. COM included angle had the greatest positive effect on walk speed.
4 Discussion
The research determined that the most significant variable(s) influencing the mixed
regression model for the dependent variable walking speed was TrunkLM, StepFreq,
StrideL, and Trail Leg IA. The Trail Leg IA negatively impacted the regression model
determining walk speed. More sensitive analyses is required to identify the principal
components or main independent variable that determines exoskeleton gait speed.
Surprisingly step width and step length were not associated to walk speed unlike
independent locomotion where both variables affect gait velocity [7, 9].
5 Conclusion
The method described in this report identifies the variables important to

exoskeleton/user control for the effect of exoskeleton training and gait speed. Future
work using this technique could also be used for analyses of dependent variables
associated with recovery outside of the device.
318 G. F. Forrest et al.
References
1. Koller, J.R., David Remy, C., Ferris, D.P.: Comparing neural control and mechanically
intrinsic control of powered ankle exoskeletons. IEEE Int. Conf. Rehabil. Robot. 2017, 294–
299 (2017). https://doi.org/10.1109/ICORR.2017.8009262
2. Wu, A.R., Dzeladini, F., Brug, T.J.H., Tamburella, F., Tagliamonte, N.L., van Asseldonk, E.
H.F., van der Kooij, H., Ijspeert, A.J.: An adaptive neuromuscular controller for assistive
lower-limb exoskeletons: a preliminary study on subjects with spinal cord injury
3. Forrest, G.F., Spungen, A.M., Bauman, W.A., et al.: Muscle changes after Exoskeleton
training. ISCOS Meeting, Dublin, Ireland (2017)
4. FDA. Evaluation of Automatic Class III Designation (De Novo) For Argore Walk. 15 A.D
5. FDA. http://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfrl/ldetails.cfm?lid=482440. FDA.
16 A.D
6. Tefertiller, C., Hays, K., Jones, J., Jayaraman, A., Hartigan, C., Bushnik, T., Forrest, G.F.:
Initial outcomes from a multicenter study utilizing the Indego powered exoskeleton in spinal
cord injury. Top. Spinal Cord Inj. Rehabil. 2018 Winter 24(1), 78–85 (2018). https://doi.org/
10.1310/sci17-00014. Epub 20 Nov 2017
7. Ramanujam, A., Cirnigliaro, C.M., Garbarini, E., Asselin, P., Pilkar, R., Forrest, G.F.:
Neuromechanical adaptations during a robotic powered exoskeleton assisted walking session.
J. Spinal Cord Med. 20, 1 (2017)
8. Ramanujam, A., Spungen, A., Asselin, P., Garbarini, E., Augustine, J., Canton, S., Barrance,
P., Forrest, G.F.: Training response to longitudinal powered exoskeleton training for SCI. In:
Wearable Robotics: Challenges and Trends, pp. 361–366. Springer International Publishing
(2017)
9. Ramanujam, A., Momeni, K., Husain, S.R., Augustine, J., Garbarini, E., Barrance, P.,
Spungen, A.M., Asselin, P.K., Knezevic, S., Forrest, G.F.: Center of mass and postural
adaptations during robotic exoskeleton assisted walking for individuals with spinal cord
injury. Submitted to WeROB, Italy (2018)
Biorobotics Approaches to Understand
and Restore Human Balance
Integrating Posture Control in Assistive
Robotic Devices to Support Standing Balance
T. Mergner1(&) and V. Lippi2

1
Neurological University Clinic, University of Freiburg,
Freiburg im Breisgau, Germany
thomas.mergner@uniklinik-freiburg.de
2
Technical University Berlin, Control Systems, Berlin, Germany
vittorio.lippi@tu-berlin.de
Abstract. To date, exoskeletons typically only allow paraplegic users to stand

or walk quadruped-like with crutches to maintain balance. The problem with
today’s robotic assistive devices that are supporting or restituting stance and
walking in paraplegic users is inadequate posture control, which endangers
balance and increases the likelihood of a fall occurring. We address this issue in
this Methods article by describing the posture-movement interrelations in
humans, suggesting the inclusion of posture control in assistive robotic devices,
and recommending their experimental testing prior to application for biped use.
1 Introduction
When humans perform voluntary movements such as reaching for an object, they
consciously control hand, arm and body movements while subconsciously performing
postural compensations. The postural compensations in this example are primarily the
gravitational torques arising in the ankle, hip, and shoulder joints. The reaching motion
is under conscious control of cortical brain centers, whereas the compensation of the
self-produced disturbances tends to be performed subconsciously by posture control
centers in the extrapyramidal system (EPS), which is located mainly in the brainstem,
basal ganglia and cerebellum. When a physically handicapped patient uses a robotic
device to augment or perform body movements, the device should provide its own
postural adjustments. In the case where the patient’s deficits include very severe sen-
sory and/or motor defects (e.g., missing sensory information from the feet), the device
might even provide the balancing of both the exoskeleton and patient. Danger of falling
may arise if the patient erroneously interprets the robot’s postural actions as external
disturbances (having an impact on both body and device) and thereby, tries to enforce a
compensation that is inadequate (a ‘user-device conflict’, possibly with disastrous
positive feedback). These considerations apply to patients who use an exoskeleton for
biped balancing, which we consider a desired goal for the future.
Understanding posture control mechanisms in humans and the humanoid robot is a
prerequisite for research of this topic. In retrospect, many falls in DARPA challenges may
be due to insufficient or inappropriate consideration of posture control in biped stance or
walking [1]. Therefore, posture control has been included in recent proposals for robot

https://doi.org/10.1007/978-3-030-01887-0_61
322 T. Mergner and V. Lippi
benchmarking, which suggested using human balancing as a gold standard for humanoids
[2]. A recent article from our laboratory [3] targeted the issue of balancing stance. Therein,
we noted that the large diversity of software and hardware solutions in robots may be
successfully handled given that the applied balancing tests address (a) the four basic,
physical disturbances (support surface rotation and translation, contact and field forces)
and their compensation, and (b) the most critical joints used for standing balance (ankle
and hip joints in the body’s sagittal plane; see [3, 4]). We conceive that implementation of
robotic assistance for biped stance and walking will become possible in the future.
However, we maintain that this still requires considerable research on how to provide
postural stability in the human-robot interaction and cooperation.
In the following article, we outline the postural control tests that address the four
basic external or self-produced posture disturbances common in both humans and
humanoid robots. We posit that these tests can be similarly applied to humans using
robotic devices for biped stance and walking. Balancing of biped walking and the
disturbance compensation in ankle and hip joints, such as balancing stance by making
steps or holding with the arms, or by adjusting foot placements during walking, are not
discussed in this article.
2 Results
2.1 Tests of the Four Basic Physical Disturbances

The four basic physical disturbances affecting standing balance which are well captured
by established balancing tests often used for the evaluation of human balance control
include: Support surface rotation (1) and translation (2), as well as contact forces
applied as a pull or push perturbation (3) and field forces, such as gravity (4). The
stimuli tend to evoke body lean in space and thereby produce or enhance gravitational
torque in the ankle joints. While the experimental implementation of 1–3 is intuitive
(Fig. 1A–C), testing of 4 is often realized as a selective testing of the vestibular sensor
using the ‘body-sway referenced platform’ (BSRP) paradigm. In this test (Fig. 1D),
support surface motion is locked to body sway that arises internally (e.g., by internal
noise). In the absence of visual orientation cues, this selectively evokes the vestibular
signal of body-in-space sway, which then dominates the balancing. The sensory dis-
turbance estimation and compensation of the four basic physical disturbances repre-
sents the core principle of the DEC model of human postural control [4], which has
been implemented as modular control architecture and successfully tested in humanoid
robots (see [3, 5–7]). In the human solution, the control of voluntary movements and
the disturbance compensation are combined in a conflict-free way for each degree of
freedom (DoF) of the musculoskeletal system.
In upright stance, the human balancing control is mainly concerned with sway in
the body’s sagittal plane around the ankle joints and with trunk sway around the hip
joints, as well as the hip-ankle coordination. Plantar force cues play no major role for
the balancing. Rather, the balancing draws mainly on joint proprioceptive, vestibular
and visual cues, which allow humans to balance even on rough terrains where plantar
force distribution may be irregular (cf. [8]).
Integrating Posture Control in Assistive Robotic Devices to Support Standing Balance 323
Fig. 1. The four basic tests of biped postural control (SS, support surface)
2.2 Achieving Conflict-Free Interaction and Co-operation Between

Robotic and Human Postural Controls
The above described DEC principles identified in human control, as well as their
successful implementation in humanoid robots, suggest that these principles can also be
used in a robotic assistive device for maintaining standing balance by paraplegic and
paraparetic patients. Among patients, one faces a variety of sensorimotor deficits,
where pyramidal and extrapyramidal systems may be impaired to various degrees
(while reduced muscle force alone is an exception). Referencing human similarity may
help to model the sensorimotor conditions specific for a patient and to take these into
account when designing or adjusting the control of the assistive system. The DEC
concept allows for the modelling of specific conditions in terms of sensory inputs and
control strategies, and sensory availability. However, DEC control in robotic devices
requires the use of impedance-controlled actuation. Using human-like low loop gain,
maintained at a level sufficient to resist gravity, provides human-like compliance,
which is advantageous for human-robot interaction, collision, and energy consumption.
Current versions of DEC are implemented in Simulink/Matlab, which eases the
migration across PCs for simulations and for controlling robotic platforms in ‘real
world’ tests.
It remains to be shown experimentally to what extent the matching of control
concepts in the patient and the assistive device helps to avoid user-device conflicts such
as the one mentioned above (where the patient experiences postural adjustments of the
assistive device as external disturbances). We conceive that patients may learn to deal
with such conflicts. The learning likely requires several evaluation and training ses-
sions, in which the balancing tests are repeatedly performed and appropriate skills are
developed with the help of cognition and vision.
324 T. Mergner and V. Lippi
3 Conclusions
A paraplegic or paraparetic patient controlling balance in biped stance using an

assistive device may face user-device conflicts in posture control, which requires
testing for each individual case and eventually training the patient to cope with the
conflict. This issue requires further research into such user-device interactions and
training for co-operation. To this end, we suggest applying specific tests, which are
already in use in postural control research of both human and humanoid robots. We
maintain that designing the control of the assistive device should not only take into
account actuation and biomechanics abilities of the user, but also the user’s notion to
use sensory-based compensation of external disturbances. Neurorobotics can provide
models (see [6, 9]) that can be used to integrate the corresponding sensory-based
mechanisms into the device.
References
1. https://www.bbc.com/news/technology-33045713, https://www.youtube.com/watch?v=g0Ta
YhjpOfo
2. Torricelli, D., Gonzalez-Vargas, J., Veneman, J.F., Mombaur, K., Tsagarakis, N., del-Ama,
A.J., et al.: Benchmarking bipedal locomotion: a unified scheme for humanoids, wearable
robots and humans. IEEE Rob. Autom. Mag. 22, 103–115 (2015)
3. Mergner, T., Lippi, V.: Posture control - Human-inspired approaches for humanoid robot
benchmarking: Conceptualizing tests, protocols and analyses. Front. Neurorobotics 12, 21
(2018)
4. Mergner, T.: A neurological view on reactive human stance control. Annu. Rev. Control 34,
77–198 (2010)
5. Hettich, G., Assländer, L., Gollhofer, A., Mergner, T.: Human hip-ankle coordination
emerging from multisensory feedback control. Hum. Mov. Sci. 37, 123–146 (2014)
6. Lippi, V., Mergner, T.: Human-derived disturbance estimation and compensation
(DEC) method lends itself to a modular sensorimotor control in a humanoid robot. Front.
Neurorobotics 11, 49 (2017)
7. Ott, C., et al.: Good posture, good balance: comparison of bioinspired and model-based
approaches for posture control of humanoid robots. IEEE Robot. Autom. Mag. 23(1), 22–33
(2016)
8. Mergner, T., Peterka, R.J.: Human sense of balance. In: Goswami, A., Vadakkepat, P. (eds.)
Humanoid Robotics: A Reference, pp 1–38. Springer, Dordrecht (2017)
9. Zebenay, M., Lippi, V., Mergner, T.: Human-like humanoid robot posture control. In: 2015
12th International Conference on Informatics in Control, Automation and Robotics
(ICINCO), vol. 2, pp. 304–309. IEEE (2015)
A Computational Framework for Muscle-Level
Control of Bi-lateral Robotic Ankle
Exoskeletons
Guillaume Durandau, Herman van der Kooij, and Massimo Sartori(&)

P.O. Box 217, 7500 AE Enschede, The Netherlands
m.sartori@utwente.nl
Abstract. Recent effort in exoskeleton control resulted in reduction of human

metabolic consumption during ground-level walking. In this context, solutions
that would enable biomechanical and metabolic benefits across large repertoires
of motor tasks would be central in supporting the human in both medical and
industrial scenarios. With this idea in mind we created a muscle-driven con-
troller based on electromyography (EMG)-driven musculoskeletal modeling that
we interfaced with the robotic bi-lateral Achilles ankle exoskeleton previously
developed in our group. Preliminary results on one healthy individual show the
possibility of continuously decoding EMG-dependent muscle force and result-
ing ankle joint moment patterns in real-time across a range of diverse motor
tasks. We demonstrate that this information can be used to establish a human-
exoskeleton interface with high-resolution at the level of single muscle
mechanics.
1 Introduction
Exoskeleton can be a game changer in preventing injury in industry setting and in

rehabilitation of paretic patients. Unfortunately, in rehabilitation settings the most
advanced gait retainers commercially available (e.g. Lokomat Hocoma, Switzerland)
still do not provide consensus on their benefit with respect to classic physiotherapy-
based rehabilitation training [1]. In industry settings, heavy working tasks supported by
the assistance of an exoskeleton are still not common. Two major unresolved chal-
lenges may be identified. The first one, is the robust decoding of the user’s intention
and the second one is the ability of operating robotic exoskeletons synchronously to
human intentions and providing assistance that can vary depending of the wearer’s
need. To tackle these challenges, we propose the integration of a computational
muscle-driven framework for the control of a bi-lateral ankle joint robotic exoskeleton.
This concept is based on electromyography (EMG)-driven musculoskeletal modelling.
Our proposed method allows continuous joint moment decoding. Furthermore, by
giving a percentage of the computed human joint moment to the exoskeleton controller,
a variable assistance scheme can be established. In this context, it has been recently
shown that the timing of the assistance delivered by an exoskeleton is one of the key
element for reducing metabolic consumption [2]. We also recently showed two major

https://doi.org/10.1007/978-3-030-01887-0_62
326 G. Durandau et al.
results. The real-time computation of joint torque using EMG-driven modeling and the
possibility to extrapolate the joint moment computation to unknown task (i.e. beyond
the calibration task) [3]. In this paper we explain how we interfaced with the bi-lateral
Achilles ankle exoskeleton [4] and discuss preliminary results showing that we can
compute muscle-level information continuously and synchronously between the user
and the exoskeleton.
Fig. 1. Schematics representation of the framework and its connection with the Achilles
exoskeleton.
Fig. 2. Right ankle muscles moment for multiple continuous task (gait to stair ascending) with
minimal impedance control.
2 Method
We developed a real-time EMG-driven musculoskeletal modeling framework. This

framework is building upon a previously developed algorithm in [5] and later enhanced
to work in real-time [3]. Our EMG-driven musculoskeletal modelling framework
allows the computation of joint moments and intermediate muscle information (muscle
force, pennation angle, fiber length) from EMG signals and joint position.
A Computational Framework for Muscle-Level Control 327
The model includes 14 musculo-tendon unit (MTU) consisting bilaterally of the

Gastrocnemius Medialis, Gastrocnemius Lateralis, Soleus, Tibialis Anterior, Peroneus
Brevis, Peroneus Longus and Peroneus Tertius. In Fig. 1, we present a schematic
representation of the framework and its connection to the Achilles exoskeleton [4]. The
Achilles exoskeleton is a bi-articular exoskeleton for the ankles joint. It allows to
record ankle position and ankle interaction forces between the user and the
exoskeleton. More information can be found in [4]. The inter-communication of the
Achilles is based on the real-time network protocol Ethercat.
The main input of our framework are the ankle joint and the experimental EMG
signal from 10 muscles bilaterally (all previously cited muscles except Peroneus
group). These values are recorded via an Ethercat plugin (Fig. 1A) in real-time and at a
frequency of 1000 Hz. This information is transmitted to our framework which will
convert the joint position to muscle-tendon length (LMT) and muscle moment arm
(MA) for the considered MTU (Fig. 1B). The ankle muscles forces are computed using
Hill-type muscle model with the EMG and LMT as an inputs. The resulting forces are
projected to the ankle joint using the MA (Fig. 1C). The resulting moment is then send
via the Ethercat protocol to the Achilles low-level moment controller (Fig. 1F). To give
subject-specific result the muscle model have to be personalized to the user. A cali-
bration procedure is used to determine the user muscle parameter. This calibration
procedure consist of an optimization that try to reduce the error between moment
computed via our framework and moment recorded by the sensor from the Achilles in
isometric condition.
Fig. 3. Timing to send and receive information between the EMG-driven framework and the
Achilles low level controller.
328 G. Durandau et al.
3 Result and Discussion
Preliminary experiments were done on one user. After calibration, the user performed
over-ground walking followed by stair ascending. Tasks were performed with the
robotic exoskeleton being controlled to enable minimal impedance behavior. The real-
time computed resulting muscle moment contribution can be visualized in Fig. 2. In
Fig. 3, the communication speed to send and receiving information between our
framework and the low-level controller of the Achilles can be visualized. These pre-
liminary results were done to first test the possibility for the framework to compute in
real-time joint and muscle information including moment and force continuously when
interfaced with the Achilles exoskeleton. The second hypothesis is that the computation
and communication were fast enough to be under the muscle electromechanical delay
(EMD) (between 30–100 ms [6]) as synchronization between the assistance provided
by the exoskeleton and the production of force by the user is primordial. Previous result
[3] shows that our framework can compute joint moment with a timing <10 ms and
here, we show that the communication time was <7 ms which is under the EMD.
4 Conclusion
We show in this abstract that our framework can be interfaced with the robotic Achilles
exoskeleton in real-time to compute muscle force and resulting ankle moment con-
tinuously. We also show that these outputs are computed under the EMD which is
central for allowing synchronization between the user and the Achilles. Future work
will test the assistance provided by the framework with different assistance level and
asses the quality of outputted moments.
References
1. Hidler, J., et al.: Multicenter randomized clinical trial evaluating the effectiveness of the
Lokomat in subacute stroke. Neurorehabil. Neural Repair 23(1), 5–13 (2009)
2. Atkeson, C.G., Zhang, J., Fiers, P., Witte, K.A., Jackson, R.W., Poggensee, K.L., Collins, S.H.:
Human-in-the-loop optimization of exoskeleton assistance during walking. Sci. Robot (2017)
3. Durandau, G., Farina, D., Sartori, M.: Robust real-time musculoskeletal modeling driven by
electromyograms. IEEE Trans. Biomed. Eng. (2017)
4. Meijneke, C., van Dijk, W., van der Kooij, H.: Achilles: an autonomous lightweight ankle
exoskeleton to provide push-off power. In: 5th IEEE RAS/EMBS International Conference
Biomedical Robotics and Biomechatronics, pp. 918–923 (2014)
5. Sartori, M., Reggiani, M., Farina, D., Lloyd, D.G.: EMG-driven forward-dynamic estimation
of muscle force and joint moment about multiple degrees of freedom in the human lower
extremity. PLoS One 7(12) (2011)
6. Cavanagh, P.R., Komi, P.V.: Electromechanical delay in human skeletal muscle under
concentric and eccentric contractions. Eur. J. Appl. Physiol. Occup. Physiol. 42(3), 159–163
(1979)
A Conductive Fabric Based Smart Insole
to Measure the Foot Pressure Distribution
with High Resolution
Xinyao Hu(&), Chuang Luo, Dongsheng Peng, and Xingda Qu
Institute of Human Factors and Ergonomics, College of Mechatronics

and Control Engineering, Shenzhen University, Shenzhen, China
huxinyao@szu.edu.cn
Abstract. This study presents a smart insole system made by pressure sensitive
fabric. This fabric insole has 360 sensing arrays. Therefore, it can measure the
foot plantar pressure distribution with a high resolution. An experiment was
carried out to validate this system with the reference measurement system
(F-scan). Ten participants were involved in an experimental study. The results
showed that the fabric insole can measure the foot plantar pressure distribution
accurately, with an average RMSE ranged from 22.46 kPa to 38.54 kPa for
different balancing tasks.
1 Introduction
FOOT pressure distribution is an important parameter in biomechanics. For wearable

robotics, it can provide vital information regarding to the foot-ground interaction.
Clinically, it has been widely used to identify the pathological changes in gait and
assess the balance and fall risks for frail people [1].
The foot pressure distribution can be measured by insole imbedded pressure
sensing system. These systems are commercially available. However, they are gener-
ally very expensive, and not suitable for personal use. More importantly, they are
difficult to be integrated with other wearable systems. Many low-cost smart wearable
insole systems have been developed recently. However, most of them were limited by
the pressure sensing resolution. The pressure sensing resolution is determined by the
number of sensing units available on each insole. Currently, the number of sensing
units is ranged from 4 to 64 among different studies [2, 3]. However, this is far from
enough to register foot plantar pressure distribution clearly.
This study presents a low cost, smart wearable insole system made by conductive
fabric. The insole consists of 360 sensing arrays. Thus, this system can monitor the foot
plantar pressure distribution with high data resolution.
This work was supported in part by the Natural Science Foundation of China [11702175,
31570944], the Natural Science Foundation of Guangdong Province [2016A030310068],
Shenzhen Science Technology, and Innovation Council [JCYJ20160422145322758,
CYJ20150525092940994], and the Shenzhen Peacock Program.

https://doi.org/10.1007/978-3-030-01887-0_63
330 X. Hu et al.
2 Method
2.1 System Design
The smart insole system has three parts: the sensing array, the control unit, and a PC-
end GUI, as illustrated in Fig. 1. An industrial standard conductive fabric (NW170-
SLPA, Eeonyx Inc., CA, USA) was used as the pressure sensing unit. This conducive
fabric was made by a proprietary coating technology. When force is applied onto the
surface, its inner fibers will be squeezed and its resistance becomes smaller [4].
Therefore, the pressure can be measured by obtaining the resistance change. This fabric
was tailored into an 8 mm 8 mm square unit. Each unit can measure the pressure
independently. They were sandwiched and connected by conductive tapes (5 mm in
width) to form a 30 (row) 12 (column) array with 2 mm gaps in between (Fig. 1).
The control unit consisted of a micro controller (STM32 32-bit ARM Cortex,
ARM, Ltd., Cambridge, UK), a customized A2D module, a Bluetooth module (HC-06,
Wavesen Co. Ltd. Guangzhou, China), and a battery. A peripheral circuit was designed
and integrated with the control unit to eliminate the cross-talk effect. The cross-talk
effect is due to the networked resistors (i.e. the sensing units). The resistance mea-
surement could be interfered by the electrical current run through other sensing units.
This was solved by a Voltage Feedback Non-Scanning Electrode (VF-NSE) method
with the peripheral circuit [5]. The details of this implement were described in a prior
study of ours [6].
The data were then transmitted wirelessly to PC, where the pressure distribution
was visualized by a GUI in real time, with the sampling rate of 25 Hz.
Fig. 1. Illustration of the smart insole system
2.2 Experiment and Validation

Prior to experiment, the fabric sensing unit was conditioned and calibrated with the
techniques described in [6]. The force (N) measured by each sensing unit was calcu-
lated based on its voltage output (V) following a second order polynomial equation [6]
and then converted into pressure (kPa).
A Conductive Fabric Based Smart Insole to Measure the Foot Pressure Distribution 331
The F-scan system (Tekscan, Inc., Boston, MA, USA) was used as the reference for
the system validation. The F-scan insole was placed underneath the fabric insole and
cautiously aligned at its top-left corner (this allowed us to establish a coordinate system
with the origin at top-left corner to compare the data from the two systems). The two
systems were synchronized by manually applying external pressure onto them at the
same time, right before the participants stepped in.
Ten young participants were involved in the experiment. They were instructed to
perform 4 balancing tasks, including: normal standing, standing with single leg sup-
port, standing up from a chair (wooden chair, seat height 435 mm, armrests height
250 mm), and sitting down to the same chair; while the two insole systems were
collecting data simultaneously at the sampling time of 25 Hz.
After the experiment, the data obtained by the fabric insole were processed by a
connected domain filter to minimize the signal noise [6]. The data were then shown in
contour form in the GUI. The F-scan system and the fabric insole have different
number of sensing units. To facilitate the comparison, the data from the fabric insole
were interpolated to obtain a 59 (row) 21 (column) data matrix to match the number
sensing units of the F-scan. The Root Mean Square Errors (RMSEs) were calculated
between the measurement obtained by each fabric sensing unit and F-scan sensing unit.
The averaged RMSE (out of 10 participants at all the sampling time) were reported for
each balancing task.
3 Results
An example of the visualized data comparison was depicted in Fig. 2. As can be seen
in this example, the fabric insole can measure the foot pressure distribution accurately
compared with the reference measurement. The average RMSE for different balancing
tasks were reported in Table 1. The RMSE are ranged from 22.46 kPa to 38.54 kPa. In
general, the fabric insole yielded better results with smaller RMSE for the static bal-
ancing task, such as normal standing and standing at single leg, than that in the
dynamic tasks, such as the standing up and sitting down.
4 Discussion
The preliminary results suggested that this smart insole can measure the foot plantar
distribution accurately with a high resolution (i.e. 360 pixels per insole) for different
balancing tasks. Therefore, it can be potentially used to continuously assess the balance
of people during different applications.
The novelty of this system is two-fold. Firstly, the pressure sensing resolution is
enhanced by having 360 sensing units. To our knowledge, this is the highest sensing
resolution achieved by low-cost wearable insole. With such high resolution, the
pressure distribution can be monitored with more details. And the subtle information
regarding to change in foot plantar pressure can be captured. Secondly, this insole is
made by conductive fabric. This material is soft and thin, its textile is similar to regular
fabric. Therefore, this smart insole is less intrusive. Also, with the fabric material, this
332 X. Hu et al.
pressure sensing array can be easily integrated with other wearable system. Thus, it is
more versatile to be used in wearable robotics system. Besides, this fabric is low-cost.
The overall cost of this smart insole system is approximately 40 USD, which makes it
feasible to be used in home settings.
Table 1. The average RMSE for each balancing task

Tasks Average RMSE (kPa)
Standing 22.46
Single leg 24.53
Standing up 35.24
Sitting down 38.54
Fig. 2. An example of comparison between the foot plantar distribution obtained by the fabric
insole and the F-scan system
5 Conclusion
This study presents a smart insole system based on conductive fabric. This insole can
measure the foot plantar distribution accurately with a high resolution. This can benefit
the future research for the real time assessment of the foot plantar pressure. And this
system can be used during gait analysis or rehabilitation applications.
A Conductive Fabric Based Smart Insole to Measure the Foot Pressure Distribution 333
References
1. Kim, S.G., Hwangbo, G.: The effect of obstacle gait training on the plantar pressure and
contact time of elderly women. Arch. Gerontol. Geriatr. 60(3), 401–404 (2015)
2. Shu, L., Hua, T., Wang, Y., Li, Q., Feng, D.D., Tao, X.: In-shoe plantar pressure
measurement and analysis system based on fabric pressure sensing array. IEEE Trans. Inform.
Tech. Biomed. 14(3), 767–775 (2010)
3. Hu, X., Zhao, J., Peng, D., Sun, Z., Qu, X.: Estimation of foot plantar center of pressure
trajectories with low-cost instrumented insoles using an individual-specific nonlinear model.
Sensors 18(2), 421 (2018)
4. www.eeonyx.com. Accessed 18 May 2019
5. Saxena, R.S., Bhan, R.K., Aggrawal, A.: A new discrete circuit for readout of resistive sensor
arrays. Sens. Actuators A Phys. 149(1), 93–99 (2009)
6. Hu, X., Shen, F., Peng, D., Luo, C., Mo, S., Qu, X.: A portable insole for foot plantar pressure
measurement based on a pressure sensitive e-textile and voltage feedback method. A paper
submitted to IEEE-DSP2018
Training Balance Recovery in People
with Incomplete SCI Wearing a Wearable
Exoskeleton
E. H. F. van Asseldonk1(&), A. Emmens1, T. J. H. Brug1, I. Pisotta2,

M. Arquilla2, F. Tamburella2, M. Masciullo2, N. L. Tagliamonte2,
R. Valette2, M. Molinari2, and H. van der Kooij1,3
1
Enschede, The Netherlands
e.h.f.vanasseldonk@utwente.nl
2
Laboratory of Robotic Neurorehabilitation, Neurorehabilitation 1 Department,
Fondazione Santa Lucia, Rome, Italy
3
Department of Biomechanical Engineering, Delft University of Technology,
Delft, The Netherlands
Abstract. Improving stability of people wearing a lower extremity Wearable

Exoskeleton (WE) is one of the biggest challenges in the field. The goal of this
preliminary study was to improve balance recovery from perturbations in people
with incomplete Spinal Cord Injury (SCI) assisted by a WE with specifically
developed balance controller. The WE has actuated ankle and knee joints, which
were controlled by using a body sway-based balance controller. Two test pilots
participated in 5 training sessions, devoted to enhance the use of the robot, and
in pre/post assessments. Their balance during quiet standing was perturbed
through pushes in forward direction.
The controller was effective in supporting balance recovery in both tests pilots
as reflected by a smaller sway amplitude and recovery time when compared with
a minimal impedance controller. Moreover, the training resulted in a further
reduction of the sway amplitude and recovery time in one of the test pilots
whereas it had not an additional beneficial effect for the other subject.
In conclusion, the novel balance controller can effectively assist people with
incomplete SCI in maintaining standing balance and a dedicated training has the
potential to further improve balance.
1 Introduction
Nowadays, there are several commercial and research Wearable Exoskeletons

(WEs) that provide people with SCI with the necessary support to restore their ability to
walk. The WE does not yet provide the user with support on balance maintenance.
Consequently, people walking in a WE have to rely on crutches losing the possibility of
using their arms and hands for other activities of daily living like carrying a cup of
coffee. Adding balance control capabilities to WEs is currently one of biggest chal-
lenges in improving the use and functionality of WE.

https://doi.org/10.1007/978-3-030-01887-0_64
Training Balance Recovery in People with Incomplete SCI 335
Human balance control during stance is all about controlling the Center of Mass
(CoM) movements over the base of support. In a previous study [1], we have devel-
oped a balance control strategy for an ankle WE based on the CoM related body sway.
This balance control strategy generated human-like response torques to perturbations
and was effective in supporting healthy subjects in maintaining balance. Here, we will
extent this control strategy to also include stabilization of the knee joint and evaluate it
in people with an incomplete SCI. The goal of this study was to improve balance
recovery during quiet stance from perturbations in people with an incomplete SCI
through training in an ankle/knee WE with this novel balance controller.
2.1 Subjects
Two male test pilots with an incomplete SCI (AIS D) participated in this experiment.
The experiments were approved by the Ethical Committee of Fondazione Santa Lucia.
2.2 Wearable Exoskeleton

The modular Symbitron WE was used [2]. This is a torque controlled WE consisting of
8 actuation modules. As it is modular it can be configured to the patient’s needs. Since
subjects with a good voluntary control of their hip joints were enrolled, we used only
two modules for each leg to actuate knee flexion/extension and ankle
plantar/dorsiflexion. The WE is controlled using EtherCAT. The EtherCAT Master
runs on a computer in a backpack. This master communicates with the custom-made
EtherCAT slaves in each of the actuation modules. During the training, subjects did not
wear the backpack.
The WE was used to assist subjects in balance recovery from perturbations. This
was achieved by a body sway-based control strategy (PDCoM). Body sway is defined as
the angle between the line from the CoM to the ankle and the vertical. The desired
ankle torque sA;d was the same for both ankle joints and was computed from a PD
control law using the body sway hsway as input.

sAd ¼ PA hswayd hsway þ DA h_ swayd h_ sway ð1Þ
where PA and DA are proportional and derivative gains respectively and the subscript
d indicates desired.
The controller for the knee joints tries to keep the knee stretched by using a PD
control law with the knee angle as input

sKd ¼ PK /Kd /K þ DK /_ Kd /_ K ð2Þ
where /K is the knee angle, and PK and DK are proportional and derivative gains,
respectively.
336 E. H. F. van Asseldonk et al.
The gains were tuned for each test pilot in a separate session taking into account the
pilot’s feedback
2.3 Protocol
Both subjects participated in 5 training sessions and pre/post assessments. During all
the sessions, the test pilots received forward pushes while standing at the level of the
sacrum by using a push stick equipped with a force sensor (FUTEK LCM 300, FUTEK
Advanced Sensor Technology Inc, Irvine, USA). Test pilots were instructed to recover
balance without taking a step. The experimenter giving the pushes varied the push force
between 70 and 100% of the maximal push force that the test pilot could withstand. In
each training session, tests pilots received pushes for 10 min with a resting time of
about 5 s between pushes. In each assessment session, test pilots received 40 pushes
when the WE was controlled with the PDcom and 40 pushes when the WE was con-
trolled in minimal impedance mode.
2.4 Data Analysis

Body sway was calculated using the estimated orientation of the different body seg-
ments obtained using the encoders in the exoskeleton and two MTx IMUs (Xsens
Technologies B.V., Enschede, the Netherlands) placed on the right thigh and trunk.
From the body sway, two outcome variables were derived. First, the body sway
amplitude was defined as the maximal deviation in body sway in response to the push.
Second, the recovery time (r.t.) was defined as the time needed to get the body sway
velocity back below 0.015 rad/s.
3 Results
In the pre-test, we evaluated whether the use of the novel PDcom assisted in counter-
acting the perturbations. Both test pilots showed an improved ability to recover after
the pushes as reflected in a smaller r.t. and sway amplitude when the PDcom was used
compared to the case minimal impedance controller was used (see Fig. 1).
Fig. 1. Body sway angles in response to forward pushes for test pilot 1 during the pre-
assessment for two different WE controllers. Lines are averaged trajectories across pushes with
the same amplitude. The marker indicates the average recovery time (r.t.).
Training Balance Recovery in People with Incomplete SCI 337
Training for five sessions with the WE controlled with PDcom resulted in a further
improvement of the ability to withstand the push perturbations in one of test pilots (as
depicted in Fig. 2) but not in the other one (not shown here). For test pilot 1, the sway
amplitude further decreased from the pre-assessment to the post-assessment. As the
WE ankle torque was directly dependent on the sway, this decrease in sway was
accompanied by a reduction in the WE ankle torques. This indicates that the reduced
sway can be attributed to the test pilot who learned how to better counteract the
perturbation with the aid of the WE.
Fig. 2. Push force, body sway and torques exerted by the WE in response to forward pushes for
test pilot 1 during the pre and post assessments. In both tests the WE was controlled with PDcom.
Lines are averaged trajectories across pushes with the same amplitude and shaded areas indicate
the standard deviation.
4 Conclusions
The novel PDcom results in torques that are effective in assisting people with an
incomplete SCI to counteract push perturbations during quiet stance. Training with
PDcom has the potential to further improve the ability to withstand perturbations. Future
studies will focus on extending the balance controller to the hip such that we can
investigate whether people with a complete SCI can be supported during balance tasks.
Acknowledgment. SYMBITRON is supported by EU research program FP7-ICT-2013-10

(contract #611626). SYMBITRON is coordinated by University of Twente.
338 E. H. F. van Asseldonk et al.
References
1. Emmens, A., van Asseldonk, E.H.F., Kooij, H.V.D.: Effects of a powered Ankle-Foot
Orthosis on perturbed standing balance. Journal of Neuroengineering and Rehabilitation, to be
published
2. Meijneke, C., Wang, S., Sluiter, V., van der Kooij, H.: Introducing a modular, personalized
exoskeleton for ankle and knee support of individuals with a spinal cord injury. In: González-
Vargas, J., Ibáñez, J., Contreras-Vidal, J.L., van der Kooij, H., Pons, J.L., (eds.) Wearable
robotics: challenges and trends, vol. 16, no. 1, pp. 169–173. Springer International Publishing,
Cham (2016)
Modular Composition of Human Gaits
Through Locomotor Subfunctions
and Sensor-Motor-Maps
Andre Seyfarth(&), Maziar A. Sharbafi, Guoping Zhao,

and Christian Schumacher
Lauflabor Locomotion Laboratory, Technische Universität Darmstadt,

Darmstadt, Germany
seyfarth@sport.tu-darmstadt.de
Abstract. Human locomotion is a complex movement task, which can be

divided into a set of locomotor subfunctions. These subfunction comprise stance
leg function, swing leg function and balance. Each of these locomotor sub-
functions requires a specific control of individual muscles in the human body.
We propose a novel method based on sensor-motor-maps to identify the
appropriate motor control settings based on sensory feedback loops. Based on
template models, both the biomechanical as well as the neuromuscular dynamics
of gait can be studies and described at different levels of detail.
1 Introduction
Legged locomotion is a daily activity in humans, which is crucial for a good living
quality. In the case of limb impairments or even limb loss several treatments including
orthotic and prosthetic systems can help to restore locomotor function. Recently, the
German amputee long jumper Markus Rehm [1, 2] succeeded to outperform non-
amputee long jumper and won the German championship using a highly elastic carbon
fiber Cheetah prosthesis.
A prosthetic limb can be considered as a model for the substituted leg function. In
the case of Rehm, lower leg function was replaced by a highly elastic carbon fiber
spring. The comparable performance of Markus Rehm and non-amputee elite jumpers
shows the role of elastic stance leg function in the take-off phase of long jump.
2 Locomotor Subfunctions
We consider legged locomotion to be composed of three locomotor subfunctions. The

first subfunction is stance leg function, comprising axial loading and unloading of the
limb. This leg function can be described by a prismatic spring and is represented by the
spring-loaded inverted pendulum model [3]. In this highly simplified template model,
regions for stable walking and running can be predicted depending on the forward
speed, leg stiffness and leg angle of attack. The second subfunction is the swing leg

https://doi.org/10.1007/978-3-030-01887-0_65
340 A. Seyfarth et al.
function. Here, the leg realigns for the next touch-down. This rotational leg adjustment
can be described by the action of thigh muscles acting like antagonistic springs [4].
The third locomotor subfunction is balance. Here, the goal is to keep the body
orientation aligned upward. This can be achieved based on vestibular sensory feedback
providing a reference for body posture in space. An alternative way for maintaining
postural balance was suggested by Maus et al. [5]. Here the leg forces are aligned to
point to a virtual pivot point (VPP). During human walking, the ground reaction forces
intersect at this specific point, which is located in the upper body above the center of
mass. This VPP concept can be easily implemented in biomechanical models for
walking and running. It was shown to provide postural balance during locomotion.
Recently, this concept was transferred into the FMCH concept [6], which assumes leg
force-modulated compliant hip muscles to achieve balance. This was implemented in a
model, which predicted similar hip torque patterns as previously obtained with the VPP
model [5]. In human walking, such a linear dependency of hip torques normalized to leg
force on the hip joint angle (Fig. 1) was found for a large range of walking speeds [7].
Fig. 1. Hip torques normalized to leg force as a function of hip angle during different walking
speeds (25 to 125% of preferred transition speed, PTS, between walking to running gait). The
predictions of the FMCH model match well the experimental data.
3 Sensor-Motor-Maps
In the last section, locomotor subfunctions were described on the mechanical level
using physical leg, joint or muscle parameters. In the next step, we shift our focus to the
neuromuscular description of the axial stance leg function. However, also for the other
locomotor subfunction (balance, swing leg function) matching neuromuscular models
can be identified. This research is, however, not yet completed.
Modular Composition of Human Gaits 341
The existence of biomechanical template models (e.g. SLIP, VPP) may suggest
matching low-dimensional neuro-muscular control models. Then the neural system
could take full benefit of the dimension reduction provided on the mechanical level.
Assuming the leg spring model to represent the stance leg function, a matching neural
control structure could be a reflex pathway based on sensory information (e.g. from
muscle spindles or Golgi organs). The leg spring is described by a few parameters (leg
stiffness, rest length of the unloaded spring). Similarly, reflex pathways can be char-
acterized by reflex gain and offset. In both cases we find a similar algebraic structure.
For walking and running, the SLIP model predicts different combinations of leg
stiffness and angle of attack to result in stable locomotion. Similarly, we can now study
combinations of different reflex pathway parameters to achieve a continuous series of
rebounds like in human hopping (or running). As different reflex pathways may lead to
stable hopping, we consider a blending scheme integrating muscle force, length and
velocity feedback with optimized gains and offset values [8]. All stable solutions can be
represented in the so-called sensor-motor-map (Fig. 2B) indicating compact regions of
appropriate blended feedback pathway combinations. The corners represent ‘pure’
muscle force, length and velocity feedback, respectively.
Fig. 2. The sensor-motor map (B) represents possible combinations of reflex pathways based on
proprioceptive muscle sensory information (fiber length L, fibre velocity V and muscle force F).
Here, one monoarticular extensor muscle is considered during hopping (A). The muscle is
modeled as a serial arrangement of the contractile element (CE) and the serial elastic element
(SE). The reflex signals from the different sensors are superimposed, time delayed and fed back
to stimulate the muscle together with the prestimulation signal PreStim.
Similar to the region for stable running regarding leg stiffness and leg angle of
attack, also on the neuro-muscular level different combinations of the three blended
reflex pathways can be used to predict energetically stable hopping. Please note that the
biomechanical SLIP model is energetically neutral, hence it cannot be used to address
energy stability during locomotion.
The analysis of the neuromuscular hopping model reveals that stable hopping requires
dominant force or length feedback pathways. The force feedback pathway is able to
optimize hopping performance (i.e. maximum height, [9]) whereas length feedback
increases hopping efficiency (i.e. metabolic costs in relation to hopping height). Velocity
342 A. Seyfarth et al.
feedback is disabling hopping. Already a moderate contribution of the velocity reflex is

sufficient to interrupt the cyclic movement transferring the system into a resting config-
uration. This establishes a safety measure as it enables a sudden controlled stopping of the
movement.
The region for stable hopping found in the sensor-motor map is robust with respect
to morphological changes, such as changed body mass, segment lengths and tendon or
ground compliance. This indicates that the task-specific selection of sensory reflex
pathways is universal and not much dependent on specific system properties.
4 Conclusion
In this paper we describe the modular organization of the biomechanical and neuro-
muscular system during legged locomotion with the help of locomotor subfunctions
and sensor-motor-maps. The underlying assumption is that complex movement tasks
can be decomposed into a set of elementary subfunctions. Each of these subfunctions
requires an appropriate matching blending of the neuromuscular reflex pathways,
which can be represented with the help of sensor-motor-maps. We described the
biomechanical template models for two locomotor subfunctions (stance and balance)
and the corresponding sensor-motor-map for stance leg function during hopping. To
finally approve these concepts, all locomotor subfunctions need to be represented on
both the biomechanical and neuromuscular level. It is required to prove that all sub-
functions can work in parallel without losing their ability to fulfill their specific tasks.
Finally, these concepts need to be applied to a robot testbed to prove their validity in
real world. After successful approval, a transfer to assistive systems such as prosthetic
or orthotic systems becomes feasible.
Acknowledgment. This research was supported by the EU FP7 project BALANCE and by the
DFG grants awarded to A.S. and M.A.S. (SE1042/6, SE1042/8, SE1042/31 and AH 307/2-1).
References
1. Deutschen Meisterschaften Ulm, 26 June 2014
2. Willwacher, S., Funken, J., Heinrich, K., Müller, R., Hobara, H., Grabowski, A.M.,
Brüggemann, G.P., Potthast, W.: Elite long jumpers with below the knee prostheses approach
the board slower, but take-off more effectively than non-amputee athletes. Sci. Rep. 7(1),
16058 (2017)
3. Geyer, H., Seyfarth, A., Blickhan, R.: Compliant leg behaviour explains basic dynamics of 28
walking and running. Proc. R. Soc. B 273(1603), 2861–2867 (2006)
4. Sharbafi, M.A., et al.: Reconstruction of human swing leg motion with passive biarticular
muscle models. Hum. Mov. Sci. 52, 96–107 (2017)
5. Maus et al.: Nature Communications (2010)
6. Sharbafi, M.A., Seyfarth, A.: Fmch: a new model for human-like postural control in walking.
In: IEEE/RSJ International Conference on Intelligent Robots and Systems, pp. 5742–5747
(2015)
Modular Composition of Human Gaits 343
7. Sharbafi, M.A., Seyfarth, A.: How locomotion sub-functions can control walking at different
speeds? J. Biomech. (2017)
8. Schumacher, C., Seyfarth, A.: Sensor-motor maps for describing linear reflex composition in
hopping. Front. Comput. Neurosci. 11, 108 (2017)
9. Geyer, H., Seyfarth, A., Blickhan, R.: Positive force feedback in bouncing gaits? Proc. R. Soc.
Lond. B Biol. Sci. 270(1529), 2173–2183 (2003)
Model-Based Posture Control
for a Torque-Controlled Humanoid Robot
Maximo A. Roa(&), Bernd Henze, and Christian Ott
Institute of Robotics and Mechatronics, German Aerospace Center (DLR),

82234 Wessling, Germany
{maximo.roa,bernd.henze,christian.ott}@dlr.de
Abstract. This talk presents an overview of the development of a full-body

model-based passivity approach for posture control of a humanoid robot. The
controller exploits passivity properties to provide suitable control inputs for the
humanoid robot without requiring a solution to the inverse kinematics problem
of the full kinematic chain. The controller has been validated in numerous
experiments using the torque-controlled humanoid robot TORO, developed at
the German Aerospace Center (DLR).
1 Introduction
Humanoid robots have been developed for mimicking human appearance and capa-
bilities, but at the same time, they have served to validate biological or neurological
findings of the human body in an artificial testbed. Practical applications of humanoid
robots include replacing humans in a large variety of tasks that entail dull, dirty or
dangerous scenarios. Controlling such robots is a challenging task, mainly due to the
large number of degrees of freedom involved, and to the paramount requirement of
keeping the balance while performing some useful task.
Different approaches have been proposed for providing whole-body and balance
control, mainly (1) solving the inverse kinematics and dynamics problem for the full
robot, or (2) exploiting passivity-based approaches that avoid the explicit solution of
the inverse dynamics problem. This talk is centered on a passivity-based approach to
control the full posture, i.e. position and orientation, of a humanoid robot. It provides
an overview of the theoretical framework and validation experiments previously
published at different venues [4, 5, 7].
2 Balancing Controller
Passivity-based compliance controllers based on joint torque sensing have been tra-
ditionally applied to manipulation tasks [1, 2]. The extension of such controllers to the
problem of balancing a humanoid robot was first proposed in [3], by controlling the
This work was partially supported by the European Commission under grant H2020-ICT-645097,
project COMANOID.

https://doi.org/10.1007/978-3-030-01887-0_66
Model-Based Posture Control for a Torque-Controlled Humanoid Robot 345
position of the center of mass (CoM) of the robot. These ideas were later extended in
[4] to control the full posture of the robot, by exploiting the observation that the
problems of grasping an object and balancing a robot are fundamentally similar. For
pure balancing, the robot uses all the end-effectors in contact with the environment to
stabilize a desired configuration through the application of suitable forces at the end-
effectors. The framework can be extended for multi-contact scenarios, where some of
the end-effectors are used to perform an intended task (e.g. manipulation), and the
perturbation created due to such task must then be considered by the balancing
controller.
In general, given a desired equilibrium rd for the CoM position r and a desired
orientation Rd of the trunk, the required balancing wrench Fd applied at the CoM of the
robot is computed according to a compliance control law given by

mg r_ Kt ðr rd Þ
Fd ¼ D ð1Þ
0 x sk ðR; Rd Þ
where R and ɷ denote the orientation and the angular velocity of the trunk, usually
measured by an onboard IMU. The matrix D indicates a damping term. Kt and sk
denote the translational stiffness and the torque from a virtual rotational spring between
R and Rd.
This required wrench Fd is distributed to the supporting contacts. The mapping P
between the contact wrenches Fi and the total wrench Fd is obtained as Fd ¼ Gi Fi .
The desired contact wrenches Fi are then computed based on a constrained mini-

mization of the cost function Fi Fidef , which minimizes the error between the
contact wrenches Fi and a default wrench distribution Fidef . Unilateral constraints on the
contact forces, friction constraints, and the Center of Pressure (CoP) being limited to
the contact area can be considered.
The desired contact wrenches Fc can be realized through corresponding joint tor-
ques sd. An exact implementation of this mapping requires the consideration of the full
multi-body robot dynamics. However, a simpler solution uses a kinetostatic mapping
based on the relevant Jacobian matrices, which can be derived by considering the
interaction of the isolated CoM dynamics with the remaining multi-body dynamics.
This leads to
X
sd ¼ JiT Fi ð2Þ
i
where the Ji correspond to the Jacobian matrices for the end-effectors with respect to
the CoM location r and hip orientation R.
The proposed control approach does not require any measurement of the contact
forces at the feet. The estimation of the state of the CoM is obtained using information
coming from internal sensors, namely an onboard IMU and the kinematic information
of the robot, which makes the controller robust against uncertainties in the ground
geometry [5].
346 M. A. Roa et al.
3 Experimental Validation
Different experiments have been carried out to test the proposed controller, using the
torque-controlled humanoid robot TORO. In the current version, TORO has 27 DoF
(excluding the hands), a height of 1.74 m and a weight of 76 kg [6]. Figure 1 shows
some of the scenarios that have demonstrated the capabilities of the controller,
including:
• Pure 3D balancing while resisting external perturbations
• Multi-contact control
• Behavior in front of tilt disturbance and translational acceleration in the supporting
surface
• Behavior on a compliant support surface
Fig. 1. Different scenarios for validation of the passivity-based balancing controller using the
DLR humanoid robot TORO.
The experiments confirm the benefits of compliant control based on torque sensing.
The passivity-based controller has been used to compare bio-inspired and model-based
approaches for posture control [7]. It has lately been combined with hierarchical multi-
task control, which not only allows for dynamic decoupling but also to handle multiple
contact points distributed across the entire body of the robot [8, 9].
Model-Based Posture Control for a Torque-Controlled Humanoid Robot 347
References
1. Ott, C.: Cartesian Impedance Control of Redundant and Flexible-Joint Robots. Springer
Tracts in Advanced Robotics 49, Springer (2008)
2. Dietrich, A., Wimböck, T., Albu-Schäffer, A.: Dynamic whole-body mobile manipulation
with a torque controlled humanoid robot via impedance control laws. In: IEEE/RSJ
International Conference on Intelligent Robots and Systems – IROS 2011, pp. 3199–3206
(2011)
3. Hyon, S., Hale, J., Cheng, G.: Full-body compliant human-humanoid interaction: balancing in
the presence of unknown external forces. IEEE Trans. Robot. 23(5), 884–898 (2007)
4. Ott, Ch., Roa, M.A., Hirzinger, G.: Posture and balance control for biped robots based on
contact force optimization. In: IEEE-RAS International Conference on Humanoid Robots
2011, pp. 26–33 (2011)
5. Henze, B., Roa, M.A., Ott, C.: Passivity-based whole-body balancing for torque-controlled
humanoid robots in multi-contact scenarios. Int. J. Robot. Res. 35(12), 1522–1543 (2016)
6. Engelsberger, J., Werner, A., Ott, C., Henze, B., Roa, M.A., Garofalo, G., Burger, R., Beyer,
A., Eiberger, O., Schmid, K., Albu-Schäffer, A.: Overview of the torque-controlled humanoid
robot TORO. In: IEEE-RAS International Conference on Humanoid Robots 2014, pp. 916–
923 (2014)
7. Ott, C., Henze, B., Hettich, G., Seyde, T.N., Roa, M.A., Lippi, V., Mergner, T.: Good
Posture, good balance: comparison of bioinspired and model-based approaches for posture
control of humanoid robots. IEEE Robot. Autom. Mag. 23(1), 22–33 (2016)
8. Henze, B., Dietrich, A., Roa, M.A., Ott, C.: Multi-contact balancing of humanoid robots in
confined spaces: utilizing knee contacts. In: IEEE/RSJ International Conference on Intelligent
Robots and Systems – IROS 2017, pp. 697–704 (2017)
9. Henze, B., Dietrich, A., Ott, C.: An approach to combine balancing with hierarchical whole-
body control for legged humanoid robots. IEEE Robot. Autom. Lett. 1(2), 700–707 (2016)
Exoskeleton Research in Europe
XoSoft - Iterative Design of a Modular
Soft Lower Limb Exoskeleton
Jesús Ortiz(B) , Christian Di Natali, and Darwin G. Caldwell

via Morego, 30, 16163 Genova, Italy
jesus.ortiz@iit.it
Abstract. XoSoft is a modular soft lower-limb exoskeleton to assist

people with mobility impairments. Being a modular system, it com-
prises of ankle, knee and hip elements, which can be used in different
configurations. XoSoft follows a user centered design strategy achieved
by involving primary, secondary and tertiary end users as participatory
stakeholders in the design and development process. This paper presents
the evolution of the different prototypes developed during the project, as
well as the testing stages. From the Alpha prototype, built from available
technologies, to the Gamma prototype, which includes advanced textiles
technologies, smart materials for sensing and actuation, biomimetic con-
trol and connected health monitoring and feedback system.
1 Introduction
The proportion of the world’s elderly population is expected to reach 2 billions
by 2050 [1]. Many elderly experience varying degrees of mobility impairment,
due to decline in voluntary muscle strength. For this reason, remaining active
and mobile during ageing is crucial to overall health and cognitive function
[2]. Consequently, assistive devices play a pivotal role in their lives and impact
on their ability to live independently and perform basic tasks of daily living.
However, many assistive aids, such as powered wheel chairs, do not encourage
or support activation of legs.
Mobility assistance is also required by patients, such as stroke sufferers or
patients with incomplete Spinal Cord Injuries (SCI). Globally circa 16 million
people per year experience a stroke for the first time, of which 5 million expe-
rience varying degrees of mobility difficulty, which significantly impacts their
ability to perform tasks of daily living. SCI lesions are mostly caused by acci-
dents, of which about the 51% are incomplete, i.e. the person is partially disabled
[3]. Patients with an incomplete SCI do not suffer complete loss of sensory-motor
function in the lower limbs but they may still require assistance to walk.
XoSoft is an EU project that is currently developing a soft lower-limb
exoskeleton to assist people with mobility restrictions due a partial loss of sen-
sory or motor function. Typically, the existing exoskeletons have a rigid structure
This work has received funding from the European Union’s Horizon 2020 framework
programme for research and innovation under grant agreement No. 688175.
https://doi.org/10.1007/978-3-030-01887-0_67
352 J. Ortiz et al.
that is heavy weight and bulky [4]. Following current trends [5,6], XoSoft has
a flexible and adaptable structure [7]. The proposed system is not intended to
substitute complete loss of function like already existing exoskeletons, but rather
assist the user in a tailored manner. In order to reach this goal, the prototypes
are based on smart soft robotics, biomimetic controlled actuation and connected
health data feedback and interface.
Fig. 1. XoSoft concept, evolution of the prototypes and testing stages during the
project, following the UCD approach.
2 System Development
2.1 User Centered Design
A core feature of XoSoft is that it follows a User Centered Design (UCD) app-
roach [8] as shown in Fig. 1. UCD employs design ethnography and participatory
stakeholder involvement as key drivers for the technology development to ensure
user needs are at the forefront XoSoft’s development [9]. As a starting point for
the technical developments, an Alpha prototype was built using off the shelf com-
ponents. This prototype consisted in a series of independent modules divided in
two categories: hard and soft. All the modules were developed with the main pur-
pose of understanding the requirements of the following prototypes, simulating
or replicating the desired functionalities using alternative available technologies.
The Alpha prototype lied the foundation for the development of three versions
of XoSoft, which are identifiable based on the module and subsystem developed,
as described in the following subsections.
XoSoft - Iterative Design of a Modular Soft Lower Limb Exoskeleton 353
2.2 Beta 1 Prototype

The Beta 1 prototype was the first full prototype using novel technologies and
actuation principles. Regarding the actuation, different technologies were eval-
uated: (i) variable stiffness based on Textile Jamming (TJ) [10]; and (ii) quasi-
passive actuation based on elastic bands and clutches. The quasi-passive actu-
ation was implemented using commercial electromagnetic clutches transmitting
the motion to the elastic bands through a bowden cable. The selection of the
actuation arrangement was based on the optimization method described in [11],
including actuation for the hip and knee flexion only in a single mode config-
uration. This prototype included a first version of the textile based knee brace
using soft capacitive sensors [12] for the measurement of the joint angle and a
sensorized insole with embedded FSR sensors for the identification of the walking
gait. The garment was built using commercial textiles with inelastic reinforce-
ment bands to transmit and distribute the assistive forces.
2.3 Beta 2 Prototype

The main technological upgrade of the Beta 2 prototype was the substitution
of the commercial electromagnetic clutches by novel custom made soft clutches
based on similar pneumatic principles of TJ, but functioning as a linear clutch.
They were integrated in series with the elastic bands providing assistance to the
hip (flexion and extension), knee (flexion and extension) and ankle (dorsi and
plantar flexion). The modular design of this prototype allowed to reconfigure
the system to accomodate different patient profiles, allowing a maximum of 8
simultaneous actuators (clutches). This prototype was also equipped with IMU
sensors for the measurement of the joint angles, besides the insole FSR sensors
and soft capacitive sensors already integrated in the Beta 1 prototype.
2.4 Gamma Prototype

The final prototype developed in the project is the Gamma prototype and it
includes all the features of the previous prototypes. Most of the components
have been improved with respect to the Beta 2 prototype, including a completely
new design of the garment, which improves significantly the wearability. The
actuation has been limited to 6 simultaneous actuators in order to reduce the
weight and energy requirements. The control is able to work with both the
capacitive soft sensors integrated in the garment and the IMU sensors. Finally,
the monitoring and feedback system, and the offline activity and task recognition
are being integrated into the system.
3 Testing and Validation

3.1 Laboratory Testing
The testing protocol defined at the early stages of the project was executed
during the testing of the Beta 1, Beta 2 and Gamma prototypes in a laboratory
354 J. Ortiz et al.
environment. The purpose of this feasibility study was to understand the impact
of the exoskeleton on different biomechanical aspects (such as gait improvement)
and overall system performance and ergonomic aspects. The Beta 1 prototype
was tested with one post-stroke patient, the Beta 2 with four subjects with
different conditions, and the Gamma will be tested with at least four subjects.
3.2 Clinical Testing
The Beta 2 prototype is being tested in a clinical environment with elderly

people. The main purpose of this testing is to gather information about the
use of the system in a rehabilitation environment and to collect feedback from
primary users. The number of participants is estimated to be between 10 and
15.
3.3 Home-Simulated Testing
Finally, the Gamma prototype will be tested in a home simulated environment

to understand the performance of the system in daily live activities. Between 10
and 15 incomplete SCI and stroke patients will participate in the trials.
4 Results and Conclusions
This paper describes the different prototype stages of the XoSoft project. Each
prototype presents a different evolution of the technologies developed during
the project. Each prototype is tested in laboratory conditions, including a val-
idation of the final prototypes in clinical and home simulated environments.
Preliminary results of the Beta 1 laboratory testing with a post-stroke patient
[13] show that the proposed system provides a positive assistance to the users
in terms of gait performance during straight walking. Further studies are being
conducted in order to quantify the effects with different subjects in several tasks
and conditions.
References
1. World Health Organization. Facts about ageing (2014). http://www.who.int/
ageing/about/facts/en/
2. Volkers, K.M., de Kieviet, J.F., Wittingen, H.P., Scherder, E.J.A.: Lower limb
muscle strength (LLMS): why sedentary life should never start? A review. Arch.
Gerontol. Geriatr. 54(3), 399–414 (2012)
3. National Spinal Cord Injury Statistical Center, et al.: Annual Statistical Report-
Complete Public Version. University of Alabama, Tuscaloosa (2013)
4. Yan, T., Cempini, M., Oddo, C.M., Vitiello, N.: Review of assistive strategies in
powered lower-limb orthoses and exoskeletons. Robot. Auton. Syst. 64, 120–136
(2015)
XoSoft - Iterative Design of a Modular Soft Lower Limb Exoskeleton 355
5. Awad, L.N., Bae, J., Odonnell, K., De Rossi, S.M., Hendron, K., Sloot, L.H.,
Kudzia, P., Allen, S., Holt, K.G., Ellis, T.D., et al.: A soft robotic exosuit improves
walking in patients after stroke. Sci. Transl. Med. 9(400), eaai9084 (2017)
6. Schmidt, K., Duarte, J.E., Grimmer, M., Sancho-Puchades, A., Wei, H., Easthope,
C.S., Riener, R.: The myosuit: bi-articular anti-gravity exosuit that reduces hip
extensor activity in sitting transfers. Front. Neurorobot. 11, 57 (2017)
7. Ortiz, J., Rocon, E., Power, V., de Eyto, A., OSullivan, L., Wirz, M., Bauer,
C., Schülein, S., Stadler, K.S., Mazzolai, B., Teeuw, W.B., Baten, C., Nikamp,
C., Buurke, J., Thorsteinsson, F., Müller, J.: Xosoft-a vision for a soft modular
lower limb exoskeleton. In: Wearable Robotics: Challenges and Trends, pp. 83–88.
Springer (2017)
8. Sanders, E.B.-N.: From user-centered to participatory design approaches. In:
Design and the Social Sciences, pp. 18–25. CRC Press (2003)
9. Power, V., O’Sullivan, L., de Eyto, A., Schülein, S., Nikamp, C., Bauer, C., Müller,
J., Ortiz, J.: Exploring user requirements for a lower body soft exoskeleton to assist
mobility. In: Proceedings of the 9th ACM International Conference on PErvasive
Technologies Related to Assistive Environments, p. 69. ACM (2016)
10. Brown, E., Rodenberg, N., Amend, J., Mozeika, A., Steltz, E., Zakin, M.R., Lipson,
H., Jaeger, H.M.: Universal robotic gripper based on the jamming of granular
material. Proc. Natl. Acad. Sci. 107(44), 18-809–18-814 (2010)
11. Ortiz, J., Poliero, T., Cairoli, G., Graf, E., Caldwell, D.G.: Energy efficiency anal-
ysis and design optimization of an actuation system in a soft modular lower limb
exoskeleton. IEEE Robot. Autom. Lett. 3(1), 484–491 (2018)
12. Totaro, M., Poliero, T., Mondini, A., Lucarotti, C., Cairoli, G., Ortiz, J., Beccai,
L.: Soft smart garments for lower limb joint position analysis. Sensors 17(10), 2314
(2017)
13. Poliero, T., Di Natali, C., Sposito, M., Ortiz, J., Graf, E., Pauli, C., Bottenberg,
E., de Eyto, A., Caldwell, D.G.: Soft wearable device for lower limb assistance:
assessment of an optimized energy efficient actuation prototype. In: IEEE-RAS
International Conference on Soft Robotics (RoboSoft). IEEE (2018, Accepted)
Preliminary Usability and Efficacy Tests
in Neurological Patients of an Exoskeleton
for Upper-Limb Weight Support
M. Caimmi1(&), I. Carpinella2, R. Di Giovanni3, D. Ellena3,

L. Molinari Tosatti1, D. Cattaneo2, M. Ferrarin2, and C. Solaro3
1
Institute of Industrial Technologies and Automation,
Italian National Research Council, Rome, Italy
marco.caimmi@itia.cnr.it
2
Don Carlo Gnocchi Foundation, Milan, Italy
3
C.R.R.F. Monsignor Luigi Novarese, Moncrivello, Italy
Abstract. Upper-limb paresis is a main disabling condition in stroke and

neurological diseases and rehabilitation is essential for recovering/maintaining
function. Upper-limb weight support may help/enable these patients performing
movements against gravity thus allowing for task oriented interventions. In this
framework, an exoskeleton for upper-limb weight support was developed. In
this preliminary study the system was tested in a small group of neurological
patients (N = 12) to verify the overall usability and its efficacy in assisting
patients during functional movements against gravity. Patients performed some
functional tasks of the ARAT test both with and without the exoskeleton. The
system seems effective as it enabled even the most impaired patients performing
the tasks. All patients could wear the exoskeleton and complete the tasks.
Usability of the system was assessed as adequate for a use inside a clinical
study. Future work will focus on verifying the efficacy of task-oriented inter-
vention performed using the exoskeleton.
1 Introduction
An adequate upper-limb function is essential to interact with the environment. Upper-

limb impairments in patients with stroke and other neurological diseases, such as
Multiple Sclerosis (MS), often lead to reduced manual dexterity and difficulties in
carrying out Activities of Daily Living (ADLs) [1, 2]. Pilot studies have been per-
formed to evaluate the efficacy of training with weight support of the upper-limb using
a passive exoskeleton and preliminary results showed improved upper-limb function in
both stroke [3] and MS patients [4]. In both studies, patients performed several ADLs
tasks in a digital environment with the assistance of an exoskeleton passively sup-
porting the upper-limb weight. There is also mounting evidence that functional and
task-oriented training using real objects is effective in upper-limb rehabilitation [5, 6].
Unfortunately, highly impaired patients are not able to perform movements against
gravity and their possibility to perform task-oriented therapy is limited. It is important
to verify whether weight supporting of the arm would enable these patients to perform

https://doi.org/10.1007/978-3-030-01887-0_68
Preliminary Usability and Efficacy Tests in Neurological Patients 357
movements against gravity thus allowing the administration of interventions based on

real functional tasks. In this framework, LIGHTarm, an exoskeleton supporting pas-
sively the upper-limb weight was developed.
This study aims to test the overall usability of the system and its efficacy in
assisting patients during functional movements against gravity. The paper is organized
as follows: Sect. 2 presents the exoskeleton kinematics and weigh support system
along with the method used to test its usability; Sect. 3 reports the main results; Sect. 4
reports the discussion and the main conclusions drawn.
2 Materials and Method
2.1 LIGHTarm
The kinematics and gravity compensation mechanism of LIGHTarm along with a
rendering and a picture are shown in Fig. 1. The exoskeleton allows for the perfor-
mance of complex functional movements because of its particular kinematics (deeply
described in [7, 8]) that complies with the movements of the shoulder and elbow.
Equally important, the hand is not constraint and is free of manipulating objects in
ADLs tasks. The gravity compensation mechanism is made of a counterbalance Ws and
a spring s, which balance the weight of the upper arm and forearm, respectively as
shown in Fig. 1.
Fig. 1. Kinematics and gravity balancing scheme (left panel), LIGHTarm rendering (central
panel), and a subject wearing 3 IMUS (orange boxes) performing the tests (right panel).
2.2 Participants
Stroke and MS patients were recruited among inpatients at the Department of
Rehabilitation Mons. Luigi Novarese, Moncrivello (VC) and tested in 4 nonconsecu-
tive days. All patients presented shoulder flexion weakness different levels of upper-
limb impairment from mild to severe.
2.3 Testing Procedure and Assessment

The testing session consisted in the execution of the Action Research Arm Test
(ARAT) [9] with and without exoskeleton. ARAT is a 19 item measure divided into 4
sub-tests (grasp, grip, pinch, and gross arm movement) to assess upper-limb
358 M. Caimmi et al.
functioning. Each item is timed (At) and scored As (0–3 points, 3 = best score) on the
success of the execution and the quality of the movement. To simplify the protocol,
with the exoskeleton patients performed only 5 representative actions of the ARAT,
namely grasp a wood cube with (i) a 2.5 cm edge and (ii) a 5 cm edge, (iii), pinch a ball
bearing with 2nd finger and thumb, and move these objects from the table and place
them on a plane placed at shoulder height and at a distance equal to the sum of the
lengths of the arms and forearm; starting with the hand on the table (iv) move the hand
to the mouth, and (v) bring the hand above the head. This subtest is hereafter in the text
referred to as ARAT5. All the patients were filmed during the execution of the tests.
The performance with and without LIGHTarm of ARAT5 tasks was instrumentally
assessed using 3 IMUs placed on the chest, upper arm, and arm (see Fig. 1). The
vertical displacement of the wrist was estimated following a 3-link kinematic chain
model based on the orientation of the 3 IMUs and scaled on each subject’s anthro-
pometry [10]. To calibrate the system and have reference data, the therapist adminis-
tering the tests, performed the five tasks displacing himself the hand of each patient
(passive trials).
All tests were administered by two occupational therapists, who completed the
System Usability Scale (SUS, 0–100 points) questionnaire at the end of the study;
usability is considered to be “poor” for SUS score between 0–50, “ok” for SUS score
between 50–74, and “good” for SUS score > 70.
2.4 Outcome Measures

Differences, with and without LIGHTarm, in ARAT5 (sum of scores of the 5 items),
ARAT5-time (execution time averaged over the 5 trials), and Zw (wrist maximum
vertical displacement calculated as percentage of ones in the passive trials and averaged
over the 5 trials).
3 Results
A group of 12 patients (Gr1) with upper-limb impairment (4 MS and 8 subacute stroke,

58 ± 15 year old, 8 males, 11 left more affected side, 2.8 ± 1 months from stroke)
were selected and participated to the tests. All selected patients were able to wear
LIGHTarm and complete the testing session. Six patients were highly functioning
(ARAT > 51) and were able to perform the test even without LIGHTarm. Therefore,
the data of a subgroup made of the 6 most impaired patients (Gr2) were analyzed even
separately. For those patients it was estimated, by viewing the videos and considering
the similarity among ARAT5 tasks and some other of the ARAT, how much could
have been the total ARAT score in the case LIGHTarm would have been used. Results
are reported in Table 1.
Results of the usability evaluation made by the two occupational therapists were
SUS = 30/100 and SUS = 52/100.
Preliminary Usability and Efficacy Tests in Neurological Patients 359
Table 1. Results of gr1 and gr2

Gr1 without Gr1 with Gr2 without Gr2 with
ARAT (max 57pt) 34 ± 22 – 15 ± 11 22 ± 15
ARAT5 (max 15pt) 9.5 ± 5.7 11.5 ± 4.5 5.2 ± 4.1 8.5. ± 4.2
ARAT5-time (s) 8.1 ± 6.0 8.6 ± 4.1 12.1 ± 6.5 11.0 ± 4.7
Zw (%) – – 53 ± 33 75 ± 24
Note: Total ARAT score of Gr2 with exoskeleton was estimated.
It is worth mentioning, that one stroke patient could pinch the ball bearing with the
thumb and the index fingers only using LIGHTarm, and one MS patient could perform
multiple consecutive hand-to-mouth and reaching movements with LIGHTarm she was
not able to perform without support due to fatigue.
4 Conclusions
Overall, the system was effective in supporting the arm in the most impaired patients as
demonstrated by increased wrist vertical averaged displacement (+22%), which
reflected in increased ARAT5 score (+3.3/15) and ARAT estimated score (+7/57). By
contrast, some limitations in the system transparency were found as shown by
increased execution time in the whole group (+0.5 s).
The usability was assessed as “poor/low” because of two main reasons: (i) time to
adjust the system for the patient was considered too long; (ii) evaluators felt they need
some technical support to setup the system. However, the system was evaluated as to
be used in the routine clinical practice. Although, the preparation time is not compatible
with the clinical routine, it is acceptable for using the system in a clinical study. In
addition, above matters can be partially overcome with a short technical training and
some practice.
In conclusion, this short study demonstrated the efficacy of the system in
helping/enabling highly impaired patients performing (multiple) movements against
gravity. This would allow subacute stroke patients to access task-oriented interventions
in advance. It would be interesting to verify the efficacy of such interventions in stroke
and MS to respectively improve and maintain function.
References
1. Millan, M., Davalos, A.: The need for new therapies for acute ischemic stroke. Cerebrovasc.
Dis. 22(Suppl 1), 3–9 (2006)
2. Bertoni, R., Lamers, I., Chen, C., Feys, P., Cattaneo, D.: Unilateral and bilateral upper limb
dysfunction at body functions, activity and participation levels in people with MS. Mult.
Scler. 21(12), 1566 (2015)
3. Taveggia, G., Borboni, A., Salvi, L., et al.: Efficacy of robot-assisted rehab for the functional
recovery of the upper limb in post-stroke patients: a randomized controlled study. Eur.
J. Phys. Rehabil. Med. 52, 767 (2016)
360 M. Caimmi et al.
4. Gijbels, D., Lamers, I., Kerkhofs, L., Alders, G., Knippenberg, E., Feys, P.: The Armeo
Spring as training tool to improve upper limb functionality in multiple sclerosis: a pilot
study. J. Neuroeng. Rehabil. 8, 5 (2011)
5. Choi, Y., Gordon, J., Park, H., Schweighofer, N.: Feasibility of the adaptive and automatic
presentation of tasks system for rehabilitation of upper extremity function post-stroke.
J. Neuroeng. Rehabil. 8, 42 (2011)
6. Carpinella, I., Cattaneo, D., Bertoni, R., Ferrarin, M.: Robot training of upper limb in
multiple sclerosis: comparing protocols with or without manipulative task. IEEE Trans.
Neural Syst. Rehabil. Eng. 20, 351 (2012)
7. Scano, A., Spagnuolo, G., Caimmi, M., Chiavenna, A., Malosio, M., Legnani, G., Tosatti, L.
M.: Static and dynamic characterization of the LIGHTarm exoskeleton for rehabilitation. In:
IEEE ICORR (2015)
8. Spagnuolo, G., Malosio, M., Scano, A., Caimmi, M., Legnani, G., et al.: Passive and active
gravity-compensation of LIGHTarm, an exoskeleton for the upper-limb rehabilitation. In:
IEEE ICORR (2015)
9. Carroll, D.: A quantitative test of upper extremity function. J. Chronic Dis. 18, 479–491
(1965)
10. Pérez, R., Costa, Ú., Torrent, M., Solana, J., Opisso, E., et al.: Upper limb portable motion
analysis system based on inertial technology for neurorehabilitation purposes. Sensors
(Basel) 10, 10733–10751 (2010)
11. Brooke, J.: SUS: a “quick and dirty” usability scale”. In: Jordan, P.W., Thomas, B.,
Weerdmeester, B.A., McClelland, A.L. (eds.) Usability Evaluation In Industry (1996)
Symbitron: Symbiotic Man-Machine
Interactions in Wearable Exoskeletons
to Enhance Mobility for Paraplegics
Herman van der Kooij1(&), Edwin van Asseldonk1, Gijs van Oort1,
Victor Sluiter1, Amber Emmens1, Heide Witteveen1,
Nevio Luigi Tagliamonte2, Federica Tamburella2, Iolanda Pisotta2,
Marcella Masciullo2, Matteo Arquilla2, Marco Molinari2, Amy Wu4,
Auke Ijspeert4, Florin Florin Dzeladini4, Freygardur Thorsteinsson3,
Arash Arami5, Etienne Burdet5, Hsien-Yung Huang5,
Wouter Gregoor6, and Cor Meijneke6
1
Enschede, The Netherlands
h.vanderkooij@utwente.nl
2
Laboratory of Robotic Neurorehabilitation, Neurorehabilitation Department,
Fondazione Santa Lucia IRCCS, Rome, Italy
3
Össur hf, Reykjavik, Iceland
4
Biorobotics Laboratory, EPFL, Lausanne, Switzerland
5
Human Robotics Group, Imperial College London, London, UK
6
Delft University of Technology, Delft, The Netherlands
Abstract. The main goal of the Symbitron project was to develop a safe, bio-
inspired, personalized wearable exoskeleton that enables SCI patients to walk
without additional assistance, by complementing their remaining motor func-
tion. Here we give an overview of major achievements of the projects.
1 Major Achievements
The highlights of the Symbitron project are summarized below. For pictures and
movies we refer to the website: https://www.symbitron.eu
1.1 Symbitron Framework

The different components of the Symbitron framework will be discussed below:
1.1.1 EtherCAT
As was already decided in the first year of the project, the real-time control of the
Wearable Exoskeleton will be done using EtherCAT. In this way we can easily
The study in this work was supported by the Symbitron project, partially funded under grant
661626 of the Seventh Framework Program (FP7) of the European Commission (Information and
Communication Technologies, FP7-ICT-2013-10).

https://doi.org/10.1007/978-3-030-01887-0_69
362 H. van der Kooij et al.
combine several modules (EtherCAT slaves), which is according to our goal of

developing a modular system. Furthermore, it contributes to the requirement of a
minimal wire solution for the wearable exoskeleton.
The main etherCAT slaves (stack) are placed in the backpack of the exoskeleton
together with the pc that runs the different models for the control of the EtherCAT
slaves.
Some off-the-shelf EtherCAT slaves could be used (e.g. for the motor control), but
also during the course of the project, some EtherCAT slaves have been developed/built
for specific hardware components of the wearable exoskeleton.
1.1.2 EtherLab and Symbitron Wiki

To control the EtherCAT slaves in real-time, EtherLab in first instance has been
selected as the EtherCAT master, because it is open-source and easily communicates
with the hardware and the Matlab Simulink control models.
The installation of EtherLab was not described extensively anywhere, therefore a
master student and some Symbitron members went through all the installation steps and
have documented the required steps and especially the problems they encountered. The
resulting installation manual has been put on the Symbitron wiki (www.symbitron.eu/
wiki) to easily share it with the other Symbitron members, but also to share it with other
people who want to use EtherLab. The Symbitron wiki is therefore open for everyone,
but editing is only possible after creating an account with permission of the website
manager. The Symbitron wiki shows up in the first hits on Google when searching for
“EtherLab” and is the first hit when looking for “Etherlab installation”.
1.1.3 Simulink Libraries and GIT Repository

To control the different slaves, Matlab Simulink models are being used. For each
EtherCAT module a Simulink library block has been created. These blocks can be
directly inserted in a Matlab Simulink model and by putting the correct slave number in
the blocks the data from the blocks can be read out or data can be sent to the slaves.
To be able to (1) share the different Simulink models among the consortium, (2) to
be able to use them on different pc’s (development pc and control pc) and (3) to have
version control, a Symbitron GIT repository is being used. This GIT repository is being
stored on Bitbucket. Bitbucket not only provides the hosting of the repository, but also
allows for issue tracking for every project. The Symbitron repository is only open for
Symbitron members or on request.
1.2 Symbitron Measurement Week

Ten symbitron test pilots came over from Rome to Enschede (UTwente), which
required a high level of organization. During the whole week, three different mea-
surement setups have been used:
1. The LOPES is installed at the lab of the Biomechanical Engineering group at
UTwente. For the LOPES experiments an 8 channel Delsys EMG system was used
for EMG measurements. Data from the LOPES (angles, velocities, interaction forces,
etc.) and the EMG system were synchronized and stored on a measurement pc.
Symbitron: Symbiotic Man-Machine Interactions in Wearable Exoskeletons 363
2. The Achilles experiments were performed on the instrumented treadmill that is

installed in the lab of the Biomechanical Engineering group at UTwente. The
instrumented treadmill is equipped with handrails and a safety harness is used to
ensure the safety for the patients. The Achilles (the motors to provide the torque
around the ankle) has its own control pc that works via EtherCAT and EtherLab.
This pc is also used to read out the data from three IMU’s that were placed on the
trunk and the upper and lower leg. An 8 channel Porti EMG system (TMSi, the
Netherlands) was used for EMG measurements and data was captured and stored on
a separate laptop. Motion capture data was collected by using the Visualeyez system
with four cameras, four wired markers on the knee and ankle joints and several
marker clusters on different body segments. The marker data was also collected on a
different measurement pc.
The last measurement pc was used to control the Pusher device that was used to
provide the perturbations at trunk level to the subjects and to measure the force data
from the force plates of the instrumented treadmill. From this pc also a synchro-
nization signal (random signal) was sent to the Visualeyez pc, the EMG pc and the
Achilles pc.
3. The spasticity measurement setup consisted of a bed, which could be adjusted in
height, where the subjects could lay down and the legs could be freely moved by the
experimenter in different angles and at different speeds. 3D-printed handles with
incorporated force sensors were used to move the legs. These handles were made of
thermoplastics, which allowed for a custom fit for all subjects. Besides force data,
also EMG data was collected via an 8 channel Delsys EMG system and Xsens
IMU’s were used to collect data of the leg movements.
1.3 The Symbitron Modular Exoskeleton Hardware and Software

Mechanical engineers from TuDelft and electrical engineers from the UTwente with
support from Ossur made a considerable effort to design and realise the WE1 and WE2
prototype. We developed a lightweight (1.5 kg) powerful universal joint that is con-
nected with a personalised structure. The exoskeleton can be used in different con-
figurations, like ankle only, ankle and knee, and ankle-knee-hip. All configurations can
be used for one leg or both legs. The Software automatically recognizes the hardware
configuration.
1.4 Successful Clinical Training and Evaluation with SCI Test Pilots
The extensive clinical training and evaluation with the SCI test pilots showed that
symbitron hardware and software could be personalised to the size of the different test
pilots and to specific capacities and characteristics of our test pilots. With the biological
inspired controllers all incomplete SCI subject could walk, and they improved their
walking speed and/or balance during training with the Symbitron devices. A combi-
nation of a biological inspired ankle controller and trajectory control at the knee and
hips enabled two complete SCI subject to walk again. Psychometric test showed that
Symbitron technology was well received by test pilots, and they remained all motivated
to use our devices.
364 H. van der Kooij et al.
2 Final Results and Their Potential Impact and Use
After more than 4 years of research we have developed a modular exoskeleton that can
be modified to the size of different subjects wearing the device and their capacities. The
unique modular design allows different configurations: only ankle support, ankle-knee
support, or ankle-knee-hip support. It is also possible to support only one or both legs,
which might be interesting when used by stroke survivors. This mechanical modularity
makes it possible to adapt to hardware to the specific needs of its user. Also the
electronics and software is modular. Using Ethercat the software automatically rec-
ognizes the hardware connected to it. Also the software was build in such a way that
within a few seconds it can be adopted to the configuration it is connected to, by simple
toggle on or off some software modules.
In the Symbitron project we included incomplete SCI subjects that needed only
support at the ankle, or at the ankle and knee and complete SCI subjects that needed
full support of both legs. Clinical tests showed that hardware and software could be
adjusted to the specific characteristics of these subjects, which provided a proof of the
feasibility of the unique Symbitron approach. The control of the devices was biolog-
ically inspired and rendered muscle and reflex dynamics on the exoskeletons legs. In
Symbitron we were the first to implement this successfully in wearable exoskeletons,
and we were also the first to apply this biologically inspired neuro-muscular controllers
to multiple joints on a wearable robot. The clinical tests with SCI subjects proofed the
feasibility of this approach, and showed that in contrast to conventional approaches it
allows for gait patterns that are variable (in term of speed and step length).
Although the clinical results are still preliminary, the chosen approach is promising,
in particular for incomplete SCI subjects or similar patient groups who have some
remaining function left. We showed that during training with the Symbitron devices
incomplete SCI subjects improved walking speed and/or balance. In some cases also a
rehabilitation effect was seen, i.e. they improved their function during training with the
Symbitron devices even when they did not use the device. Psychometric test also
showed that the test pilots were very motivated and satisfied using the devices.
In conclusion, the Symbitron approach exploiting modularity of hardware and
software and biological inspired control, has potential in supporting incomplete SCI
subjects but also stroke survivors during rehabilitation and as an assistive device.
The Symbitron hardware and software will be re-used in a national Dutch Wearable
Robotics research program that will start August 2018, and which involves 5 univer-
sities and more than 20 companies and end-user organisations. The Symbitron tech-
nology will also be used and further developed in the Symbitron+ team that will
compete in the Cybathlon 2020 competition in Zurich.
Beyond Robo-Mate: Towards
the Next Generation of Industrial
Exoskeletons in Europe
Jesús Ortiz(B) , Stefano Toxiri, and Darwin G. Caldwell

via Morego, 30, 16163 Genova, Italy
jesus.ortiz@iit.it
Abstract. The Robo-Mate project developed industrial exoskeletons to

reduce the risk of physical injury associated to manual material han-
dling tasks. Prototypes targeting different body areas were evaluated for
their effectiveness but also for their applicability and usability in the
field. Encouraging evidence of their effectiveness and informative feed-
back from the field have driven further research and development. This
has additionally led the initiation of a new collaborative project, which
aims at continuing technical advancements as well as at promoting the
translation to diverse areas of application.
1 Introduction
About 44 million EU workers are affected of musculoskeletal conditions resulting

in annual costs of more than 240 billion Euro to the European economy. Of all
musculoskeletal conditions the largest portion of lost working days are accounted
to back pain [1]. The costs related to musculoskeletal injuries are ever increasing,
encouraging the industry to explore and invest in new ideas and technologies.
Among the possible solutions, the use of wearable exoskeletons has been explored
in research (see review in [2]) and has resulted in an expanding commercial
landscape as well1 .
The Robo-Mate project, active between late 2013 and late 2016, focused on
devices for the reduction of lower back compression forces. This contribution
describes the project’s approach, proposed technology, and summarizes the find-
ings. Additionally, it presents an ongoing project that follows up on Robo-Mate’s
effort towards technological improvement and its applicability in the field.
(INAIL).
The Robo-Mate project received funding from the European Unions Seventh
Framework Programme for research, technological development and demonstration
under grant agreement No. 608979.
1
https://exoskeletonreport.com/product-category/exoskeleton-catalog/industrial/.
https://doi.org/10.1007/978-3-030-01887-0_70
366 J. Ortiz et al.
2 The Robo-Mate Project

2.1 Approach
The main goal of the Robo-Mate EU project2 was to develop a user-friendly
intelligent wearable exoskeleton for manual handling work in different indus-
tries. The manufacturing lines of three end-users (INDRA: end-of-life vehicle
recycler, CRF: car manufacturer, and ROPARDO: part supplier for the auto-
motive industry), which are all partners in the Robo-Mate consortium, were
analysed to identify their specific needs. The results are that, for objects heavier
than 20 kg, solutions typically already exist to support the handling and workers
are advised to use them. For manual handling objects ergonomic indicators (like
[3]) can provide a reasonable assessment of the related risks. The unanimous
opinion is that, for tasks with a repetitive character, even few kilograms may
pose a significant risk for the workers. Within the Robo-Mate’s end-users man-
ufacturing lines, many repetitive tasks with lightweight loads were identified,
resulting in the definition of the following two use cases:
• Lifting, carrying and lowering: two-handed manipulation of goods
between 5 and 15 kg;
• Postural support: one-handed balancing and positioning of an object up to
7.5 kg.
2.2 Concept and Prototypes

In order to make it adjustable to a variety of industries and tasks, Robo-Mate
was designed following a modular approach. The system was composed of three
modules which were able to work independently or in combination depending on
the target task:
• Passive parallelogram arms. This module uses a spring mechanism which
provides constant support to the arms, reducing the user’s arm effort as well
as offering safe handling of medium/heavy loads.
• Active parallelogram arms. Following a similar concept of the passive
parallelogram arms, this module represents a technological advance including
a state-of-the-art wire pull functions, allowing the user to benefit from the
assistive support for pick and place tasks.
• Active trunk. This module reduces lower back musculoskeletal loading by
applying a supporting torque at the hip. The module permits the worker to
adopt natural but assisted movements as they perform their lifting activities.
2.3 Testing
During the Robo-Mate project, the different prototypes were evaluated under
different aspects, such as biomechanics, usability and integration in industry. For
this reason, the testing was divided in three levels:
2
www.robo-mate.eu.
Beyond Robo-Mate: Towards the Next Generation 367
• Laboratory testing. The first prototypes of the exoskeleton were tested in

laboratory conditions in order to understand their physical effectiveness in
simulated manual material handling tasks. A revised version of the active
trunk module was later tested in order to evaluate the technical improve-
ments.
• Usability evaluation. The final prototypes were tested in a simulated indus-
trial scenario, focusing on the usability of the system, measuring subjective
factors and global variables such as energy expenditure and process time.
• Field testing. Finally, the prototypes were tested with factory workers in
the same tasks identified in the first stages of the project. The feedback from
the workers was collected using the System Usability Scale (SUS) forms and
other subjective qualitative measurements.
2.4 Main Outcomes
Two modules were tested in the lab: the trunk module and the passive arms. Lab
tests showed the first prototype of the trunk module to be effective with regard
to physical exertion during dynamic lifting tasks. While holding static bent pos-
ture, the device was only beneficial for bending angles over 20deg, although
it appeared to increase physical exertion for the legs. Usability is clearly the
area where this prototype needed most improvement, not reaching an accept-
able score. Passive arms were effective at reducing perceived exertion for the
arms, but slightly increased exertion for legs and trunk. Usability scores for the
arms were significantly higher than for the trunk module: Half of test subjects
reached an acceptable usability score. Further testing of the revised version of
the trunk module showed a clearer benefit in lifting and lowering tasks [4], and
a better user acceptance emerged during the industrial testing.
All of Robo-Mate’s three modules were tested in three companies: CRF con-
cluded that all three modules could have a beneficial effect on workers’ health
and on efficiency – following a few improvements in usability. Workers preferred
the trunk module since it reduced physical effort and also decreased task dura-
tion. INDRA concluded that the passive arm module would be a good choice for
tasks where objects have to be carried or held for a longer period of time. The
active arms was valued for its support at lifting objects with a straight back.
The trunk module, again, was the favourite of workers, since it provided the best
support for bending and lifting tasks and did not restrict movements. COMPA
was met with a lot of enthusiasm of workers who were testing the modules. Their
key recommendation for further development of the modules was in line with lab
tests: Reduce weight and dimensions of the modules.
3 Ongoing and Future Research

Following the end of the project, several aspects were identified for the improve-
ment of the Active Trunk module. In particular, although the parallel spring
mechanism had been beneficial for the actuators [5], was temporarily left out on
368 J. Ortiz et al.
a revised version aimed at testing campaigns focused on control strategies. This

change allowed to reduce the overall weight (by over 1 kg) and to save some space
laterally and at the back. Revised electronics additionally improved robustness
and compactness of cables. Figure 1 shows the revised version. Regarding control
strategies, one of the limitations of the final Robo-Mate prototype was that it
considered only the trunk bending angle and not the variation of the load weight.
The approach of estimating the load by measuring the pressure contact between
the shoes and the ground was not robust enough in industrial scenarios, mostly
due to adjustment and calibration issues [6]. A new approach based on sur-
face Electromyography (sEMG) placed on the forearm represented an important
improvement in the controllability and usability of the system. Tests conducted
recently explored the potential of this control strategy [7].
Fig. 1. The trunk module prototype, achieved shortly after the end of the Robo-Mate
project.
An ongoing follow-up project has leveraged on the collaboration with the

Italian Workers’ Compensation Authority (INAIL). The overall objectives
include further advancements on the trunk module, such as stronger assistance at
the back and possibly extending it to the knee joint. Additionally, the shoulders
will be targeted as another joint commonly affected by injuries in the workplace.
The focus for the developments is on usability (covering actuation performance,
power autonomy, and control strategies) as well as comfort (covering weight
and physical attachments). Besides technical advancements, the project aims
at extending to areas of application outside manufacturing, possibly including
logistics, maintenance and healthcare.
Beyond Robo-Mate: Towards the Next Generation 369
References
1. Bevan, S.: The impact of back pain on sickness absence in Europe. The Work Foun-
dation, Lancaster (2012)
2. de Looze, M.P., Bosch, T., Krause, F., Stadler, K.S., OSullivan, L.W.: Exoskele-
Ergonomics 59(5), 671–681 (2016)
3. Waters, T.R., Putz-Anderson, V., Garg, A., Fine, L.J.: Revised NIOSH equation
for the design and evaluation of manual lifting tasks. Ergonomics 36(7), 749–776
(1993)
4. Huysamen, K., de Looze, M., Bosch, T., Ortiz, J., Toxiri, S., OSullivan, L.W.:
Assessment of an active industrial exoskeleton to aid dynamic lifting and lowering
manual handling tasks. Appl. Ergon. 68, 125–131 (2018)
5. Toxiri, S., Calanca, A., Ortiz, J., Fiorini, P., Caldwell, D.G.: A parallel-elastic actu-
ator for a torque-controlled back-support exoskeleton. IEEE Robot. Autom. Lett.
(2017)
6. Mateos, L.A., Ortiz, J., Toxiri, S., Fernández, J., Masood, J., Caldwell, D.G.:
Exoshoe: a sensory system to measure foot pressure in industrial exoskeleton. In:
2016 6th IEEE International Conference on Biomedical Robotics and Biomecha-
tronics (BioRob), pp. 99–105. IEEE (2016)
O’Sullivan, L., Caldwell, D.G.: Rationale, implementation and evaluation of assistive
strategies for an active back-support exoskeleton. Front. Robot. AI 5(53) (2018)
The SoftPro Project: Synergy-Based
Open-Source Technologies for Prosthetics
and Rehabilitation
Cristina Piazza1(B) , Manuel G. Catalano2 , Matteo Bianchi1

Emiliano Ricciardi3 , Domenico Prattichizzo2,4 , Sami Haddadin5
Andreas R. Luft6 , Olivier Lambercy7 , Roger Gassert7 , Eike Jakubowitz8 ,
Herman Van Der Kooij9 , Frederick Tonis10 , Fabio Bonomo11
Benjamin de Jonge12 , Tomas Ward13 , Kristin D. Zhao14 , Marco Santello15 ,
and Antonio Bicchi1,2
1
University of Pisa, Pisa, Italy
cristina.piazza@ing.unipi.it
2
Istituto Italiano di Tecnologia, Genoa, Italy
3
IMT School for Advanced Studies Lucca, Lucca, Italy
4
University of Siena, Siena, Italy
5
Technische Universitat Munchen, Munich, Germany
6
University Hospital Zurich, Zurich, Switzerland
7
ETH Zurich, Zurich, Switzerland
8
Hannover Medical School, Hannover, Germany
9
University of Twente, Enschede, Netherlands
10
Hankamp Rehab, Enschede, Netherlands
11
qbrobotics, Navacchio, Italy
12
TMS International, Enschede, Netherlands
13
Bioservo Technologies AB, Kista, Sweden
14
Mayo Clinic, Rochester, MN, USA
15
Arizona State University, Tempe, AZ, USA
Abstract. Robotics-enabled technologies for assistive and rehabilita-

tive applications have gained an increasing attention, both in academic
and industrial research settings, as a promising solution for human
sensory-motor system recovery. However, many constraints remain that
limit their effective employment in everyday-life, mainly related to cost,
usability and users’ acceptance. The Softpro project proposes to com-
pletely reverse such paradigm, starting from the analysis of the needs
from patients and the careful investigation of the sensory-motor human
behaviour, capitalizing on the framework of synergistic control and soft
robotics. The final goal is to study and design simple, effective and afford-
able soft synergy-based robotic technologies for the upper limb, such as
This research has received funding from the European Union’s Horizon 2020 Research
and Innovation Programme under Grant Agreement No. 688857 (SoftPro). The con-
tent of this publication is the sole responsibility of the authors. The European Com-
mission or its services cannot be held responsible for any use that may be made of
the information it contains.
https://doi.org/10.1007/978-3-030-01887-0_71
The SoftPro Project: Synergy-Based Open-Source Technologies 371
new prostheses, exoskeletons, and assistive devices which can be useful

and accessible to a wide audience of users. To pursue such an ambi-
tious objective, SoftPro has put together research groups who laid the
neuroscientific and technological fundamentals underpinning the project
approach, a net of international collaborations and numerous and qual-
ified industrial partners, which is expected to produce a strong impact
on both research and innovation.
1 Introduction
Currently available technology-enabled rehab training and assistive devices are
very rarely manageable, intuitive, and affordable enough to be used in patients’
everyday life. The ambitious objective of the SoftPro project [1] firmly adheres
to a quantitative assessment-based philosophy in the development of innovation
concerning prosthetics and neurorehabilitation devices. Starting from homoge-
neous methodological bases, the aim of the project is to address a spectrum of
challenges ranging from technologies that, starting from a validation in relevant
environments, reach a complete and qualified status, to high-risk high-gain ideas
supported only by preliminary observations that will hopefully reach a relevant
validation stage. Three aspects are crucial in our project: the control of prosthe-
ses and rehab devices, where the human “talks” to the artificial device to con-
trol its motion through brain- and body-machine interfaces; the haptic stimulus
delivery, where the artificial “talks” to the human to render the missing or defi-
cient sensory feedback; and the implementation of elementary semi-autonomous
sensorimotor loops, such as the grip reflex in a hand prosthesis or exoskeleton
(Fig. 1).
Fig. 1. SoftPro technologies and integration

372 C. Piazza et al.

Based on the previous work of its research and industrial participants and in
close collaboration with clinical participants, The SoftPro project is working on
the development of new advanced instruments to measure the dynamics of inter-
action between the user, the assistive device and the environment (other people
and objects in the surroundings) in new and more accurate ways. In SoftPro,
starting from a solid neuroscientific theoretical basis on sensorimotor synergies
as the elementary alphabet of human motor control primitives and from the
theory and technology of soft robotics that enable controllable impedance and
adaptability in physical human-robot interaction, new methods and technologies
for prosthetics and rehabilitation can be developed. Furthermore, a principled
simplification approach allows the development of the simplest technology to
fulfill a desired assistive goal. This approach is then extended to the design of
upper limb and hand prostheses with a larger/different set of synergies, and
supernumerary limbs for assistance and rehabilitation (extratheses) to increase
upper limb functionality and subsequent independence in activities of daily liv-
ing (ADLs) in persons where no more functional motor improvements seem to
be achievable. The SoftPro principled simplification approach also provides for
hand and upper limb exoskeletons that are light-weight, low-cost, easy to don
and only minimally interfere with natural motion and interaction with the envi-
ronment. Finally, to maximize the impact of our research we are following a very
open approach to sharing our results. SoftPro promotes open access not only to
data collected, but also to technology developed, for the purpose of building a
community of users and developers which will in turn contribute to furthering
our goals of making prosthetics and rehabilitation aids more easily accessible
(Fig. 2).
3 Results
Main results currently achieved in applied neuroscience include the clarification
of the theory of sensorimotor synergies in rehabilitation [2], and the develop-
ment of novel algorithms for eliciting synergy-inspired control of robotic aids
from brain-machine interfaces. The SoftPro project is further pursuing the con-
cept of soft-synergy based hand prostheses (the SoftHand Pro, [3]), currently
being tested at a wide range of international facilities, at a pre-marketable state
(TRL8). New tools for the assessment of effectiveness of robot-enabled prostheses
and assistive devices include the maturation of the Virtual Peg Insertion Test,
the development and application of force/torque based [4], intrinsic tactile sens-
ing techniques to measure grasping and manipulative forces with tools such as
instrumented objects and the ThimbleSense [5]. Moreover, other SoftPro devices
are currently reaching a higher TRL, as e.g. the shell-based exoskeletons, and
IMU-based posture and measurement gloves [6]. Starting from previous works
by participants, as the HexoSys [7] or the HandExo [8], the SoftPro exoskeletons
will be designed to achieve a reliable construction with more optimized kinemat-
ics, using suitable materials, small form factor, low weight and high precision.
The SoftPro Project: Synergy-Based Open-Source Technologies 373
The design and validation of novel interfaces for haptic stimulus delivery in pros-
thetics and robotics rehab, represents a significant improvement for closing the
sensory-motor loop in an intuitive and simple manner, likely contributing to
device acceptance, as e.g. [9,10]. Finally, SoftPro is featuring “red label” investi-
gations and developments, where the project is undertaking higher risks testing
disruptively new hypotheses and paradigms. These include the attempt to esti-
mate from autonomic nervous system signals, subtle but important information
such as stress and/or fatigue [11]; and pioneering the idea of using robotic extra
limbs [12] for assistance to persons with chronic motor impairments.
(a) (b) (c) (d)
Fig. 2. Examples of devices developed or used within the SoftPro project: (a) SoftHand
Pro, (b) Stretch Pro, (c) HandExo and (d) Sixth Finger.
4 Conclusion
The SoftPro project addresses the scope to advance key technologies for assistive
and rehabilitative robotics. In terms of innovation, SoftPro capitalizes on pre-
vious research work by the participants, often in collaboration, to significantly
progress in engineering, clinical validation, and assess the economic viability of
systems for robotics-enabled aids. Thanks to the strong links between partici-
pants, the industrial pickup of at least some of these results is highly probable.
References
1. SoftPro Project website. http://www.softpro.eu/
2. Leo, A., Handjaras, G., Bianchi, M., Marino, H., Gabiccini, M., Guidi, A., Scilingo,
E.P., Pietrini, P., Bicchi, A., Santello, M., Ricciardi, E.: A synergy-based hand
control is encoded in human motor cortical areas. Elife 5, e13420 (2016)
3. Piazza, C., Catalano, M.G., Godfrey, S.B., Rossi, M., Grioli, G., Bianchi, M., Zhao,
K., Bicchi, A.: The SoftHand pro-H: a hybrid body-controlled, electrically powered
hand prosthesis for daily living and working. IEEE Robot. Autom. Mag. 24(4),
87–101 (2017)
4. Hofmann, P., Held, J.P., Gassert, R., Lambercy, O.: Assessment of movement pat-
terns in stroke patients: a case study with the virtual peg insertion test. In: Pro-
ceedings of the International Convention on Rehabilitation Engineering & Assistive
Technology, p. 14. Singapore Therapeutic, Assistive & Rehabilitative Technologies
(START) Centre (2016)
374 C. Piazza et al.
5. Battaglia, E., Grioli, G., Catalano, M.G., Santello, M., Bicchi, A.: ThimbleSense:
an individual-digit wearable tactile sensor for experimental grasp studies. In: 2014
IEEE International Conference on Robotics and Automation (ICRA), pp. 2728–
2735. IEEE (2014)
6. Santaera, G., Luberto, E., Serio, A., Gabiccini, M., Bicchi, A.: Low-cost, fast and
accurate reconstruction of robotic and human postures via IMU measurements.
In: 2015 IEEE International Conference on Robotics and Automation (ICRA), pp.
2728–2735. IEEE (2015)
7. Iqbal, J., Tsagarakis, N.G., Caldwell, D.G.: A multi-DOF robotic exoskeleton inter-
face for hand motion assistance. In: 2011 Annual International Conference of the
IEEE Engineering in Medicine and Biology Society, EMBC, pp. 1575–1578. IEEE
(2011)
8. Nycz, C.J., Bützer, T., Lambercy, O., Arata, J., Fischer, G.S., Gassert, R.: Design
and characterization of a lightweight and fully portable remote actuation system
for use with a hand exoskeleton. IEEE Robot. Autom. Lett. 1(2), 976–983 (2016)
9. Rossi, M., Bianchi, M., Battaglia, E., Catalano, M.G., Bicchi, A.: Hap-Pro: a wear-
able haptic device for proprioceptive feedback. IEEE Trans. Biomed. Eng. (2018)
10. Meli, L., Hussain, I., Aurilio, M., Malvezzi, M., O’Malley, M., Prattichizzo, D.: The
hBracelet: a wearable haptic device for the distributed mechanotactile stimulation
of the upper limb. IEEE Robot. Autom. Lett. 3(3), 2198–2205 (2018)
11. Peternel, L., Tsagarakis, N., Caldwell, D., Ajoudani, A.: Robot adaptation to
human physical fatigue in human-robot co-manipulation. Auton. Robots 42, 1–
11 (2017)
12. Salvietti, G., Hussain, I., Prattichizzo, D.: The robotic sixth finger: a wearable
compensatory tool to regain grasping capabilities in paretic hands. In: Robotics
Research, vol. 2, pp. 423–437. Springer (2018)
EUROBENCH: Preparing Robots for the Real
World
D. Torricelli(&) and J. L. Pons
Neural Rehabilitation Group, Spanish National Research Council, Madrid, Spain

{diego.torricelli,jose.pons}@csic.es
Abstract. Robots are entering our everyday life at an exponential pace.

Benchmarking can help researchers and developers to improve their systems, as
well as providing end-user with easy-to-understand “performance scores” able
to identify the best solution for their needs. Unfortunately, a benchmarking
methodology for robotics is still not available. The EUROBENCH project wants
to provide rigorous tools at both software and hardware level, to allow com-
panies, researchers and users to test robotic systems under multiple facets. The
project, initially focused on bipedal robotics technologies (exoskeletons, pros-
theses, humanoids), will offer funding opportunities to third parties to participate
in the development and validation of the different components of the framework.
1 Introduction
An increasing number of robotic solutions are available to in the market public for a
great variety of everyday applications. A similar trend is about to happen for wearable
robotics. Nevertheless, several roadblocks exist in this process. Some of these are
technical, while others are related to the lack of reliable performance/safety indicators
for these devices to meet international certifications and standardization requirements.
Benchmarking is taking an increasingly important role as a quantitative instrument to
assess the Technology Readiness Level (TRL) of technology and to quantify how
robotic solutions match user needs [1–3]. Recent international efforts such as
Cybathlon, RockIn [4], European Robotics League, have confirmed the interest of the
scientific and industrial communities in evaluating (and comparing) the performance of
robotic systems in real-life environments. However, a consolidated benchmarking
methodology for Robotics has not been reached yet. In order to fill these gaps, the
H2020 project EUROBENCH aims to create the first international framework for
robotic benchmarking. This framework, mainly (but not only) focused on bipedal
robotic technologies, will include methods and tools to measure System Ability Levels
on a rigorous, quantitative and replicable way.
This work is supported by the project EUROBENCH (European ROBotic framework for bipedal
locomotion BENCHmarking) funded by H2020 Topic ICT 27-2017 under grant agreement no:
779963.

https://doi.org/10.1007/978-3-030-01887-0_72
376 D. Torricelli and J. L. Pons
2 Why Benchmarking?
Costumers need to compare available solutions to identify which one better fits their
needs. This process is normally facilitated by “scores” that quickly define the product
performance. e.g. the efficiency class of electrical appliances, the acceleration of a car,
or the data transfer speed of an internet-based provider. Based on these simple num-
bers, customers can easily understand the key product functions, and make the right
choice. While this is a common process for most products available in the market,
when it comes to robotics, the picture is very different. Indicators of robot performance
are very heterogeneous and, when available, are conceived to be used by expert users,
e.g. engineers. Simple, easy-to-understand metrics of robotic performance are in gen-
eral not available [5].
In the research scenario, assessment protocols, methods and metrics are usually
specified around system-specific or lab-specific conditions and goals. In this respect,
the creation of a benchmarking methodology will help the robotics community to
increase reproducibility and comparison of robotics systems to each other, and truly
“stand of giants’ shoulders”.
As a first, important, step in this direction, the Multi-Annual Robotics Roadmap for
Robotics [6] proposed a powerful method to quantify the technical progression of
robotic systems, using a three-level taxonomy composed of Technologies, System
Abilities, and Domains (see Fig. 1). While Technologies represent the “building
blocks” of a robotic system, System Abilities describe the performance of the whole
system under a global and functional perspective. System Abilities can help developers
to understand to what extent technological improvements can contribute to the global
performance of a system. At the same time, System Abilities can also provide useful
indications to non-experts about the level of technological readiness (TRL) of a system
in a particular domain. Nevertheless, the MAR does not specify how System Abilities
should be quantitatively assessed. Due to the high variability of robotic applications
and technologies, it is still not clear how System Ability levels can be quantified and
measured on a realistic and application-specific basis.
The EUROBENCH project aims at establishing standard benchmarks of robot
performance and making them available to the general public, the research community,
and the industry. The main goal is to convert the System Abilities in measurable items
with purposeful meaning for both technical developers and end-users.
The project is focused on two main expected outcomes.
The first outcome is a methodological framework, which will include methods and
metrics to calculate the System Ability levels. These methods will be integrated in a
software suite to permit its wide use across domains and laboratory conditions. The
main goal of the software will be to facilitate the use of benchmarking methodology at
all levels from research to commercial prototyping.
The second outcome is represented by an experimental framework, which will
concentrate the state-of-the-art testbeds in two facilities, one for wearable robots (in-
cluding exoskeletons and prostheses), and one for humanoid robots. These facilities
will allow companies and/or researchers to perform standardized tests on advanced
EUROBENCH: Preparing Robots for the Real World 377
DO ROBOTSMEET
USERNEEDS?
TECHNOLOGIES SYSTEM ABILITIES APPLICATION DOMAINS
CURRENT SOLUTIONS
USERS’ NEEDS
Benchmarking: Quantifying system abilitiesand make them
available to researchers, companiesand end-users
Fig. 1. The EUROBENCH questions and goal.
robotic prototypes in a unique location, saving resources and time, and preparing for
the certification processes.
3 The Cascade Funding Approach
3.1 Developing the Framework

The EUROBENCH Consortium will count on the collaboration of the robotic com-
munity by means of a cascade funding modality named “Financial Support for Third
Parties” (FSTP). Under this scheme, external companies and research entities can take
active part in designing and developing specific benchmarking tools and methods. The
best solutions will be included in the framework, as components of the EUROBENCH
Software and/or Facilities. This first FSTP action will be focused on the following
outcomes:
• Testbeds, i.e. devices/structures able to reproduce in a standardized and replicable
way typical ‘out-of-the-lab’ environments (e.g. uneven terrains, stairs, obstacles) as
well as external perturbations (e.g. pushes, unstable support surfaces).
• Sensors, i.e. devices able to record kinematic, kinetic and/or physiological variables
from bipedal systems (i.e. humans, humanoids, exoskeletons, and/or prostheses)
across a wide range of motor tasks.
• Benchmarking routines, i.e. computational algorithms able to quantify one or more
system abilities during bipedal locomotion functions (walking and/or balance).
• Datasets, i.e. experimental data obtained by human and/or robot systems during the
execution of different motor skills.
378 D. Torricelli and J. L. Pons
3.2 Validating the Framework

Once the Software tool and the Facilities will be operative (estimated by the end of
2020), the Consortium will make them available as Beta testing tools to Third Parties
interested in using them to test the performance of their commercial or research bipedal
robotic prototypes/algorithms. This second FSTP Action is addressed to Third Parties
interested in either:
• Performing experiments in the Facilities, to: (1) test their own robot prototype or
(2) implement their control algorithm into standard bipedal platforms available in
the facility. The grant will cover expenses for two weeks of tests (called ‘PRE’ and
‘POST’) separated by a period of at least 3 months, in which the Third Party is
requested to improve its system before performing the second round of trials
(‘POST’ test session).
• Performing the tests in their own laboratory settings, by using only the Bench-
marking Software. The software will support the researcher to design and run a
number of tests on their robotic platforms and return a set of performance scores.
In both modalities, the EUROBENCH Consortium will support the Third Parties in
all the experiments, data processing and interpretation of the results, as well as advising
on how to improve their systems. The Consortium will use the outcome of this FSTP
action to validate the benchmarking framework and carry out the subsequent
exploitation of the results. All these procedures will be developed in total respect of
confidentiality and industrial secret requirements of participants.
4 Conclusion
This paper presents the main goals of the European project EUROBENCH, which aims
to create the first framework for the benchmarking of bipedal robotics technologies. In
the medium-term period, this project will set the foundations of standardized bench-
marking tools for a wide range of robotic technologies and application domains.
References
1. Strategic Research Agenda (SRA) 2014–2020: Technology assessment, pp. 962–993
2. Torricelli, D., et al.: Benchmarking lower limb wearable robots: emerging approaches and
technologies. In: Proceedings of the 8th ACM International Conference on PErvasive
Technologies Related to Assistive Environments, pp. 51–55 (2015)
3. Torricelli, D., et al.: Benchmarking bipedal locomotion: a unified scheme for humanoids,
wearable robots, and humans. IEEE Robot. Autom. Mag. 22(3), 103 (2015)
4. Amigoni, F., et al.: Competitions for benchmarking: task and functionality scoring complete
performance assessment. IEEE Robot. Autom. Mag. 22(3), 53–61 (2015)
5. Multi-Annual Roadmap (MAR) For Robotics in Europe Horizon 2020, Release B
02/12/2016, page 4 and 166
6. del Pôbil, A.P.: Why do we need benchmarks in robotics research? In: Proceedings of the
Poster Session
Actuation Requirements for Assistive
Exoskeletons: Exploiting Knowledge
of Task Dynamics
Stefano Toxiri1(B) , Andrea Calanca2 , Tommaso Poliero1,3 ,

Darwin G. Caldwell1 , and Jesús Ortiz1
1
stefano.toxiri@iit.it
2
Department of Computer Science, University of Verona, Verona, Italy
3
Department of Informatics Bioengineering Robotics and Systems Engineering,
University of Genoa, Genoa, Italy
Abstract. When selecting actuators for assistive exoskeletons, design-

ers face contrasting requirements. Overdimensioned actuators have draw-
backs that compromise their effectiveness in the target application (e.g.
performance, weight, comfort). In some cases, the requirements on the
powered actuator can be relaxed exploiting the contribution of an elastic
element acting in mechanical parallel. This contribution considers one
such case and describes an approach to fit the actuation requirements
closely to the task dynamics, thereby mitigating the drawbacks of overdi-
mensioned actuators.
1 Introduction
Exoskeletons are being explored to provide physical assistance in a wide variety
of applications ranging from enabling gait of paraplegic patients to increasing
endurance of military troops. The concept of physical assistance implemented
on several assistive exoskeletons may be described as the device following the
wearer’s movements and automatically detecting when and to what extent assis-
tance is necessary, in a way that provides meaningful assistance [1,2]. In modu-
lating the interaction forces the device may need to “feel transparent” or to gen-
erate large interaction forces. These two cases lead to opposite actuation design
requirements. Indeed, large force output with DC motors is typically achieved
using high-ratio transmission gears, which degrade transparency and make force
control more difficult [3–5]. They also tend to introduce extra efficiency losses,
costs, and weight. In order to limit the drawbacks of contrasting needs, it is
important to establish appropriate requirements in the early design stages.
The actuation requirements to implement such assistive action may be
described as consisting of (a) the intended output torque range, and (b) the
(INAIL).
https://doi.org/10.1007/978-3-030-01887-0_73
joint velocities. A device with misdimensioned actuators may prove ineffective

(not enough torque) or uncomfortable (not enough speed). In some cases, a solu-
tion in this direction is to reduce the need for active power by relying on passive
elastic elements to generate part of the physical assistance. This solution, known
as parallel-elastic actuation (PEA), has been explored and analyzed on wearable
robots in recent literature [5–9].
This contribution presents an approach to determine the actuation require-
ments for an exoskeleton that protects the lower back in lifting tasks [10],
although an attempt is made to generalize the approach to a generic actuated
joint on a wearable robot. A perspective on complementary necessary compo-
nents is also discussed.
2 Methods
The proposed approach exploits available knowledge of the task dynamics to

derive the actuation requirements. The target task consists of lifting and lower-
ing objects up to 15 kg. The exoskeleton aims at generating forces between the
torso and the upper legs, thus supporting back extension and thereby reduc-
ing back compression. Knowledge of the task dynamics is obtained by recording
motion and estimating joint torques via inverse dynamics on a multibody model,
as in [11]. The ranges of joint motion and torque are [0, 2.8] rad, [−5, 5] rad/s,
2
[−10, 20] rad/s , [0, 240] Nm. Following the anticipated concept of assistance and
based on the authors’ experience, the desired behavior is set as contributing
about one third of the total torque while not limiting movements. Considering
one actuator on each side, the desired torque range for one actuator becomes
[0, 40] Nm, while the speed is [−5, 5] rad/s.
50 50
w/ PE w/o PE
40 w/o PE 40 w/ PE
Joint Desired Torque [Nm]
Joint Desired Torque [Nm]
PE profile 109W
53W
30 30
20 20
10 10
0 0
-10 -10
-20 -20
0 0.5 1 1.5 2 2.5 3 -6 -4 -2 0 2 4 6
Joint Angular Position [rad] Joint Angular Speed [rad/s]
Fig. 1. The desired torque output is shown on the left against joint angle. The linear
model approximating their relationship (dashed blue) represents a the torque-angle
profile of an elastic element acting in parallel to the geared motor, thereby reducing its
peak torque. On the right, the torque for the two cases is shown against joint speed,
highlighting a reduction in peak power contribution from the geared motor.
Actuation Requirements for Assistive Exoskeletons 383
2.1 Static Requirements: Maximum Torque

Figure 1 on the left shows the desired torque (gray lines) against the measured
joint angle. The torques are distributed almost exclusively on positive values
(joint extension) and show correlation with the joint angles (static loads). For the
target task, a linear model (dashed blue) approximates the relationship between
torque and angle. A passive elastic element implementing the dashed profile
would generate a substantial portion of the desired torque on the target joint.
As a result, a geared motor acting in mechanical parallel (PEA) would only
need to generate the remaining portion (illustrated by the solid blue lines). In
other words, the necessary torque contribution from the geared motor would be
substantially lower in a PEA than in a non-elastic actuator. In this illustrative
case, the required active torque is approximately reduced from [0, 40] Nm to
[−15, 15] Nm.
2.2 Dynamic Requirements: From Speed to Power

Figure 1 on the right shows the desired joint torque output against the measured
joint speed. Similarly to Sect. 2.1, the gray lines represent the full joint torque
whereas the blue lines represent the portion of the torque required to the geared
motor in a PEA configuration. Considering the active mechanical power as the
product of motor torque and joint speed, in this illustrative case the required
peak active power decreases from 109 W to 53 W.
3 Discussion
Section 2.1 highlights the asymmetric distribution of desired joint torques to
support the target task, and how a PEA may be exploited to relax the torque
requirements on the geared motor [5,6,8]. The approximated linear relationship
between torque and angle is a convenient feature of this specific task (it emerges
in Fig. 1 on the left), but this may not be the case in a generic different joint
and task. In loose terms, it may be said that PEA is appropriate in those cases
where the elastic element alone would partially meet the requirements. The
contribution of the active portion provides the ability to modulate the assistance
around the baseline provided by the elastic element based on parameters such as
the type and intensity of the physical activity. The need to produce lower active
torques while following the same joint speed leads to lower power requirements as
well, as in Sect. 2.2. With respect to our previous work in [5], the perspective has
shifted from an elastic element displacing the operating point of a geared motor
to the contribution of a geared motor augmenting the capabilities of an elastic
element. In this situation, we no longer require the motor to dominate the elastic
element at all joint angles, but rather the spring and motor are dimensioned as
a whole to meet the task requirements. This enables the choice of a smaller
and lighter geared motor, thereby minimizing the disadvantages associated to
overdimensioned actuators (e.g. unnecessary weight, poor efficiency and dynamic
performance).
In order to further rationalize and improve the selection of actuators, the

proposed approach should be complemented with additional considerations that
have relevance to the function of the resulting device. Among these considera-
tions, the different losses from input to output power need to be understood and
accounted for [4,12]. Also, the factors affecting force control performance in the
interaction with the user’s variably soft tissues may provide helpful information
to the design process.
4 Conclusion
We propose an approach to define exoskeleton actuation requirements exploiting
the dynamics of the assisted task. If the task presents substantial static loads,
parallel-elastic actuation emerges as a solution to reduce the need for active
power, thus mitigating the associated drawbacks of overdimensioned actuators
in practical applications.
References
O’Sullivan, L., Caldwell, D.G.: Rationale, implementation and evaluation of assis-
tive strategies for an active back-support exoskeleton. Front. Robot. AI 5(53)
(2018)
2. Chen, B., Grazi, L., Lanotte, F., Vitiello, N., Crea, S.: A real-time lift detection
strategy for a hip exoskeleton. Front. Neurorobotics 12(April), 1–11 (2018)
3. Kong, K., Bae, J., Tomizuka, M.: A compact rotary series elastic actuator for
human assistive systems. IEEE/ASME Trans. Mechatron. 17(2), 288–297 (2012)
4. Verstraten, T., Mathijssen, G., Furnémont, R., Vanderborght, B., Lefeber, D.:
Modeling and design of geared DC motors for energy efficiency: comparison
between theory and experiments. Mechatronics 30, 198–213 (2015)
5. Toxiri, S., Calanca, A., Ortiz, J., Fiorini, P., Caldwell, D.G.: A parallel-elastic
actuator for a torque-controlled back-support exoskeleton. IEEE Robot. Autom.
Lett. 3(1), 492–499 (2018)
6. Wang, S., Van Dijk, W., Van Der Kooij, H.: Spring uses in exoskeleton actuation
design. In: IEEE International Conference on Rehabilitation Robotics, pp. 0–5
(2011)
7. Beckerle, P., Verstraten, T., Mathijssen, G., Furnemont, R., Vanderborght, B.,
Lefeber, D.: Series and parallel elastic actuation: influence of operating positions
on design and control. IEEE/ASME Trans. Mechatron. 22(1), 521–529 (2017)
8. Jimenez-Fabian, R., Geeroms, J., Flynn, L., Vanderborght, B., Lefeber, D.: Reduc-
tion of the torque requirements of an active ankle prosthesis using a parallel spring.
Robot. Auton. Syst. 92, 187–196 (2017)
10. Toxiri, S., Ortiz, J., Masood, J., Fernandez, J., Mateos, L.A., Caldwell, D.G.: A
wearable device for reducing spinal loads during lifting tasks: biomechanics and
design concepts. In: International Conference on Robotics and Biomimetics, pp.
2295–2300 (2015)
Actuation Requirements for Assistive Exoskeletons 385
11. De Looze, M.P., Kingma, I., Thunnissen, W., Van Wijk, M.J., Toussaint, H.M.:
The evaluation of a practical biomechanical model estimating lumbar moments in
occupational activities. Ergonomics 37(9), 1495–1502 (1994)
12. Verstraten, T., Geeroms, J., Mathijssen, G., Convens, B., Vanderborght, B.,
Lefeber, D.: Optimizing the power and energy consumption of powered prosthetic
ankles with series and parallel elasticity. Mech. Mach. Theory 116, 419–432 (2017)
Grasping Detection with Force Sensor
Embedded in a Hand Exoskeleton
Jorge A. Dı́ez(B) , José M. Catalán, Andrea Blanco, Juan Barios,

Santiago Ezquerro, Arturo Bertomeu-Motos, and Nicolás Garcı́a-Aracil
Biomedical Neuroengineering Group, University Miguel Hernandez of Elche,

Elche, Spain
jdiez@umh.es
Abstract. This paper presents the results of the force measurements

performed with an industrial-grade load cell embedded in the linkage of
a hand exoskeleton. The force sensor is placed such that it measures the
interaction force between the index finger of the user and the actuator
that controls its motion. This architecture has been used in an experi-
mental test in which users had to grasp an object (cup or bottle), interact
with it and then release it. Force measurements shows that this disposi-
tion allows to discern between successful and unsuccessful grasping.
1 Introduction
The recent development of new miniaturized force sensing technologies [1–4]
provides the designers of wearable devices with a wide range of sensor alter-
natives that can be integrated in their devices with a minimal impact in their
functionality and wearability.
A concrete application where the miniaturization of the sensors is specially
critical in terms of functionality is the design of hand exoskeletons. Specially, if a
hand exoskeleton is aimed to assist an impaired person during the performance
of activities of daily living, it must be as compact as possible in order to be
useful in real environments where a bulky device might easily collide with object
in the surrounding of the working area. For this application, it is interesting to
have a measurement of the force between the hand exoskeleton and the object
which it is interacting with.
In this paper, the authors study the potential applications of embedding
a single force sensor in a finger mechanism of a hand exoskeleton in terms of
detection of grasping. This setup is expected to allow the detection of a correct
This work has been supported by the European Commission through the project
AIDE: Adaptive Multimodal Interfaces to Assist Disabled People in Daily Activities
(Grant agreement no. 645322); by the AURORA project (DPI2015-70415-C2-2-R),
which is funded by the Spanish Ministry of Economy and Competitiveness and by
the European Union through the European Regional Development Fund (ERDF),
“A way to build Europe” and by and by Conselleria d’Educació, Cultura i Esport of
Generalitat Valenciana through the grants ACIF2016/216 and APOTIP/2017/001.
https://doi.org/10.1007/978-3-030-01887-0_74
Grasping Detection with Force Sensor Embedded in a Hand Exoskeleton 387
or unsuccessful grasping rather than obtaining an estimation of the quality and

stability of it, which will be object of further studies.
The presented tests are part of a bigger experimental setup, which pretends
to study the feasibility of using a complete multimodal system composed by
an arm exoskeleton [5], a hand exoskeleton, an object detection and tracking
system, combined with an EOG-EEG interface to assist an impaired user during
the performance of certain activities of daily living such as drinking, pouring or
eating.

The presented tests where performed with 6 users with certain degree of impair-
ment (Stroke, traumatic brain injury, and cerebral palsy). As a part of the global
experimentation mentioned above, they were asked to perform two series of ten
trials in which they had to control the different systems involved to reach and
object (glass or bottle), grasp it, perform a simple task and then release it in
its original position. All the stages of the trial were coordinated by a finite state
machine (FSM) that activated the required devices for each subtask. Since only
the information related to the grasping is relevant for the presented study, the
states can be grouped as shown in Table 1.
Table 1. FSM states description
State ID Descriptio Exoskeleton state

1 to 3 Previous to grasp Open
4 Object grasping Closing
5 to 7 Task with object Closed
8 Object releasing Opening
9 After object releasing Open
2.2 Hand Exoskeleton
To perform the grasping tasks, a hand exoskeleton with four active degrees
of freedom has been used [6,7]. The actuator corresponding to each degree of
freedom controls the movement of one finger (thumb, index and middle), with the
exception of ring and little fingers that are moved jointly with a single actuator.
In particular, each degree of freedom corresponds to an independent mod-
ule that is controlled by an electric linear actuator Actuonix PQ12-100-12-P.
388 J. A. Dı́ez et al.
These actuators have a maximum load capacity of 50 N and a stroke of 20 mm.

The linear movement of the actuator is transferred and converted to a rotation
movement around finger’s joints through a linkage and a pair of circular guides.
The main frame of the hand exoskeleton is composed by a hand orthosis that
holds the fixation elements required to attach the corresponding finger modules.
User’s fingers are fixed to the end of each finger module. Each module is fixed to
the user’s fingers by a custom-size elastic ring that is connected to the linkage
by a snap-in fastening mechanism.
Fig. 1. Hand exoskeleton with embedded load cell. Force to be measured is represented
by the blue and red arrows. Extension forces (red) are considered as positive values
while compression forces (blue) correspond to the negative values.
2.3 Force Sensor
The finger module that controls the motion of the index finger of the user has
been modified to host an industrial grade load cell OMEGA LCM201-100N,
which has a measurement range from −100 N to 100 N. This load cell is placed
between the attachment interface of the distal phalanx of the user’s finger and the
end of the transmission linkage of the exoskeleton (Fig. 1), thus it will measure
the interaction force between both human finger and finger exoskeleton.
3 Results
Figure 2 shows the average of the force measured by the sensor placed in the
index finger of the users computed for each FSM state through all the trials. For
all six users the states 5, 6 and 7 show an increasing of the measured force in
the negative direction (compression) associated to the grasped object. States 1
Grasping Detection with Force Sensor Embedded in a Hand Exoskeleton 389
to 3 and 9 show that the measured force is near to zero (There is an offset that
depends on the user). Though the trend is similar for all users, the mean force
in the states where the object is held varies significantly between users, this is
due to many factors such as number of successful trials or the force with which
the object is being held.
Fig. 2. Mean value of the measured forces in each state of the finite state machine.
Dashed lines show the data for each user, continuous line represents the average of all
users.
Fig. 3. Example of two experimental trials that show the state of the finite state
machine (FSM) that controls the system, the measured force and the movement of
the hand exoskeleton in terms of the fraction of hand closure, which is only measured
in the states where the exoskeleton is moving (In state 4: 0 corresponds to totally
open hand and 1 to totally closed. In state 8: 0 corresponds to totally closed hand
and 1 to totally open). Top figure shows an example of successful grasping, in which
the measured force is constant along the states where the exoskeleton is kept closed.
Bottom figure presents and example of unsuccessful grasping in which the object slips
after closing the hand, so that the measured force drops to zero while the exoskeleton
remains still closed.
390 J. A. Dı́ez et al.
The dynamic response of the sensor along the time gives a much clearer idea
of how the task is being performed. Figure 3 shows two examples corresponding
to a successful grasping and a failure.
4 Conclusion
These results show that the installation of only one sensor can be useful to discern
successful or unsuccessful grasping and can improve the reliability of a hand
exoskeleton in assistance environments. Additionally, this force measurement
might be used as a feedback of closed-loop control system to allow the robotic
device to manipulate complex objects.
However, authors consider that the quality and usefulness of the interaction
force feedback can be greatly improved by adding sensors in each finger so that
grasping stability can be evaluated. The inclusion of similar sensors to the used
in this research would increase the cost of the device in an unaffordable way. The
development and use of sensors like the mentioned in the introduction sections
may allow to overcome this inconvenience and help the design of more practical
and affordable exoskeletons.
References
1. Fontana, M., Marcheschi, S., Salsedo, F., Bergamasco, M.: A three-axis force sensor
for dual finger haptic interfaces. Sensors 12, 13598–13616 (2012)
2. Palli, G., Pirozzi, S.: A miniaturized optical force sensor for tendon-driven mecha-
tronic systems: design and experimental evaluation. Mechatronics 22(8), 1097–1111
(2012)
3. Jeong, S.H., Lee, H.J., Kim, K.R., Kim, K.S.: Design of a miniature force sensor
based on photointerrupter for robotic hand. Sens. Actuators A Phys. 269, 444–453
(2018)
4. Dı́ez, J.A., Catalán, J.M., Blanco, A., Garcı́a-Perez, J.V., Badesa, F.J., Gacı́a-Aracil,
N.: Customizable optical force sensor for fast prototyping and cost-effective appli-
cations. Sensors 18(2), 493 (2018)
5. Lauretti, C., Cordella, F., Ciancio, A.L., Trigili, E., Catalan, J.M., Badesa, F.J.,
Garcia Aracil, N., : Learning by demonstration for motion planning of upper-limb
exoskeletons. Front. Neurorobotics 12(5) (2018)
6. Dı́ez, J.A., Blanco, A., Catalán, J.M., Bertomeu-Motos, A., Badesa, F.J., Garcı́a-
Aracil, N.: Mechanical design of a novel hand exoskeleton driven by linear actuators.
In: Iberian Robotics Conference, pp. 557–568. Springer, Cham, November 2017
7. Dı́ez, J.A., Blanco, A., Catalán, J.M., Badesa, F.J., Lled’ó, L.D., Garcı́a-Aracil, N.:
Hand exoskeleton for rehabilitation therapies with integrated optical force sensor.
Adv. Mech. Eng. 10(2), 1687814017753881 (2018)
XoSoft Connected Monitor
(XCM) Unsupervised Monitoring
and Feedback in Soft Exoskeletons of 3D
Kinematics, Kinetics, Behavioral Context
and Control System Status
Chris T. M. Baten1(&), Wiebe de Vries1, Leendert Schaake1,

Juryt Witteveen1, Daniel Scherly2, Konrad Stadler1,
Andres Hidalgo Sanchez3, Eduardo Rocon3, Danny Plass-Oude Bos4,
and Jeroen Linssen4
1
Roessingh Research and Development, Enschede, The Netherlands
C.Baten@RRD.nl
2
ZHAW, Winterthur, Switzerland
3
CSIC, Madrid, Spain
4
Saxion University of Applied Sciences, Enschede, The Netherlands
Abstract. Intelligent soft exoskeletons are developed to be used unsupervised

and continuously on a large scale in normal daily situations. As they miss the
stiffness of the structural components of traditional robotic devices, traditional
robotic movement assessment are rendered useless, as they assume structural
segment rigidity. This all requires a radical different approach towards (remote)
monitoring and feedback of data relevant to a host of different type users:
clinicians and therapists responsible for training and well-being of patient,
caregivers, maintenance technicians and even the exoskeleton’s control system.
This paper proposes such a system, one implementation of which is developed
and tested within the XoSoft soft exoskeleton project. It provides continuous
remote (partly IMMU based) assessment of 3D kinematics and kinetics, control
system activity, subject activity pattern and derived movement pattern param-
eters. It also is structured in a maximally flexible way facilitating the ever-
shifting, optimal distribution of functional software modules over more
peripheral and central hardware to accommodate for fast changes in specifica-
tions and technical and practical constraints.
1 Introduction
Novel soft exoskeleton technology facilitate large scale application for (elderly) people
with mild mobility and balance problems. The shift from ‘rigid’ to ‘soft’ structural
component puts new challenges to kinematics data assessment in exoskeletons.
This project has been supported by Horizon 2020 program project XoSoft (688175) and by
projects AmbuLab, FreeMotion and Fusion from Ministry of Economic Affairs, the Netherlands.

https://doi.org/10.1007/978-3-030-01887-0_75
392 C. T. M. Baten et al.
In traditional robotics and ‘rigid’ exoskeletons the high (assumed) stiffness of the
structural segments allows for meaningful kinematic assessment based on data acquired
in the joints and/or in the actuators. For soft exoskeletons (and light weight robotics) this
stiffness is absent and meaningful kinematics assessment can only be derived with
methods very similar to kinematics assessment in human subjects not wearing
exoskeletons. So, moving from ‘hard’ to ‘soft’ exoskeletons causes a shift from traditional
robotics paradigms for kinematics and kinetics assessment towards typical ambulatory
3D analysis of human movement paradigms as applied in regular rehabilitation, sports
and ergonomics.
The foreseen scale of (unsupervised) use of soft exoskeleton poses an additional
challenge to wearable kinematics assessment as a natural consequence of the limited
space/weight bearing capacity in soft exoskeletons. This requires an optimal distribu-
tion over multiple devices of all computational hardware and software, that is required
for the (remote) monitoring of kinematics, soft sensor data plus control systems
status/action data, plus all required calculations and feedback for kinematics, kinetics,
subject behavior and intention estimation.
This calls for an optimization of functional software modules distribution over local
and peripheral locations under constraints of functional specifications, bandwidth and
power consumption limitations and maximally acceptable delay of information
required in ‘real time’. Also, in the high tech driven young field of soft exoskeleton
research technology and solutions are developing quickly, causing the functional
specifications and constraints to change strongly ‘by the day’, causing continuously
shift in optimal solution.
Fig. 1. Schematic indication of the sensing in XoSoft soft exoskeleton. Internal IMMUs (4):
white boxes on both Tibia and outside of upper legs, External IMMUs (8+): orange boxes on
same as internal IMMUs plus pelvis (backside) and thorax, soft joint angle sensor strips (4):
green over knees and ankles, foot pressure sensors (8): green in wireless embedded insoles
XoSoft Connected Monitor (XCM) Unsupervised Monitoring 393
The foreseen continuous and unsupervised mode of usage of intelligent soft

exoskeletons in daily life also calls for continuous automated assessment of usage
context, in the form of the (intended) user activity pattern, to increase versatility,
robustness and safety of its control strategies.
1.1 Research Question

How can remote automated monitoring of 3D kinematics and kinetics, control system
actions and subject activity pattern and intention in a scalable number of free roaming
unsupervised soft exoskeletons be optimally organized to keep up in an ever changing
combination of specifications, capabilities and constraints? (Fig. 1)
2 Methods
This paper proposes a flexible and scalable solution for (remote) monitoring and
feedback for 3D body segment kinematics, kinetics, subject activity and intention
detection, additional soft sensor data and control system state and subject activity
pattern, that is capable to deliver all required functionality, while dealing with all
technical/practical constraints in a maximally flexible way, by allowing easy relocation
of functional software components from local to peripheral hardware. Feedback is
generated numerical (to control system), through graphical user interfaces on tablets or
PCs based on specific authorized user roles or vibro-tactile (to patient).
This solution is initially developed and implemented for the XoSoft soft-
exoskeleton system developed in the EU-Horizon 2020 project of the same name.
It includes proof central data storage and offers data access facilities to users with
different roles and authority, stored and protected according to the latest European legal
and ethical legislation and also provides exoskeleton control system configuration
options plus data feedback to the soft exoskeleton control system or the user.
Manager, a Data Manager and a Process Manager module, a series of data
acquisition modules (Measurement Clients), a series of User Information Clients, a
series of Data Analysis clients and a central database. The Connection Manager
manages all communication and data channels between modules (in TC/IP or UDP).
This allows for great freedom of moving functional software modules between more
peripheral and more central hardware component in the XoSoft ecosystem (Fig. 2).
The Data Manager handles all data storage and retrieval requests from clients and
sends data analysis requests to the Process Manager, which then invokes recursive
chains of Data Analysis clients before data is stored or delivered to User Information
clients, e.g. on a tablet in the hands of caregiver, rehabilitation physician, physical
therapist, orthotist or maintenance technician. The Process Manager generates recur-
sively Analysis Module call chains, that process data requested from the Data Manager
and dissolve after returning results to the Data Manager. In a typical implementation
the Manager modules, database and Analysis Clients are all running on servers inside a
central data center and Measurement Clients and User Information Clients are located
more peripherally on multiple devices close, or on, the subject or at the site of a user
(physician, therapist, orthotist, care giver, scientist, technician, manager). Therefore an
394 C. T. M. Baten et al.
Fig. 2. Schematic XoSoft connected motion monitoring and feedback platform. The core
elements are the Connection Manager, Data manager plus central Database and the (data
analysis) Process Manager. An authentication and authorization layer separates user related
modules from the rest of the system. Measurement clients provide multiple data input facilities,
User Information Clients provide information and feedback to primary, secondary and tertiary
users. Data Analysis clients provide body segment 3D kinematics and kinematics estimates,
activity pattern estimates and movement pattern parameters.
authentication and authorization layer is placed between these clients and the Con-
nection Manager. A machine to machine token based authorization is present between
all modules in the central server site.
Easy relocation of functionality between more peripheral and more central hard-
ware is facilitated by using the same operating system (Windows 10 OS) and the same
functional programming language (G) and development environment (LabVIEW) for
all server and client modules in combination with using centrally organized TCP/IP and
UDP based data channeling. This also facilitates running all modules on one machine
in one editor/debugging environment, providing optimal means for agile and robust
development and testing. Typically, functionality is relocated by simply moving or
copying the functional module inside an Analysis Client to a User Information Client
or Data Acquisition Client. All user interfaces were developed under continuous user
feedback.
3 Results
The XCM is currently implemented for, and tested in, monitoring and feedback in
several clinical XoSoft soft exoskeleton patient validation studies in Germany,
Switzerland and the Netherlands. It remotely monitors simultaneously internal XoSoft
IMMUs (Technaid), custom soft joint angle sensors, a series of foot pressure sensors
and control system activity data plus an additional external wireless IMMU based 3D
kinematics assessment system [1, 2] (Xsens Awinda), It provides facilities to synch
with an external recording system (e.g. Vicon motion analysis system in gait laboratory
testing). In the current implementation the analysis components deliver detailed lower
body and trunk segment and joint 3D kinematics estimates, gait spatial-temporal
parameters, minimal foot clearance estimates. Also several artificial intelligence based
XoSoft Connected Monitor (XCM) Unsupervised Monitoring 395
methods of activity pattern recognition [3, 4] are available, tested and under continuous
development, as well as methods for estimating internal low back load exposure
(requiring additional EMG recording of a few trunk muscles) that were already used [1]
to estimate the effect of a passive, rigid back support exoskeleton rigid on the actual
physical spinal load decreasing effect of the device.
4 Conclusion
The concept of reconfigurable remote monitoring has shown its value already by
allowing smooth adaptation to the quickly changing specifications and constraints
during the XoSoft exoskeleton development.
References
1. Baten, C.T.M., van der Aa, R., Verkuyl, A.: Effect of wearable trunk support for working in
sustained stooped posture on low back net extension moments. In: International Conference
on ISB, Glasgow (2015)
2. Baten, C.T.M.: Advancements in sensor-based ambulatory 3D motion analysis. J. Biomech.
40, S422–S422 (2007)
3. Recher, F., Banos, O., Nikamp, C., Schaake, L., Baten, C., Buurke, J.: Optimizing activity
recognition in stroke survivors for wearable exoskeletons. In: 7th IEEE International
Conference on Biomedical Robotics and Biomechatronics (2018)
4. Wassink, R.G.V., Baten, C.T.M.: Classifying human lifting activities automatically by
applying Hidden Markov Modeling technology. J. Biomech. 40, S428 (2007)
Tactile and Proximity Servoing
by a Multi-modal Sensory Soft Hand
John Nassour(B) and Fred H. Hamker
Artificial Intelligence, Computer Science, Chemnitz University of Technology,

Chemnitz, Germany
john.nassour@informatik.tu-chemnitz.de
Abstract. We present the manufacturing and the implementation of a

multi-modal sensory soft hand for the interaction with conductive and
non-conductive objects. The hand sensors were mounted on two fabric
layers with three sensory modularities: touch, proximity, and curvature.
Servoing behavior is generated based on the estimation of the center of
touch (force sensitive resistor) and the center of proximity (proximity
sensitive capacitor). Results are presented on human subjects wear the
hand, and a set of vibration motors that work as haptic feedback for the
center of stimulation. Driven by vibration, the system guides the subject
to explore conductive objects. Servoing behavior is generated based on
the estimation of the center of stimulations without visual feedback.
1 Introduction
Wearable robots aim to extend or compensate human capabilities in sensing and
acting with the dynamic world. Human skin hosts different sensory modalities
that inspire roboticists in building sensory skills for robots [1]. The diversity of
sensory modalities in robotics improves the way the robot perceive the world and
interact with its dynamics. Different multi-modal tactile sensor structures have
been proposed for both rigid [2] and soft robots [3–5]. Soft interaction requires
appropriate sensitive modalities that provide a bundle of information about the
interaction. For example a robotic hand equipped with tactile sensors provides
information if an object is touched, with temperature sensors it can distinguish
cold from warm objects, and with finger curvature sensors it can sketch the
shapes of grasped objects. However, the set-up of multi-modal soft sensors on
soft robots remains challenging since sensors may share same contact points. In
this paper, we propose a multi-modal sensory soft hand that hosts three different
modalities: pressure, proximity, and finger curvature, see Fig. 1. The soft sensors
were mounted on two pieces of fabric that are attached to the soft actuator.
Proximity modularity for conductive objects was produced thanks to 12 conduc-
tive fabric pieces sewed to the first cloth fabric layer forming four sensory arrays
(three sensors per finger). Pressure and curvature modularities were introduced
thanks to the superimposition of piezoresistive sensors in a multilayer set-up.
Four pressure sensor arrays were proposed (four sensors per finger). One sensor
https://doi.org/10.1007/978-3-030-01887-0_76
Tactile and Proximity Servoing by a Multi-modal Sensory Soft Hand 397
was also hosted in each finger, it is dedicated for curvature estimation. There-
fore, each finger of the hand hosts eight sensors three for proximity, four for
pressure, and one for curvature. Section 2 represents the methods used to esti-
mate the curvature and the center of stimulation. Section 3 demonstrates the
servoing experiment. We present the conclusion in Sect. 4.
Fig. 1. The multi-modal sensory soft hand under actuation with GUI that demon-
strates the sensors activation for proximity and tactile modularities (top). Capacitive
sensor layer (conductive fabric and conductive threads) are mounted on a blue fabric.
Piezoresistive sensor layer (Velostat and conductive threads) mounted on a yellow fab-
ric. The control diagram of the hand and a human subject wears the hand and a set
of four vibration motors (middle). Some of the manufacturing processes pictures are
presented at the bottom, see the video link [7].
The manufacturing process of the hand is presented in the video link [7]. Three
sensory modularities have been implemented with the actuator. The actuator
TM
is made of silicone rubber Ecoflex 00-30. Conductive threads are used to rout
the capacities and the resistors into the related electronic boards (MPR121,
398 J. Nassour and F. H. Hamker
MCP3008 respectively). The curving estimation is calculated from (1):
∂VRC ∂VRC
dVRC = dC + dVRP 0 + dVRP 1
∂VRP 0 ∂VRP 1
(1)
∂VRC ∂VRC
+ dVRP 2 + dVRP 3 ,
∂VRP 2 ∂VRP 3
where C is the curving estimation after eliminating the applied forces. The par-
tial derivatives represent the rate of change of curvature sensor reading caused
only by tactile forces. They are obtained through linear regression algorithm. V
is the ADC read for pressure and curvature sensors (RP 0 , RP 1 , RP 2 , RP 3 , RC ).
We compute the overall center of proximity and the overall center of pressure
based on readings from capacitive and piezoresistive sensors respectively while
considering the positions of these sensors (yc , zc ) with respect to the center point
of the hand, see (2) for yc in y coordinate. zc is calculated in a similar way.
Ny Ny

fy = fi , yc = fy−1 fi · yi (2)
i=1 i=1
where fi is the normalized reading for sensor i. yi is the position of this sensor
on y − axis. Ny is the number of sensors positioned on y − axis. The overall
proximity and the overall pressure are computed as in (3).
Ny +Nz

f= fi (3)
i=1
where Ny + Nz is the number of sensors (16 for pressure sensors, and 12 for
capacitive ones). Four vibration motors are used as haptic feedback for the center
of stimulation (either pressure or proximity). Each motor is dedicated to one
direction in y-z plane. Each motor is controlled with a PWM signal that reflects
the amount of activation of the related sensors. Each motor vibrates with respect
to four pressure sensors in case of tactile servoing, and three capacitive sensors
in case of proximity servoing. Motors are placed in a sequential order on the arm
of human subject wears the hand, see Fig. 1. In order to provide a recognizable
spatial stimulation, the distance between two successive motors is 4.5 [cm] (as
suggested to be more than 2.4 [cm] in [6]).
3 Results
Figure 1 shows a human pressing on two fingers of the sensory hand in actuation.
The center of pressure and the center of proximity are presented by the floating
red dots. Left GUI shows the normalized activations of 12 proximity sensitive
capacitors. Right GUI shows 16 force sensitive resistors and 4 curvature sensors.
The radius of the floating dot represents the overall activation, see (3), while the
positions represent the side of stimulations, see (2), it is used to guide the arm
Tactile and Proximity Servoing by a Multi-modal Sensory Soft Hand 399
movement. To demonstrate the servoing behavior, we run multiple experiments

on human subjects wear the hand and the set of four vibration motors. Subjects
are first asked to associate the stimulation with the related vibrations, this is
done by self-touching the worn hand fingers separately. Then, two attempts have
been performed in the absence of visual feedback which has been eliminated by
hiding the experiment set-up from the subjects’ view. Subject were asked to
move the worn hand over a transparent surface where a conductive material
forming a path was hidden and covered by white papers from the subject side.
Subjects were asked to follow the conductive material path based on the vibra-
tion motors activations. A straight conductive path has been used in the first
attempt, while an angular path in the second. Figure 2 illustrates the hand tra-
jectories of two attempts for four subjects. We notice that servoing behaviors
have been improved in the second attempt for subjects 1, 2, and 4 (e.i., areas
between hand trajectories and the conductive paths are reduced).
Fig. 2. Experiment set-up (top). Hand trajectories (in green) are presented for four
subjects. The area between the hand and the conductive paths from the start of vibra-
tion until the end of the task are in yellow. Video link [7].
4 Conclusions
We present the design, the fabrication, and the set-up of a multi-modal sen-
sory soft hand. Vibration motors are used as tactile and proximity displays for
400 J. Nassour and F. H. Hamker
the servoing task. The system robustness was presented through an experiment
where a human subject wears the hand and follows non visible conductive paths.
Although the sensory soft hand has no anthropomorphic shape, we could present
that a wearable hand does not require a visually guided movement. The logistic
arm movements can be visually guided, while the servoing movements are sen-
sory guided. Future work aim to transfer this approach into an anthropomorphic
hand and perform grasping task.
References
1. Dahiya, R.S., Metta, G., Valle, M., Sandini, G.: Tactile sensing-from humans to
humanoids. IEEE Trans. Robot. 26(1), 1–20 (2010)
2. Mittendorfer, P., Cheng, G.: Integrating discrete force cells into multi-modal arti-
ficial skin. In: IEEE-RAS International Conference on Humanoid Robots, pp. 847–
852, Osaka (2012)
3. Park, Y.L., Chen, B.R., Wood, R.J.: Design and fabrication of soft artificial skin
using embedded microchannels and liquid conductors. IEEE Sens. J. 12(8), 2711–
2718 (2012)
4. Din, S., Xu, W., Cheng, L.K., Dirven, S.: A stretchable multimodal sensor for soft
robotic applications. IEEE Sens. J. 17(17), 5678–5686 (2017)
5. Tada, Y., Hosoda, K., Yamasaki, Y., Asada, M.: Sensing the texture of surfaces by
anthropomorphic soft fingertips with multi-modal sensors. In: IEEE/RSJ Interna-
tional Conference on Intelligent Robots and Systems, pp. 31–35 (2003)
6. Jones, L.A., Sofia, K.: Measuring surface wave propagation during vibrotactile stim-
ulation. In: IEEE Haptics Symposium (HAPTICS), pp. 457–461. Vancouver, BC
(2012)
7. The paper descriptive video. https://youtu.be/96Y8KoyeJPs
Improved Fabrication of Soft Robotic Pad
for Wearable Assistive Devices
Yi Sun1,3, Aaron Jing Yuan Goh2, Miao Li1, Hui Feng1,

Jin Huat Low2, Marcelo H. Ang Jr.1, and Raye Chen Hua Yeow2(&)
1
Department of Mechanical Engineering, National University of Singapore,
9 Engineering Drive 1, Singapore 117575, Singapore
2
Department of Biomedical Engineering, National University of Singapore,
4 Engineering Drive 3, Singapore 117583, Singapore
rayeow@nus.edu.sg
3
NUS Graduate School for Integrative Sciences and Engineering, Singapore,
Singapore
Abstract. Soft Robotic Pad (SRP), as a new class of soft pneumatic actuator
(SPA), is a two-dimensional pad-like SPA that can be programmed to achieve
different surface morphing. Recently, the successful fabrication has proven the
feasibility of functional SRPs. However, there are issues to be solved so that the
SRP can withstand high pressure for practical applications. This paper, based on
the first version of the SRP fabrication method, presents some modifications in
the method and discusses their pros and cons. Firstly, the incorporation of
stiffness customization and patterning method into the SRP fabrication not only
simplifies the SRP morphing design, but also makes many morphing modalities
possible. Furthermore, the use of larger carbon-fiber rods and the channel filling
process improve the SRP strength, which qualifies them to many applications.
As an envisioning step, we presents a design of a wearable assistive SRP for
elbow flexion. With this fabrication method, the SRP with its unique shape and
morphing capabilities has great potential in wearable robotics especially for
human joint rehabilitation.
1 Introduction
Silicone based soft pneumatic actuator (SPA), as a signature genre of soft robotics, has
been prevailing for a few years [1–5]. Recent developments in fabrication have
diversified the SPAs in many ways [4, 6–9]. Despite the diversification, one common
feature in the existing SPAs is that they mostly shaped like one straight silicone rod,
regardless of their fabrication methods. This one-dimensional design is favored as it is
easy to fabricate. However, there are limitations: monotonous motion types due to lack
of dimension, and poor force stability because of the low torsion resistance [10].
Recently, two-dimensional flat-shaped SPA in flat shape has emerged in concomi-
tant with certain applications [5, 11–13], however, they are mostly the combination of
Research supported by MOE Tier 2 Grant (R-397-000-281-112) awarded to Dr. Chen Hua Yeow.

https://doi.org/10.1007/978-3-030-01887-0_77
402 Y. Sun et al.
one-dimensional SPAs. The true two-dimensional SPA, termed as soft robotic pad
(SRP), was developed in [14] with a novel and well-designed fabrication processes.
Only silicone and cotton fiber were used and the fiber were arranged in different formats
to constrain the SRP thickness and program the 2D surface morphing. Three types of
SRPs were fabricated, bending, saddle and wrapping, to demonstrate the feasibility and
the flexibility of the fabrication technique.
However, the key issue is that the SRP sometimes breaks at very low pressure and
thus severely affects its application prospects. In this paper, we present some modifi-
cations in the fabrication processes and articulate the key benefits and some minor
drawbacks. Conclusively, the new fabrication simplifies the SRP motion design,
widens the size range up to 40 cm in one dimension and, most importantly, strengthen
the SRP by at least 30%. The new fabrication essentially facilitate the utilization of
SRP in many applications, especially wearable assistive devices. As a demonstration,
we also present a design of a wearable SRP for elbow flexion assist.
2 Improved Fabrication of Soft Robotic Pad
Compared to the previous fabrication method [14], we adopt a method called stiffness
customization and patterning (SCP) [9] which is to adhere patterned fabric sheets to the
top and bottom surface of the SRP to for motion program. This method not only
considerably simplifies the application of the surface constraint, but also makes the
motion diversification indeed achievable. In addition, we utilize laser cutting to prepare
the patterned fabric sheets.
Fig. 1. Improved SRP fabrication. a:mold preparation, b:first layer molding, c:second layer
molding, d:halfway SRP after demolding, e:channel filling process, f:final SRP after wall sealing.
Improved Fabrication of Soft Robotic Pad 403
As reported in [14], the SRP from the old fabrication suffers from low breaking
pressure due to the channel inflation and fiber detachment. To address these issues, our
solution is to utilize thicker CF rod (3 3 mm2) instead of the thin one (U1 mm).
Using larger CF rod makes the winding frame stronger and supports longer fiber
winding (up to 40 cm). During the fabrication, an addition step is needed after the
removal of the CF rod which is to fill the channel with fresh silicone. After the curing
process, the channels are finally eliminated and the SRP strength can be improved by
30% according to our failure tests. Figure 1 illustrates the new fabrication technique
that includes all the modifications.
With the utilization of larger CF rod, the fabrication becomes more tedious because
of the time-consuming channel filling process. Another drawback is that the final
thickness of the SRPs becomes slightly thicker. However, overall, the benefits of the
new fabrication surly outweighs the drawbacks.
3 Design of a Soft Assistive Device for Elbow Flexion
Figure 2 shows the components (top) and final assembly (bottom) of the device. Similar
to other SRPs, this elbow-assisting SRP also features a long thin silicone body with a
through-all air chamber, vertical fiber matrix for thickness constraint, a non-patterned
fabric to constrain the top surface and a patterned fabric to generate the desirable SRP
motion as the combination with the top surface constraint. From the bottom fabric pattern,
we can see that, at the two ends, the SRP can curl up, which is consciously designed with
the purpose of securing the upper arm and forearm during actuation. On the other hand,
the major bending in the middle will assist the actual elbow flexion motion. Moreover, the
through-all chamber in the SRP can achieve an entire-body stiffening during actuation so
that the bending of the SRP can be achieved without collapsing, and the force can be
effectively applied onto the upper arm and the forearm. With this SRP, we can achieve
multi-motion in one body for a more effective assist.
Fig. 2. Design of a wearable assistive SRP for elbow flexion.

404 Y. Sun et al.
4 Conclusion
This paper presents the improved SRP fabrication method with some modifications to
the previous fabrication processes: using patterned fabric for surface constraint, using
larger CF rods and an additional channel filling process. The pros and cons of the
modification has been discussed and, conclusively, the pros outweigh the cons as the
modifications simplify the SRP motion programming, support larger SRP size and most
importantly enable the SRPs to withstand higher pressure.
Acknowledgment. Research supported by MOE Tier 2 Grant (R-397-000-281-112) awarded to

Dr. Chen Hua Yeow. The authors would like to thank NUS Graduate School for Integrative
Sciences and Engineering for providing scholarship to support Yi Sun’s Ph.D. study.
References
1. Ilievski, F., Mazzeo, A.D., Shepherd, R.F., Chen, X., Whitesides, G.M.: Soft robotics for
chemists. Angew. Chem. 123(8), 1930–1935 (2011)
2. Shepherd, R.F., Ilievski, F., Choi, W., Morin, S.A., Stokes, A.A., Mazzeo, A.D., Chen, X.,
Wang, M., Whitesides, G.M.: Multigait soft robot. Proc. Natl. Acad. Sci. 108(51), 20400–
20403 (2011)
3. Sun, Y., Song, Y.S., Paik, J.: Characterization of silicone rubber based soft pneumatic
actuators. In: 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems
(IROS), pp. 4446–4453. IEEE (2013)
4. Rus, D., Tolley, M.T.: Design, fabrication and control of soft robots. Nature 521(7553),
467–475 (2015)
5. Roche, E.T., Horvath, M.A., Wamala, I., Alazmani, A., Song, S.E., Whyte, W., Machaidze,
Z., Payne, C.J., Weaver, J.C., Fishbein, G., et al.: Soft robotic sleeve supports heart function.
Sci. Transl. Med. 9(373), eaaf3925 (2017)
6. Sun, Y., Song, S., Liang, X., Ren, H.: A miniature soft robotic manipulator based on novel
fabrication methods. IEEE Robot. Autom. Lett. 1(2), 617–623 (2016)
7. Liang, X., Sun, Y., Ren, H.: A flexible fabrication approach toward the shape engineering of
microscale soft pneumatic actuators. IEEE Robot. Autom. Lett. 2(1), 165–170 (2017)
8. Polygerinos, P., Wang, Z., Overvelde, J.T., Galloway, K.C., Wood, R.J., Bertoldi, K.,
Walsh, C.J.: Modeling of soft fiber-reinforced bending actuators. IEEE Trans. Robot. 31(3),
778–789 (2015)
9. Sun, Y., Yap, H.K., Liang, X., Guo, J., Qi, P., Ang Jr., M.H., Yeow, C.-H.: Stiffness
customization and patterning for property modulation of silicone-based soft pneumatic
actuators. Soft Robot. 4(3), 251–260 (2017)
10. Sun, Y., Liang, X., Yap, H.K., Cao, J., Ang, M.H., Yeow, C.-H.: Force measurement
towards the instability theory of soft pneumatic actuators. IEEE Robot. Autom. Lett. 2, 985–
992 (2017)
11. Park, Y.L., Chen, B.R., Majidi, C., Wood, R.J., Nagpal, R., Goldfield, E.: Active modular
elastomer sleeve for soft wearable assistance robots. In: 2012 IEEE/RSJ International
Conference on Intelligent Robots and Systems (IROS), pp. 1595–1602. IEEE (2012)
12. Robertson, M.A., Sadeghi, H., Florez, J.M., Paik, J.: Soft pneumatic actuator fascicles for
high force and reliability. Soft Robot. 4, 23–32 (2016)
Improved Fabrication of Soft Robotic Pad 405
13. Wirekoh, J., Park, Y.-L.: Design of flat pneumatic artificial muscles. Smart Mater. Struct. 26
(2016)
14. Sun, Y., Guo, J., Miller-Jackson, T.M., Liang, X., Ang, M.H., Yeow, R.C.H.: Design and
fabrication of a shape-morphing soft pneumatic actuator: soft robotic pad. In: 2017
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 6214–
6220. IEEE (2017)
The Exosleeve: A Soft Robotic Exoskeleton
for Assisting in Activities of Daily Living
Rainier F. Natividad1,2, Sin Wai Hong1, Tiana M. Miller-Jackson1,2,

and Chen-Hua Yeow1,2,3(&)
1
Department of Biomedical Engineering,
National University of Singapore (NUS), Singapore, Singapore
bieych@nus.edu.sg
2
Advanced Robotics Center at NUS, Singapore, Singapore
3
Singapore Institute for Neurotechnology at NUS, Singapore, Singapore
Abstract. The shoulder is one of the most complex joints in the human body
due to its extensive range of motion. Exoskeletons must accommodate the
shoulder’s capabilities in order to be effective. Soft robotic actuators have found
their way into upper limb exoskeletons; however, current designs do not provide
a mechanism for adjusting the structure of the exoskeleton in order to tailor-fit it
onto the user. We have created a modular, pneumatic, soft robotic exoskeleton
that is capable of mechanical and structural reconfiguration: the Exosleeve.
Reconfiguration provides the potential to ensure that the device correctly mat-
ches the user’s requirements. The ability of the Exosleeve to provide assistance
in performing limb motion was preliminarily assessed through a pilot test of
three healthy subjects. Subjects were instructed to perform shoulder abduction
exercises while surface electromyography measured muscle activation under
various conditions. The test showed that utilization of the Exosleeve reduces
muscle activation.
1 Introduction
The complexity of the human shoulder presents unique challenges when applying
traditional robotic components in exoskeletons. This has resulted in the development of
a number of exoskeletons that employ novel power transmission systems, typically
utilizing soft robotic techniques [1]. A prominent archetype of soft robots, fluidic
actuators [2, 3] utilize fluid pressure to create linear or rotational motions; however,
most fluidic actuators are limited to a single actuation profile due to their construction.
Exoskeleton actuators must agree with the unique anthropomorphic measurements of
various users in order to minimize the risk of injury while maximizing the device’s
mechanical efficacy. In their current state, it is therefore difficult to adapt an
exoskeleton utilizing a fluidic actuator to the anthropomorphic measurements of
multiple patients; hence, a specialized actuator will have to be constructed for each
patient.
To attempt to circumvent this shortcoming, we have created a soft robotic shoulder
exoskeleton powered by a modular, pneumatic actuator based on the design first pre-
sented in [4]. The primary feature of the exoskeleton is to provide the ability to perform

https://doi.org/10.1007/978-3-030-01887-0_78
The Exosleeve: A Soft Robotic Exoskeleton 407
structural adjustments on the exoskeleton in order to alter and adapt its mechanical
performance based on the requirement of a particular user. A pilot test has been
conducted to measure its effect on muscular effort, in accordance with the protocol
approved by the university’s institutional review board (N-17-103).

2.1 Exoskeleton and Actuator Design
Figure 1 illustrates the actuator’s structure and its integration onto the exoskeleton. The
actuator is an assembly of two primary components: a fully fabric flexible spine, and a
series of inflatable, modular, replaceable fabric bladders. The bladders are attached
through custom-made quick-release connectors, while the modules possess their own
pneumatic lines to improve reliability and customizability. The modules can be
replaced and resized depending on the musculoskeletal structure and the biomechanical
requirements of the user.
Bladders
Spine
A
B C
Fig. 1. (A) A 3-module actuator is assembled. (B) The fabric bladders unfold as they are
inflated. (C) A test subject performs shoulder abduction while assisted by the Exosleeve.
Two actuators are integrated: an abduction actuator, and an adduction actuator. The
abduction actuator is situated on the lateral side of the torso and extends to the upper
arm, while the adduction actuator is placed on the shoulder, extending from the base of
the neck to the elbow. The modules in each actuator are pneumatically connected as to
ensure that they activate simultaneously. However, the design of the device allows for
408 R. F. Natividad et al.
the individual inflation of the modules, but investigation of the feature is beyond the
current scope of the study.
2.2 Healthy Subject Testing

Three healthy subjects were recruited to perform a preliminary assessment of the
Exosleeve’s capability to perform shoulder abduction. The subjects were instructed to
sit on a chair and place their upper body in the standard anatomical position, but with
the palm facing towards the torso. They were then instructed to perform 90° shoulder
abduction of the right arm over 6 s, followed by adduction at the same speed. All the
movements were monitored by a passive, optical, motion capture system (Vicon
Motion Systems). Surface electromyography (sEMG) sensors were attached to the
anterior deltoid, posterior deltoid and the pectoralis major (Delsys, Trigno Wireless).
Marker and sEMG signals were sampled at rates of 100 Hz and 2 kHz respectively.
The subjects were asked to perform three sets of the exercise under different conditions.
Set A saw subjects do the motion unaided. Set B asked the subjects to repeat the
motions while wearing the Exosleeve; during this condition, the exoskeleton was
unpowered. Finally, the subjects performed Exosleeve-powered assisted shoulder
motion during set C. The exoskeleton’s pressure input was programmed to increase
linearly over six seconds until it reached 50 kPa during shoulder abduction, while it
was programmed to function inversely during adduction. The subjects performed three
repetitions of the motion during each set.
3 Results
The acquired sEMG data were full-wave rectified and were passed through a third-
order Butterworth filter with a cutoff frequency of 5 Hz. Kinematic data from set B and
C were compared against set A to ensure that the subjects performed the motions
consistently. Figure 2 shows the voltage readings from the lateral deltoid as the
abduction-adduction motions were performed. As expected, sEMG gradually increased
as abduction was performed and subsequently decreased in a similar manner during
adduction. For set A and B, peak voltage readings corresponded to 90° shoulder
Fig. 2. The surface electromyography readings of the subjects’ lateral deltoid are presented. The
readings over three abduction-adduction motions are averaged. The results from the three test sets
are superimposed in each graph. 50% motion progress corresponds to 90° of shoulder abduction
The Exosleeve: A Soft Robotic Exoskeleton 409
abduction while the peak values for set C were skewed towards the latter half of the
motion. The results from the three subjects clearly show a reduction in the peak sEMG
voltage during set C motions, indicating a reduction in the required muscular activation
during both abduction and adduction. However, readings from set B featured higher
voltage readings compared to both set A and C.
4 Discussion and Future Work
The intertwining nature of a soft robotic actuator’s structure and mechanical operation
makes it difficult to address potential problems that may arise from the unique
anthropomorphic and mechanical requirements that various users possess. A modular,
soft actuator was integrated into an exoskeleton in order to provide shoulder support to
a user whilst maintaining the ability to reconfigure the actuator in accordance to that
user’s characteristics. A pilot test was conducted in order to ascertain if the Exosleeve
is able to reduce the required muscular effort to perform shoulder abduction by
attaching sEMG sensors to the lateral deltoid of the test’s participants.
The results of the test clearly indicate the Exosleeve’s capability to assist users in
executing shoulder motion. The skewed nature of the EMG signals during set C
activities suggest that there is a disjoint between the activation of profile of the Exo-
sleeve during the tests and the motion of subject. Nevertheless, such a result can be
remedied by providing users with a greater degree of control on the activation of the
exoskeleton as compared to the open-loop scheme that was employed. Subsequent tests
will be performed to verify these preliminary findings and to measure the efficacy of
mechanical reconfiguration on the operation of soft exoskeletons.
References
1. Gopura, R.A.R.C., Bandara, D.S.V., Kiguchi, K., Mann, G.K.I.: Developments in hardware
systems of active upper-limb exoskeleton robots: a review. Robot. Auton. Syst. 75, 203–220
(2016)
2. Simpson, C.S., Okamura, A.M., Hawkes, E.W.: Exomuscle: an inflatable device for shoulder
abduction support. In: 2017 IEEE International Conference on Robotics and Automation
(ICRA), pp. 6651–6657 (2017)
3. O’Neill, C.T., Phipps, N.S., Cappello, L., Paganoni, S., Walsh, C.J.: A soft wearable robot for
the shoulder: design, characterization, and preliminary testing. In: IEEE International
Conference on Rehabilitation Robotics, vol. 2129, pp. 1672–1678 (2017)
4. Natividad, R.F., Del Rosario, M.R., Chen, P.C.Y., Yeow, R.C.H.: A hybrid plastic-fabric soft
bending actuator with reconfigurable bending profiles. In: Proceedings of the IEEE
International Conference on Robotics and Automation, pp. 6700–6705 (2017)
Exoskeleton with Soft Actuation and Haptic
Interface
Ivanka Veneva(&), Dimitar Chakarov, Michail Tsveov,

and Pavel Venev
Mechatronics Department, Institute of Mechanics, Bulgarian Academy of

Sciences, Sofia, Bulgaria
veneva@imbm.bas.bg
Abstract. This work presents an active orthotic device with a wearable

structure corresponding to the natural motions of the human that can be used for
motion capturing and mobility assisting. The exoskeleton structure includes left
and right upper limb, left and right lower limb fabricated with light materials and
powered by pneumatic artificial muscles. The proposed exoskeleton provides
more than fifteen degrees of freedom and can operate in three modes: Motion
tracking and data exchange with virtual reality (VR); Haptic device with force-
feedback that can display sensory information from a virtual reality to the user;
Assistive and rehabilitation device in cases of impaired muscles.
The design and development has been carried out in Institute of Mechanics,
Bulgarian Academy of Sciences, Sofia, Bulgaria.
1 Introduction
The main purpose of the exoskeletons is to compensate the lack of force in the joints
and support the user’s body weight so as to minimize the loading on the joint supplying
assistive torque during dynamic activities. Exoskeletons are currently being developed
to assist in rehabilitation, to increase the mobility of the elderly or to support factory
workers while performing manual work. Haptic or force-reflecting interfaces are
robotic devices used to display touch or force-related sensory information from a
virtual or remote environment to the user. There are many prototypes of exoskeletons
with different mechanical structure and actuation [1, 2]. The known exoskeletons
working in virtual reality are commonly for one or both upper limbs [3, 4]. In most
cases, these devices are grounded and they provide a limited range of interaction in
virtual reality.
The main task of our work was to create an exoskeleton for the whole body with
soft wearable structure and anthropomorphic workspace including an exoskeleton for
upper limbs as a haptic device providing force feedback of the limbs during the
interaction in a virtual reality and an exoskeleton for lower limbs providing assistive
torques and control the virtual avatar movements.

https://doi.org/10.1007/978-3-030-01887-0_79
Exoskeleton with Soft Actuation and Haptic Interface 411
The whole exoskeleton includes left and right upper and lower limbs and central
structure for torso and waist. The Lower limb exoskeleton has 3 d.o.f. corresponding to
the natural motion of the human lower limb from the hip to the foot. The Upper limb
exoskeleton has 4 d.o.f. for the shoulder and elbow. The total number of actuated joints
is ha ¼ 4 þ 4 þ 3 þ 3 ¼ 14:
The Upper limb exoskeleton acts as a haptic device that displays force-related
sensory information from the virtual avatar. The Local control system interprets the
applied forces at the end effector of the avatar and sends the commands to drive the
corresponding segments of the exoskeleton. As a result a Joint torque control is
implemented in each joint [5] (Fig. 1).
Fig. 1. Real prototype: (a) Upper Limb Exoskeleton; (b) Lower Limb Exoskeleton
The joint control is achieved by producing appropriate antagonistic torques through

antagonistic action of pneumatic actuators applying Impedance Control (Fig. 2).
f
Ze Force
- Pa sensor
Filling
f dPa=│dP│+Pb dPa-Pa Valve
∆Q + Air PAM“a”
+ ∆q
+ Vending p0
dPb=0 -
Valve
+ dP=∆Q/r dP Joint Position
sensor
Qd Vending
- dPa=0 - Valve
Air PAM“b”
+ Filling p0
dPb=│dP│+Pa dPb-Pb
Valve
- Force
sensor
Pb
Fig. 2. Joint torque control

412 I. Veneva et al.
Agonist muscle group is active in one direction, when antagonist is passive and
vice versa. This is achieved by switching the filling valve to achieve the desired joint
moment. However, the muscle bundles with zero pressure always participate with a
force in the joint antagonistic equilibrium, as they are elastic. If we denote by Pa and
Pb forces in passive muscle groups (at zero pressure), the desired force of an active
muscle bundles is calculated by equations:
dPa ¼ Pb þ jdPj; dPb ¼ Pa þ jdPj; ð1Þ
dP ¼ DQ=r ð2Þ
where │dP│ is the force module set by the desired torque Qd in the joint, r is the
radius of the pulley, Zef represents the dynamic model of the exoskeleton in joint space.
The range of each muscle depends on the operating pressure. The torque regulation
in the joint position are analysed by means of pressure variation which are used to
control the moments, Q, in the joints.
Pa ¼ fa ðQ; qÞ; Pb ¼ fb ðQ; qÞ ð3Þ
Gravitational components of joint moments were rated according to the model for
gravity compensation of the exoskeleton arm.
3 Results
The System is actuated by pneumatic artificial muscles and can operate in three modes:
– Haptic device with force-feedback that can display sensory information from a
virtual reality to the user. The exoskeleton for upper limbs with force-reflecting
interface;
– Assistive Mode in cases of impaired muscles. The exoskeleton for lower limbs with
adjustable joint torque;
– Motion tracking system with data exchange with virtual reality (Fig. 3). A Graph-
ical User Interface, ExoInterface has been created for exoskeleton calibration,
communication and data exchange with the virtual reality. ExoInterface program is
the main interface between the exoskeleton’s controller, the virtual reality envi-
ronment and the user. Data transfer between Exoskeleton and Virtual avatar has
been performed in 3D Unity virtual Engine. For realizing the communication
between ExoInterface program and virtual reality a UB data exchange server was
used. HMD Oculus Rift is used for visualization in the virtual system.
Actual tests have been performed between IM-BAS in Sofia and UB in Barcelona
with very low latency between the Exoskeleton and the Virtual avatar. Data transfer is
performed at 170 Hz. The avatar and the exoskeleton were moving synchronously.
A Virtual GIM suitable for realization of the physical exercise (from a first person
perspective) has been created. Weight Lifting experiment has been completed in VR
where participants were either embodied in a weak or strong body. Real experiments
Exoskeleton with Soft Actuation and Haptic Interface 413
Fig. 3. Active exoskeleton for upper and lower limb actuated by pneumatic muscles exchange
data with Virtual Avatar
with the exoskeleton have been performed with 30 middle aged male volunteers. The
effect of embodiment in different types of bodies on personal strength has been
examined. In order to evaluate the progression during the sessions and to compare with
the control groups the measurements have been taken and following set of variables
recorded - forces applied and retrieved from the exoskeleton, tracking data for later
playback and evaluation, embodiment questionnaires.
4 Conclusion
The exoskeleton mechanical structure as a wearable device has to fulfil the design
requirements for low mass/inertia, safety, comfort, anthropomorphic extensive range of
motion, etc. One of the main achievements is the multifunctionality of the system
combining the motion tracking, haptic and rehabilitation device in one wearable
structure with force feed-back and anthropomorphic workspace covering the full range
of human motions.
The proposed system would be of great importance to people with limited mobility
for assistive and rehabilitation tasks both physically and mentally during human
interaction with virtual environments.
Acknowledgment. This research was supported by the European Commission through FP7
Integrated Project VERE [grant number 257695]; Bulgarian Science Found [grant number DN
07/9].
References
1. Kobayashi, H., Sho, H., Hirokazu, N.: Development and application of a muscle force
enhancement wear: muscle suit. In: Proceedings of the 11th Symposium on Construction
Robotics in Japan, pp. 93–100 (2008)
2. Mineo, I., Keijiro, Y., Kazuhito, H.: Stand-alone wearable power assist suit–development and
availability. J. Robot. Mechatron. 17(5), 575–583 (2005)
3. Carignan, C.R., Cleary, K.R.: Closed-loop force control for haptic simulation of virtual
environments. Haptics-e 1(2), 1–14 (2000)
414 I. Veneva et al.
4. Frisoli, A., Salsedo, F., Bergamasco, M., Rossi, B., Carboncini, M.: A force-feedback
exoskeleton for upper-limb rehabilitation in virtual reality. Appl. Bionics Biom. 6(2), 115–
126 (2009)
5. Veneva, I., Chakarov, D. Tsveov, M., et al.: Active assistive orthotic system: (Exoskeleton)
enhancing movement. In: Handbook of Research on Biomimetics and Biomedical Robotics,
pp. 48–75. IGI Global eEditorial Discovery (2017)
Comparison of a Soft Exosuit and a Rigid
Exoskeleton in an Assistive Task
Domenico Chiaradia1(B) , Michele Xiloyannis2 , Massimiliano Solazzi1 ,

Lorenzo Masia3 , and Antonio Frisoli1
1
PERCRO Lab, Tecip Institute, Sant’Anna School of Advanced Studies, Pisa, Italy
{domenico.chiaradia,antonio.frisoli}@santannapisa.it
2
School of Mechanical and Aerospace Engineering, Nanyang Technological
University, Singapore, Singapore
michele001@ntu.edu.sg
3
Department of Biomechanical Engineering, University of Twente, Enschede,
Netherlands
lormasia@gmail.com
Abstract. Rigid and soft wearable robots have promising complemen-

tary properties that could be exploited to cover a broad range of appli-
cations and needs. While the former are ideal when high forces, accurate
position and high dynamics are required, soft devices are more practical
when portability and comfort are demanded. In this preliminary study,
we quantify this duality by comparing the technical characteristics of a
soft exosuit and a rigid exoskeleton and measuring their biomechanical
and physiological effect on the elbow movements of healthy subjects.
1 Introduction
In the never-ending quest to push the boundaries of human motor abilities and
restore lost motor functions, humans have developed a wealth of wearable robotic
devices. Some of the earliest designs, probably fuelled by the imagination of
science-fiction writers, where visionary but overly-bulky machines that never
accomplished their purpose.
Technological advances have since provided lighter materials, smarter con-
trollers and more compact power supplies: exoskeletons have been used to aug-
ment human strength, endurance during locomotion, provide rehabilitation and
assistance to subjects suffering from neuromuscular diseases and study principles
of human movement [1].
The recent introduction of soft materials in this context gives wearable robots
the potential to penetrate in our daily lives. By using inherently compliant mate-
rials such as fabric and elastomers we can design assistive devices that resemble
our clothes but work in parallel with our muscles to provide assistance [2].
In an elegant comment on the growth and outlook of this field [3], one of its
pioneers rightly underlines that soft wearable robots are complementary and not
substitutive to their rigid counterparts. The weakness of soft devices stands in
https://doi.org/10.1007/978-3-030-01887-0_80
416 D. Chiaradia et al.
their very same strength: the use of soft materials greatly limits the amount of
forces/torques that the device can transmit to the human body and the velocity
that it can move at. This makes them suitable only for applications that require
small levels of assistance, with the wearer having no bone or joint conditions.
Rigid exoskeletons, on the other hand, can deliver higher forces more quickly
and accurately, and their linkage structure allows to do so even in the extreme
case of paralysis of the user.
Fig. 1. Rehab-Exos, the rigid exoskeleton for the upper limbs (a) and the soft exosuit
for assistance of the elbow joint (b). In this study we only provide assistance through
joint 4 (elbow) of the Rehab-Exos.
The objective of this study is to quantify this complementarity by highlight-

ing the strength and weaknesses of rigid and soft wearable robots. To do so, we
compare the technical characteristics of a soft exosuit and a rigid exoskeleton
and measure their biomechanical and physiological effect on the elbow move-
ments of healthy subjects. Both devices provide assistance by compensating for
the gravitational force acting on the arm and shadow the wearer’s motion using
a force/torque sensor.

The Rehab-Exos is an active robotic exoskeleton (Fig. 1a) that exhibits a serial
architecture, isomorphic with the human kinematics, it comprises 3 actuated
Comparison of a Soft Exosuit and a Rigid Exoskeleton 417
Fig. 2. Comparison of the assistive torque and percentage reduction in muscular activ-
ity when wearing the rigid exoskeleton and soft exosuit. (a) The rigid exoskeleton
provided nearly the entire torque required to support the elbow position; the torque
that the wearer needed to exert was higher when wearing the exosuit. (b) Amplified
electromyography from the biceps brachii, comparison between the case of no assistance
and assistance with the exosuit. (c) The bar plot presents the reduction in muscular
activity resulting from wearing the exoskeleton or exosuit.
DOF for the shoulder joint, an active elbow joint and a passive wrist prono-
supination joint [4]. Its joints embed a brushless motor, a torque sensor and a
compact harmonic drive (100:1 reduction).
The exosuit for assistance of the elbow joint (Fig. 1b) consists of three wear-
able components (i.e. a chest, an arm and a foream strap) and an electric motor
(Maxon EC-i 40, 70 W, 55:1 reduction), driving a pair of tendons to assist in
both flexion and extension of the joint. The suit is equipped with a load cell in
series with the flexing tendon and an absolute encoder to monitor the elbow’s
position.
Each device’s control algorithm is designed so that it can move in concert
with its wearer with minimal interaction forces between the two. The Rehab-
Exos control is based on a direct force feedback loop and a full-state observer that
estimates the interaction torques [5]. The soft exosuit implements an admittance-
based control to compensate for the arm’s gravity with an internal velocity
loop [6].
To evaluate the biomechanical and physiological effect on the elbow move-
ments of the two device, 2 healthy subjects were asked to perform repetitive
flexion/extension movements, while holding a 1.25 Kg load in their hand. A ref-
erence trajectory was shown on a screen in order to standardize the range of
motion and speed. The experiments were conducted in three distinct phases,
randomised to avoid potential order effects: with the exoskeleton, with the exo-
suit and without any assistance. We monitored the assistive torque and estimated
the muscular effort from the Root Mean Square (RMS) of the Electromyography
(EMG) of the biceps brachii.
418 D. Chiaradia et al.
3 Results and Discussions

Figure 2a shows the effect on the torque required to flex the elbow resulting
from wearing either the Rehab-Exos or the soft exosuit. Values are reported in
percentage of the total torque required to perform the movement. The Rehab-
Exo provides a slightly higher assistive torque, relieving its werarer from a greater
part of the effort required to flex the joint.
Similar considerations were obtained from analysing the EMG activity of
the biceps brachii by comparing it with the no-assistance case. Figure 2b shows
the muscular activity when flexing the elbow both with and without the soft
exosuit, the difference between them is used to compute the % reduction of
effort from wearing the device. A similar analysis for both systems, averaged
across repetitions and subjects, is shown in Fig. 2c: both produce an average
reduction in the activation of the biceps brachii, of −61.63% and −63.97% for
the rigid and soft device respectively.
Table 1 summarises some of the key aspects that characterise rigid and soft
wearable robots. While the exosuit stands out for its low weight and power
consumption, making it ideal for mobile applications, the rigid device has higher
torque rating, bandwidth and efficiency.
Table 1. Performance of the exoskeleton versus the exosuit
Rigid Soft
Characteristics
Frame material Aluminium Fabric
Motor location Joints Waist
Weight [Kg] 17 1.2
DOF 5 1
Maximum torque [Nm] 150 10
Bandwidth [Hz] 39.7 1.1
Power Consumption [W] 25.65 13.71
Efficiency 0.26 0.14
Physiological effects (% of the no-exo case)
Muscular activity −61.63 −63.97
Human torque −97.05 −76.63
As expected, the Rehab-Exos provides a higher part of the torque required to

move the arm but this did not result in a greater reduction of muscular activity.
A higher sample size, an investigation of the effect on the triceps muscle and
an analysis of the effect of each device on the kinematics of movement would
provide further interesting insights.
Comparison of a Soft Exosuit and a Rigid Exoskeleton 419
References
1. Guizzo, E., Goldstein, H.: The rise of the body bots. IEEE Spectr. 42(10), 42–48
(2005)
2. Asbeck, A.T., De Rossi, S.M., Galiana, I., Ding, Y., Walsh, C.J.: Stronger, smarter,
softer: next-generation wearable robots. IEEE Robot. Autom. Mag. 21(4), 22–33
(2014)
3. Walsh, C.: Human-in-the-loop development of soft wearable robots. Nat. Rev.
Mater. 3, 78 (2018)
4. Vertechy, R., Frisoli, A., Dettori, A., Solazzi, M., Bergamasco, M.: Development of
a new exoskeleton for upper limb rehabilitation. In: IEEE International Conference
on Rehabilitation Robotics, pp. 188–193 (2009)
5. Solazzi, M., Abbrescia, M., Vertechy, R., Loconsole, C., Bevilacqua, V., Frisoli, A.:
An interaction torque control improving human force estimation of the rehab-exos
exoskeleton. In: IEEE Haptics Symposium, HAPTICS, pp. 187–193 (2014)
embedded control of a soft elbow exosuit. In: Proceedings of the IEEE International
Conference on Soft Robotics (RoboSoft), pp. 565–571. IEEE (2018)
Design of Soft Exosuit for Elbow
Assistance Using Butyl Rubber Tubes
and Textile
John Nassour(B) , Sidhdharthkumar Vaghani, and Fred H. Hamker
Artificial Intelligence Lab, Computer Science, Chemnitz University of Technology,

Chemnitz, Germany
john.nassour@informatik.tu-chemnitz.de
Abstract. Soft materials show numerous advantages compared to rigid

ones in exosuit devices. We present the design of a soft wearable elbow
assistance device with flexion and extension actuations. Commercially
available butyl rubber tubes have been used as a pneumatic actuator.
Tubes are enveloped by a lightweight polyester fabric to eliminate a non-
homogeneous expansion. The surrounding fabric in turn is mounted on a
clothes fabric as zigzag paths. The exosuit is lightweight, shock resistant,
simple to manufacture, and low cost. The subjective experiments show a
reduction average of 48% in the Rectified & integrated raw electromyog-
raphy signal of the brachialis muscle during a rhythmic flexion/extension
sequence while lifting weights (3 and 5 [kg]). Results indicate a significant
assistance with respect to the other existing soft elbow exosuits.
1 Introduction
One trend in the wearable robotics research is to develop assistance devices
to support human motion. These devises contribute in human’s daily life, e.g.,
supporting elderlies, rehabilitation, lifting and manipulating heavy objects in
the production lines, and other performance augmentation applications. Rigid
wearable devices show limitations in terms of the wearing duration, the device
weight, and the soft interaction with the human body [1,2]. In order to be more
accepted by human, compact solutions based on lightweight soft materials have
been developed recently. With respect to the actuation technique, the soft wear-
able devices are classed into two main categories: electric [3–6] and pneumatic
[7,8]. As an example for pneumatic soft wearable devices, Suzumori and col-
leagues proposed a “Muscle Textile” exosuit that reduces the integrated EMG
signal of the brachialis muscle by 33% [7]. A recent work used high strength
bubble artificial muscle to assist walking [10], 33% of the required torques were
delivered by a group of three actuators. As an example for electric soft wear-
able devices, the lower body wearable device proposed by Poliero et al. reduced
the mechanical energy requirement for walking up to 30% [9]. In this paper, we
present a pneumatic elbow assistance device, see Fig. 1. The actuator uses butyl
tubes organized in zigzag horizontal patterns for flexion and vertical patterns for
https://doi.org/10.1007/978-3-030-01887-0_81
Design of Soft Exosuit for Elbow Assistance 421
Fig. 1. The control diagram of the soft exosuit. The device is mounted on the elbow
of a human subject while lifting weights. Sensory information for joint bending and
flexion/extension air pressure are transferred into the control unit (Raspberry PiTM )
via an analog to digital converters. A rectified & integrated EMG signal is measured
with MyoWareTM muscle sensor during the experiment. The Raspberry Pi drives a
12[v] air-pump and four solenoids for flexion and extension actuators. A video link
about the paper is available in [12].
extension. Textile is used to envelop the rubber and also to hold the actuators.
Section 2 presents the fabrication process. The actuator was tested with a simple
flexion/extension sequence. Results are presented in Sect. 3. We finally discuss
the contribution in Sect. 4.

Figure 2 shows the external and the internal views of the design. The flexion
actuator is illustrated in red, the extension actuator in blue. The manufacturing
process is detailed in the video link [12]. To distribute the actuator force along the
arm and the forearm we attached a clothe fabric into two parts of commercially
available shin pads protection. A double side polyester belt (5 × 15 [cm]) hosts
two curving sensors (Short Flex/Bend Sensor Adafruit R
). The belt protects
sensors from damage while keeping the freedom for bending. The sensory belt
is mounted to the internal side of the two shin pads to measure the joint angle
of the device. We are using two curving sensors aligned in parallel to each other
and shifted by 1 [cm] each to the left and right sides of the longitudinal axis, see
Fig. 2.
3 Results
The contribution of the actuator in weight lifting tasks is presented through two
experimental parts. First, a human subject lifts weight without the assistance
(passive), then with assistance (active). We performed this experiment several
422 J. Nassour et al.
Fig. 2. Outside and inside views of the soft elbow exosuit (0.5 [kg], 0.4 × 0.18 [m]).
The sensory belt hosts two bending sensors. The horizontal zigzag air tube (in red -
top left) is responsible for the flexion. The vertical zigzag air tube (in blue - top left)
is responsible for extension.
times and with different weight (3 and 5 [kg]). During the experiment, the EMG
raw signal of the brachialis muscle was measured. Figure 3 demonstrates the
experiment set-up (top), and the results. The subjective experiments shows a
reduction of 52% and 38% in the rectified & integrated raw electromyography
signal of the brachialis muscle during a rhythmic extension and 56% and 47%
during rhythmic flexion while lifting 3 and 5 [kg] payloads respectively. Experi-
ment data are available on [11]. Middle graphs show the rectified EMG signals,
the bending sensors’ readings, and the air pressure for two experiments in lifting
3 [kg] payload with actuation (in blue) and without actuation (in red). The bar
graph shows the average of the EMG rectified integral signals over the sets of
four experiments with 3 and 5 [kg] payloads in active and passive movements.
For each exercise cycle (one flexion and one extension), we calculated the inte-
gral of rectified EMG signal based on the reading of banding sensors for the
extension and flexion movements. Since the extension and the flexion durations
may differ from one cycle to another, we use the mean value of the integral in
the comparison. The averages of that mean value over sets of the experiments
and its confidence intervals are given in the bar graph.
Design of Soft Exosuit for Elbow Assistance 423
Fig. 3. EMG measurement for active and passive movements. Experiment set-up and
the EMG electrodes locations (top). Middle graphs illustrate the rectified EMG, the
bending sensors average voltage, and the air pressure for two experiments (passive and
active) in lifting 3 [kg] payload. The bar graph shows average values of the mean of the
integrated EMG signals for flexion and extension while lifting 3 and 5 [kg] payloads
(bottom).
4 Conclusions
We present a soft exosuit for elbow assistance with 1 D.O.F. (flexion/extension).
It is made using butyl rubber tubes and textile material. The device produces
a range of motion of 95◦ and support the elbow joint to lift different weights
(48% EMG signal reduction). Results show that the assistance provided by the
exosuit during the rhythmic extension/flexion movements of the elbow joint is
high compared to the existing elbow assistance devices.
424 J. Nassour et al.
References
1. Wu, K.-Y., Su, Y.-Y., Yu, Y.-L., Lin, K.-Y., Lan, C.-C.: Series elastic actuation of
an elbow rehabilitation exoskeleton with axis misalignment adaptation. In: Inter-
national Conference on Rehabilitation Robotics (2017)
2. Manna, S.K., Dubey, V.N.: A mechanism for elbow exoskeleton for customised
training. In: International Conference on Rehabilitation Robotics (2017)
embedded control of a soft elbow exosuit. In: IEEE International Conference on
Soft Robotics (2018)
4. Guo, J., et al.: A soft robotic exo-sheath using fabric EMG sensing for hand rehabil-
itation and assistance. In: IEEE International Conference on Soft Robotics (2018)
5. Awad, L.N., et al.: Soft exosuits increase walking speed and distance after stroke.
In: International Symposium on Wearable Robotics (2017)
6. Bae, J., et al.: Exosuit-induced improvements in walking after stroke: comprehen-
sive analysis on gait energetics and biomechanics. In: International Symposium on
Wearable Robotics (2017)
7. Abe, T., Koizummi, S., Nabae, H., Endo, G., Suzumori, K.: Muscle textile to
implement soft suit to shift balancing posture of the body. In: IEEE International
Conference on Soft Robotics (2018)
8. Koh, T.H., Cheng, N., Yap, H.K., Yeow, C.H.: Design of a soft robotic elbow sleeve
with passive and intent-controlled actuation. Front. Neurosci. 11, 597 (2017)
9. Poliero, T., et al.: Soft wearable device for lower limb assistance: assessment of an
optimized energy efficient actuation prototype. In: IEEE International Conference
on Soft Robotics, pp. 559–564 (2018)
10. Diteesawat, R.S., Helps, T., Taghavi, M., Rossiter, J.: High strength bubble arti-
ficial muscles for walking assistance. In: IEEE International Conference on Soft
Robotics, pp. 388–393 (2018)
11. The paper experiments data. https://www.tu-chemnitz.de/informatik/KI/edu/
robotik/videos/werob2018data.zip
12. The paper descriptive video. https://youtu.be/-BOO6PC-9Pc
Optimizing Body Thickness of Watchband-
Type Soft Pneumatic Actuator for Feedback
of Prosthesis Grasping Force
Masashi Sekine(&), Kazuya Kawamura, and Wenwei Yu
Center for Frontier Medical Engineering, Chiba University, Chiba, Japan

sekine@office.chiba-u.jp, kawamura@chiba-u.jp,
yuwill@faculty.chiba-u.jp
Abstract. Watchband-type soft pneumatic actuators were prototyped and tested

by motion experiments. The actuators are soft device which is made from
thermoplastic copolyester (TPC) by 3D printer. The actuator is for a device for
feedback of prosthesis grasping force. The feedback device resembles a watch,
and the parts corresponding to the watch wristband are actuators shaped like a
thumb and index finger. The actuator of finger-type bands can press an intact
wrist of amputees who use a prosthesis by wrapping around the wrist in
accordance with the grasping motion of the prosthesis hand. In experiments,
actuator’s various configurations using different combination of thickness of the
body were compared by measuring output forces and range of motion (ROM).
As a result of the comparison, the optimal configuration which improve the
force and ROM was found.
1 Introduction
Hand grasping motions are often done, thus very important in activities of daily living.
The grasping force of prosthesis should be conveyed for the users of prostheses, i.e., a
feedback system should be embedded to prosthetic hand. There is a risk that users crush
an object to be held or inflicting injury on others by using excessive grasping force.
In our previous study [1], wristband-type pneumatic device for feedback of pros-
thesis grasping power was prototyped. This is soft device which is mainly made from
thermoplastic copolyester (TPC). The feedback device resembles a watch, and the parts
corresponding to the watch wristband are actuators shaped like a thumb and index
finger. The actuator of the finger-type bands can press an intact wrist of amputees who
use a prosthesis by wrapping around the wrist in accordance with the grasping motion
of prosthesis hand. However, the actuator’s pressing force was insufficient comparing
the grasping power of hand prostheses or healthy hands. Moreover, the actuator is
considered to have room for improvement in bending angle for grasping, that is, the
range of motion (ROM). In this study, the dimension of internal structure of the
This work was supported by the JSPS Grant-in-Aid for Scientific Research (C), Grant Number
16K01537.

https://doi.org/10.1007/978-3-030-01887-0_82
426 M. Sekine et al.
actuator in previous study [1] was set as a benchmark, and the actuators with several
patterns of structure were tested by activating with air pressure and compared. Based on
a consideration of the output force and ROM of the actuator, the optimal dimension
was confirmed.
2 Design and Basic Structure of Actuator
The benchmark model of watchband-type pneumatic soft actuator is shown in Fig. 1.

The actuator of the finger-type bands is basically hollow, and the top part is shaped like
a bellow. The actuators of index finger and thumb type is swollen, bend inward like
finger grasping, and finally become like wristband by feeding air to the inside of the
actuators as shown in Fig. 1(a)–(c). The inside structure is shown in Fig. 1(d). Basi-
cally, the thickness of the top and bottom part of the actuator is 2.0 mm and 3.0 mm,
respectively. The bellow of the top part forms some chambers inside the actuator as
shown in Fig. 1(d). Such elastomer pneumatic device can contribute to enhance safety
of wearable devices with intrinsic softness from a viscoelasticity of air and elastic
materials. Soft pneumatic actuators (SPAs) with similar structure has been studies [2–
6], however, these devices were not basically used for grasping wrist as in our study.
Moreover, the grasping force below about 1.4 N of these SPAs [3–5] was considered to
be insufficient for good feedback for the grasping.
Fig. 1. Benchmark model of watchband-type soft pneumatic actuator prototype: (a) initial state
(0 s, 0 MPa); (b), (c) states after activation (0.6 MPa); (d) inside (cut surface) of index finger-like
actuator.
Optimizing Body Thickness 427
3 Method of Comparison and Experiments
In this study, configurations of the actuator were compared by using only the index
finger. The configurations are shown in Table 1. The thickness of the top and bottom
part of the actuator was set as parameters. For all configurations, the maximum bending
angle of fingertip and the pressing force at proximal phalanx (PP) part of the index
finger were measured, and a difference of the force and ROM of the actuator was
confirmed. About pressing force, the force in the direction of the red arrow in Fig. 1(d)
was measured by using a force gauge (ZTA-DPU-500 N, IMADA Corp.).
Table 1. Configurations for actuator

Configuration Thickness
(mm)
Top Bottom
I2O3 (Benchmark) 2 3
I2O4 2 4
I2O2 2 2
I1O3 1 3
The actuators made from TPC were prototyped by 3D printer (MF-2200D, Mutoh
Industries, ltd.). In motion experiments, maximum bending angle h of the tip of index
finger with air pressure 0.6 MPa against initial state (0 MPa) was shown in Figs. 2 and 3.
Moreover, Fig. 4 showed the pressing force measured at PP part of the index finger,
which generated the force for the wrist grasping with air pressure 0.6 MPa. As the
results, both ROM and pressing force of the configuration I1O3 was improved com-
paring the benchmark I2O3 as shown in Figs. 2, 3 and 4. The chambers of the I1O3 could
expand larger due to thinner wall of the top part. Therefore, the I1O3 could bend larger
Fig. 2. Motion experiments: (a) benchmark I2O3 at initial state (0 MPa); (b) I2O3 pressurized
to 0.6 MPa. h represents bending angle of fingertip; (c) I2O4 with 0.6 MPa; (d) I2O2 with
0.6 MPa; (e) I1O3 with 0.6 MPa.
428 M. Sekine et al.
than other configurations, and then it is considered to generate larger pressing force. At
the beginning, I2O4 was set in expectation of strong pressing force due to harder bottom
part of thickness 4 mm. However, it was considered that the harder wall of bottom part
caused small bending angle, and then I2O4 could not generate large force.
Fig. 3. Bending angle of fingertip for all configurations
Fig. 4. Pressing force at PP part for all configurations
5 Conclusion
In this study, watchband-type soft pneumatic actuators with various thicknesses of the
body were prototyped and tested for comparison by motion experiments. As a result of
the comparison, the optimal configuration which improve the force and bending angle,
i.e., ROM was found. However the pressing forces were considered to be still not
enough considering grasping power of human and hand prosthesis. In the future, we
will improve the structure including the size, pitch of chambers inside the body and
investigate further body thickness including practical limit of thinness in order to
accomplish stronger grasping power with lower air pressure. Moreover, we must
reconsider the setting of fabrication conditions of 3D printing because slight air leakage
from actuator body occurred.
Optimizing Body Thickness 429
References
1. Sekine, M., Shiota, K., Liu, E., Kawamura, K., Yu, W.: Prototype wristband pneumatic
device made from thermoplastic copolyester for feedback of prosthesis grasping force. In:
Proceedings and Abstracts Book of Advanced Materials World Congress (2018)
2. Ogura, K., Wakimoto, S., Suzumori, K., Nishioka, Y.: Micro pneumatic curling actuator -
nematode actuator. In: Proceedings of the 2008 IEEE International Conference on Robotics
and Biomimetics (ROBIO), pp. 462–467 (2008)
3. Wang, Z., Hirai, S.: Soft gripper dynamics using a line-segment model with an optimization-
based parameter identification method. IEEE Robot. Autom. Lett. 2, 624–631 (2017)
4. Polygerinos, P., Lyne, S., Wang, Z., Nicolini, L.F., Mosadegh, B., Whitesides, G.M., Walsh,
C.J.: Towards a soft pneumatic glove for hand rehabilitation. In: Proceedings of the 2013
IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 1512–
1517 (2013)
5. Shintake, J., Sonar, H., Piskarev, E., Paik, J., Floreano, D.: Soft pneumatic gelatin actuator for
edible robotics. In: Proceedings of the 2017 IEEE/RSJ International Conference on Intelligent
Robots and Systems (IROS), pp. 6221–6226 (2017)
6. Wang, T., Ge, L., Gu, G.: Programmable design of soft pneu-net actuators with oblique
chambers can generate coupled bending and twisting motions. Sens. Actuators A Phys. 271,
131–138 (2018)
The Effect of Negative Damping
at the Hip Joint During Level Walking: A
Preliminary Testing
Jongwon Lee(B) , Juwhan Bae, Chilyong Kwon, Seokjin Hwang,

and Gyoosuk Kim
Rehabilitation Engineering Research Institute of the Korea Workers’ Compensation

and Welfare Service, Incheon, Republic of Korea
jongwonia@gmail.com
Abstract. Walking is a fundamental but important activity for inde-

pendent daily life. We designed an intuitive controller (called negative
damping controller) for the elderly to walk more efficiently. The con-
troller behaves as a negative damping for the hip joint to eliminate nat-
ural energy dissipation terms and is capable of mimicking the biological
torque profile. The primary objective of this study is to evaluate the
effects of the controller on the biomechanical and physiological aspects
of gait. As a first step, we measured gait parameters and muscle acti-
vation levels for a subject during overground walking with and without
exoskeleton at self-selected speed. The experimental results show that
the negative damping controller has the potential to enable the wearer
to walk overground with larger hip range of motion and lower muscle
activation: the hip range of motion increased about 21.9% and the aver-
age muscle activation levels decreased about 18.0% in rectus femoris and
about 23.7% in bicep femoris.
1 Introduction
TODAY most countries are facing the challenges of population ageing and low
birth rate. Because the issues give rise to decrease in not only prime-age workers
but also the caregiver population, it has been required to develop technologies
for independent daily life of the aged. To meet the social needs, many research
groups are developing a variety of types of wearable devices, generally called
exoskeleton, to assist human locomotion [1].
However the questions about the actuation joint, timing, and its magnitude
for assistance of human locomotion still remain. Several research groups have
proposed the answer to the questions. In [2] it is revealed that providing external
This work was supported by Institute for Information & communications Technology
Promotion(IITP) grant funded by the Korea government(MSIP) (No. 2016-0-00452,
Development of creative technology based on complex 3D printing technology for
labor, the elderly and the disabled).
https://doi.org/10.1007/978-3-030-01887-0_83
The Effect of Negative Damping at the Hip Joint During Level Walking 431
power to the hip joint is more effective to reduce metabolic cost than providing
the same amount of power at the ankle joint. In [3] researchers reported that
actuating force profile mimicking the biological joint torque allows a reduction
in metabolic cost.
Based on their findings, we proposed a simple and intuitive controller to
assist hip flexion/extension movements during overground walking. The con-
troller behaves as a negative damping for the hip joint to eliminate natural
energy dissipation terms and is capable of mimicking the biological torque pro-
file. The present study evaluates the effect of the negative damping at the hip
joint on the wearer’s gait kinematics and muscle activation during level walking.
Fig. 1. Negative damping controller for a hip exoskeleton

2.1 Negative Damping Controller
Negative damping behavior can be realized by feeding the hip rotational velocity
back with a simple form:
τN D = b · θ̇hip . (1)
The virtual negative damping coefficient b is a tuning factor according to a
wearer’s preference. In addition, as shown in Fig. 1, the controller includes com-
pensation efforts to eliminate gravitational effects by the wearer’s and exoskele-
ton’s weight and nonlinear frictional effects in the mechanical transmission sys-
tems, because their effects can be considered as a critical disturbance to resist
wearer’s hip movements for the elderly people.
432 J. Lee et al.
2.2 Hip Exoskeleton
The negative damping controller is implemented in the hip exoskeleton developed

by Rehabilitation Engineering Research Institute (Incheon, Republic of Korea),
as shown in Fig. 2. The exoskeleton generates assistance torques by a pair of
BLDC motors with harmonic drive transmission mounted at the hip joints. The
generated torques are transmitted to wearer by 3D printed harnesses attached
to the waist and thigh. A battery and control unit are placed on the back. The
total weight of the exoskeleton is about 6.5 kg.
Fig. 2. The hip exoskeleton used for the experiments (left). Snapshots of level walking
experiment under exoskeleton condition (right).
2.3 Methods
To evaluate the negative damping effect at the hip joint, we conducted a level
walking experiment under two device conditions (control and exoskeleton). As
this is preliminary testing, one healthy male subject free from musculoskeletal
and neurological disorders was participated in this experiment (age 35, height
175 cm, weight 81 kg). Under the exoskeleton condition, the exoskeleton was
worn by the subject, but was not worn under the control condition. Between
the experimental conditions, an interval of 1 h was allowed for the subject to
recover from residual effects. Before the experiment under the exoskeleton con-
dition, the subject was given enough familiarization time with the device. During
familiarization, we determined negative damping coefficient (b in (1)) enabling
the subject to walk with preferred speed. During the experiments, a 3D motion
analysis system (Cortex 6.02, Motion Analysis, USA) was used to measure gait
kinematic data and a surface electromyogram (TeleMyo 2400R, Noraxon, USA)
was used to measure activation levels of the rectus femoris (RF) and bicep femoris
(BF) muscles.
The Effect of Negative Damping at the Hip Joint During Level Walking 433
3 Results
Figure 3 shows that the actuation torque for hip flexion is higher than that for
hip extension, as the negative damping effect depends on hip rotational speed.
The hip angle trajectory, muscle activation levels in the RF and BF and walk-
ing velocity are illustrated in Fig. 4. Compared to the control condition, some
significant differences were observed under the exoskeleton condition. First, the
hip range of motion (ROM) was increased about 21.9%. Second, the average mus-
cle activation levels during level walking were decreased about 18.0% in RF and
about 23.7% in BF. The maximum muscle activation levels, which is given as a
percentage of maximum voluntary contraction (MVC), were also decreased from
19% to 15% in RF and from 13% to 8% in BF. Finally, the hip flexion/extension
profile was altered and the preferred gait speed was decreased from 5.17 km/h
to 4.67 km/h.
Fig. 3. The hip torque profile computed by the negative damping controller (Sect. 2.1).
Fig. 4. Hip angle trajectory, muscle activation levels, and walking velocity: control
condition (left), exoskeleton condition (right).
434 J. Lee et al.
4 Conclusion
Our findings from the preliminary testing suggest that the negative damping at
the hip joint has the potential to enable the wearer to walk overground with
larger hip ROM and lower muscle activation level. We also found that the neg-
ative damping changed the hip flexion/extension profile and decreased the pre-
ferred walking velocity. In [4,5] it has been reported that altered gait pattern
is one of the critical factors in increasing the net metabolic cost. Therefore fur-
ther investigation is necessary to clarify the issue of the metabolic effectiveness.
Future work will also include development of a systematic way to find the opti-
mal negative damping parameter and to evaluate the negative damping effects
on the biomechanical and physiological aspects of gait among different subjects.
References
1. Yan, T., Cempini, M., Oddo, C.M., Vitiello, N.: Review of assistive strategies in
powered lower-limb orthosis and exoskeletons. Robot. Auton. Syst. 64(3), 120–136
(2015)
2. Sawicki, G.S., Lewis, C.L., Ferris, D.P.: It pays to have a spring in your step. Exer.
Sports Sci. Rev. 37(3), 130–138 (2009)
3. Ding, Y., Caliana, I., Siviy, C., Panizzolo, F.A., Walsh, C.: IMU-based iterative
control for hip extension assistance with a soft exosuit. In: IEEE International Con-
ference on Robotics and Automation (ICRA), pp. 3501–3508 (2016)
4. Zarrugh, M.Y., Radcliffe, C.W.: Predicting metabolic cost of level walking. Eur. J.
Appl. Physiol. Occup. Physiol. 38(3), 215–223 (1978)
5. Donelan, J.M., Kram, R., Kuo, A.D.: Mechanical and metabolic determinants of
the preferred step width in human walking. Proc. R. Soc. Lond. Ser. B Biol. Sci.
268(1480), 1985–1992 (2001)
Overview and Challenges for Controlling
Back-Support Exoskeletons
Maria Lazzaroni1,2(B) , Stefano Toxiri1 , Darwin G. Caldwell1 , Elena De Momi2 ,

and Jesús Ortiz1
1
Department of Advanced Robotics, Istituto Italiano di Tecnologia, Genova, Italy
maria.lazzaroni@iit.it
2
Department of Electronics, Information and Bioengineering, Politecnico di
Milano, Milan, Italy
Abstract. Exoskeletons were recently proposed to reduce the risk of mus-

culoskeletal disorders for workers. To promote adoption of active exoskele-
tons in the workplace, control interfaces and strategies have to be designed
that overcome practical problems. Open challenges regard sensors inva-
siveness and complexity, accurate user’s motion detection, and adaptabil-
ity in adjusting the assistance to address different tasks and users. Focus-
ing on back-support exoskeletons, different control interfaces and strate-
gies are discussed that aim at automatically driving and modulating the
assistance, according to the activity the user is performing.
1 Introduction
In order to reduce workers’ probability of developing musculoskeletal disorders
(MSDs) [1], wearable technologies like exoskeletons have recently attracted con-
siderable interest among the academic community and in industries [2]. As the
lumbar spine is one of the body area most affected [3], back-support exoskeletons
to assist lifting task are being developed. The aim is to reduce back overloading,
by reducing the activity of spinal muscles [4].
In contrast to passive exoskeletons, active ones can modulate the assistance
during the operation and thereby adapt to different tasks and users. The key to
adaptability is a suitable control strategy that is able to precisely detect user’s
movement intention and provide assistance with appropriate timing and amount.
Different sensors to acquire convenient input signals and the complex processing
and integration of these signals are required. Moreover, minimal sensors inva-
siveness is necessary for exoskeleton use in real work environments.
2 Methods
In order to assist the user only when necessary and to not impose undesirable
movement constraints, control decisions should be considered at the different
This work is funded by the Italian Workers’ Compensation Authority (INAIL). The
authors from Politecnico di Milano were involved through Istituto Italiano di Tec-
nologia and did not receive any direct funding.
https://doi.org/10.1007/978-3-030-01887-0_84
436 M. Lazzaroni et al.
control levels, as proposed in [5]. The high level has to classify the task the
user is performing, by the processing and the integrating of a set of proper
input signals. The control strategy (i.e. the mid level) modulates the assistance
accordingly to the current task and generates reference values of the desired
exoskeleton state outputs, such as torque or speed. The low level tracks these
values, regulating motors outputs. In this approach, the low and the mid levels
are independent of the task detection control level. Consequently, the control
strategy most convenient for each specific task can be employed.
Several strategies to modulate exoskeletons assistance have been proposed
in the literature. They use different sensors arrangements that acquire different
input signals to detect user’s intended movement. The assistance is manually
triggered by the user with extra joysticks or buttons when a system is not able
to automatically detect user’s intention (e.g., back-support Muscle Suit [6]).
Electromyography (EMG)-based strategies control exoskeletons according to the
wearer’s muscle activity. Surface EMG of related muscles is usually used, as for
the HAL Lumbar Support [7] that is controlled by the EMG of erector spinae
muscles. However, the assistance can be controlled even by the EMG of a muscle
acting on a different joint, if it is activated in coordination with the target muscle
during the task [8]. Control driven by mechanically intrinsic signals relies on
measures that are intrinsic to the device itself. User’s motion is registered thanks
to Inertial Measurement Units (IMUs) and encoders and is used to modulate the
exoskeleton assistance by compensating user’s upper body weight against gravity
(e.g., CRAY X [9]). To control a knee exoskeleton [10], IMUs and encoders were
used in combination with foot sensors that measure Ground Reaction Forces
(GRFs).
As regards back-support exoskeleton, two recent studies have examined the
problem of detecting user’s lift needs for assistance and then driving the assis-
tance accordingly.
For Robo-Mate exoskeleton [11], two different control strategies are proposed
that consider the factors related to lumbar compression: the torso inclination
and the weight of the lifted object. The first strategy provides the assistance
proportionally to the torso inclination, measured by an IMU mounted on the
exoskeleton. The second strategy provides the assistance proportionally to the
EMG of the forearm muscles, as during grasping and holding forearm muscles
activity increases with object weight.
In [12] an algorithm is proposed that detects lift movement using encoders
and an IMU embedded in the Active Pelvis Orthosis (APO). Encoders measure
the left and the right hip joint angles used to detect the transition between the
different phases of the lifting task. If a lift is detected, the estimation of the thigh
angle (provided thanks to the additional IMU) is used to confirm the current lift
phase. Knowing the current user’s movement, the assistive torque is computed
using only hip angle measures from the encoders.
In both studies, back muscles EMG activity was analysed to evaluate the
effectiveness of the exoskeleton in reducing muscular activation during lifting.
Both studies showed a significant reduction (around 30%) of muscle activity.
Overview and Challenges for Controlling Back-Support Exoskeletons 437
3 Discussion
To promote exoskeletons use in industries, the invasiveness of the sensors and
the ease of use of the device have to be achieved while ensuring adaptability in
order to address different users, tasks and assistance requirements.
As regards sensors, the aim is to minimise instrumentation complexity while
maximising the information we can extract from them to recognise movements
and tasks. Manual trigger main limitation is that users are required to use
their hands to control the system. This increases cognitive burden, makes the
task intermittent, and additionally introduces physical complications since user’s
hands are usually busy to lift objects. Considering industrial workplace, EMG
signal variability (with time, fatigue, sweat, skin artefacts) and the invasiveness
of the electrodes limit EMG-based strategies use in this context. By contrast,
mechanically intrinsic controllers employ IMUs and encoders that are easy to
integrate into an exoskeleton. Invasiveness problem would emerge with GRF
sensors, that can estimate the presence of an external weight, but cannot be
integrated into the structure and, furthermore, may limit wearer’s movement.
Nevertheless, mechanically intrinsic control main limitation is that they usually
required an accurate model of the body.
As concerns controller design, the Robo-Mate exoskeleton [11] implements
only the mid and the low levels. The assistance is thus given when a particu-
lar movement, and not a complete task, is recognised. This approach permits,
therefore, to assist tasks not standardised (e.g. asymmetric lifting) that require
some type of help as the user is bending his torso or holding an object. Indeed,
the gravity compensation assistance is given to the user both in the lowering
and in the lifting phase. Moreover, additional assistance is given proportionally
to the weight of the lifted object, estimated by forearm muscles activity. How-
ever, unwanted forces or movement constraints are possible, as the system is not
able to detect and switch off when different activities are being performed (e.g.,
walking, taking stairs, sitting for which this type of assistance is not meaningful).
By contrast, the approach introduced for the APO device [12] implements a
lift detection algorithm as a high level controller to trigger the assistance auto-
matically. The advantage of this approach is that it is possible to assist the
user specifically for the target task, avoiding constraints or undesirable assistive
forces corresponding to different activities. Embedded and minimally invasive
sensors ensure an easy implementation in real practical applications, but they
do not provide information about external objects weight. Nevertheless, con-
troller effectiveness relies on the capability of discerning accurately the lifting
task. Accuracy was proved to be higher than 97%, also for different lifting tech-
niques and speeds. However, several assumptions have been made to strictly
define the lifting task: grasping must happen before lifting and has a predeter-
mined time threshold, the hip angle has to reach the peak in the grasping phase,
the lifting is symmetric. Therefore, only standardised tasks can be assisted effec-
tively. Moreover, the high level is designed to detect the tasks at the beginning of
the lift movement, thereby the user is not assisted during the lowering phase to
support his own or a potential external weight. In this contest, a future challenge
438 M. Lazzaroni et al.
could be to design an algorithm able to classify many different tasks that require
assistance and then implement the different control strategies accordingly.
To make control strategies effective, the major causes of workers’ MSDs have
to be investigated. As concerns lifting task assistance, in [11] two key factors
have been found that mostly affect lumbar compression: torso inclination and
the mass of the object being handled.
4 Conclusion
Recent progress in research has been contributing to promoting adoption of
back-support exoskeletons in real working scenarios. The challenges that active
devices have to address were discussed, regarding integration of the acquisition
systems in the structure, strategies to modulate the assistance during the oper-
ation and control system design. In our opinion, the underlined advantages of
a task detection level should be further exploited, together with more advanced
strategy for targeted assistance. Future works will focus on the classification of
different tasks and the delineation of specific control strategies for each of them,
as authors believe that could promote exoskeletons employment significantly.
References
1. Punnett, L., Wegman, D.H.: Work-related musculoskeletal disorders: the epidemi-
ologic evidence and the debate. J. Electromyogr. Kinesiol. 14(1), 13–23 (2004)
Ergonomics 59(5), 671–681 (2016)
3. INAIL Open Data. https://dati.inail.it
4. Reeves, N.P., Cholewicki, J.: Modeling the human lumbar spine for assessing spinal
loads, stability, and risk of injury. Crit. Rev. Biomed. Eng. 31(1&2), 73–139 (2003)
5. Tucker, M.R., Olivier, J., Pagel, A., Bleuler, H., Bouri, M., Lambercy, O., del R
Millán, J., Riener, R., Vallery, H.: Control strategies for active lower extremity
prosthetics and orthotics: a review. J. Neuroeng. Rehabil. 12(1), 1 (2015)
6. Kobayashi, H. Nozaki, H.: Development of support system for forward tilting of
the upper body. In: 2008 IEEE International Conference on Mechatronics and
Automation, ICMA 2008. pp. 352–356. IEEE (2008)
7. Hara, H., Sankai, Y.: Development of HAL for lumbar support. In: SCIS &
ISIS/SCIS & ISIS 2010, pp. 416–421. Japan Society for Fuzzy Theory and Intelli-
gent Informatics (2010)
8. Grazi, L., Crea, S., Parri, A., Yan, T., Cortese, M., Giovacchini, F., Cempini,
M., Pasquini, G., Micera, S., Vitiello, N.: Gastrocnemius myoelectric control of
a robotic hip exoskeleton. In: 2015 37th Annual International Conference of the
IEEE Engineering in Medicine and Biology Society (EMBC), pp. 3881–3884. IEEE
(2015)
9. German Bionic CRAY X. https://www.germanbionic.com/product/
10. Saccares, L., Brygo, A., Sarakoglou, I., Tsagarakis, N.G.: A novel human effort esti-
mation method for knee assistive exoskeletons. In: 2017 International Conference
on Rehabilitation Robotics (ICORR), pp. 1266–1272. IEEE (2017)
Overview and Challenges for Controlling Back-Support Exoskeletons 439
O’Sullivan, L., Caldwell, D.G.: Rationale, implementation and evaluation of assis-
tive strategies for an active back-support exoskeleton. Frontiers in Robotics and
AI (in press)
12. Chen, B., Grazi, L., Lanotte, F., Vitiello, N., Crea, S.: A real-time lift detection
strategy for a hip exoskeleton. Front. Neurorobotics 12, 17 (2018)
Assessment of a Hand Exoskeleton
Control Strategy Based on User’s
Intentions Classification Starting
from Surface EMG Signals
Nicola Secciani1(B) , Matteo Bianchi1 , Alessandro Ridolfi1 , Federica Vannetti2 ,

and Benedetto Allotta1,2
1
Department of Industrial Engineering (DIEF), University of Florence,
Florence, Italy
nicola.secciani@unifi.it
2
Don Carlo Gnocchi Foundation, Florence, Italy
Abstract. Among the most challenging aspects of the current bio-

robotics trends, a place of honor is certainly reserved to the assistance to
physically impaired people during the Activities of Daily Living (ADLs).
The aim of this work is to assess the interaction between an assistive hand
exoskeleton, controlled through surface ElectroMyoGraphy (sEMG) sig-
nals, and its user. A new control strategy, which focuses mainly on
the wearability and the usability of the system, is presented. Promis-
ing results of two preliminary tests, conducted in collaboration with the
“Don Carlo Gnocchi” Foundation, are also reported.
1 Introduction
Nowadays, robots are permeating more and more aspects of everyday life and,
as a consequence, the matter of a safe and reliable human-robot interaction is
becoming a central topic in the robotic field [1]. Particular attention has to be
paid for assistive devices, which are specifically thought to physically interact
with human beings for long periods of time. Making the interaction of the human
hand with assistive robotic systems comfortable is certainly a difficult task since
the hand is an important provider of independence during Activities of Daily Liv-
ing (ADLs), which usually require high dexterity to be performed. The presented
work focuses on the assessment of the interaction between a hand-impaired per-
son and an assistive hand exoskeleton, presenting a light control strategy which
mainly aims to exploit surface ElectroMyoGraphy (sEMG) signals to improve
the simplicity, the wearability and the usability of the system.
This work has been supported by the HOLD Project, funded by the University of
Florence. The authors would also like to thank the Don Carlo Gnocchi Foundation
for the help.
https://doi.org/10.1007/978-3-030-01887-0_85
Assessment of a Hand Exoskeleton Control Strategy 441
Fig. 1. The exoskeleton developed by the researchers of the MDM Lab, Florence, worn
by a patient.
2 The Controlled Device

The exoskeleton under examination, see Fig. 1, has been designed and devel-
oped by the researchers of the Mechatronics and Dynamic Modeling Labora-
tory (MDM Lab) of the Department of Industrial Engineering (DIEF) of the
University of Florence (UNIFI) to people with disabilities in hand opening [2].
This device acts on the four long fingers by means of four planar mechanisms
and has a total of four actuated (flexion/extension) and four passive (abduc-
tion/adduction) Degrees Of Freedom (DOFs), two per finger. The kinematics
of each mechanism is customized on the user’s hand thanks to an automatic
scaling algorithm [3], which, starting from anatomical measurements, calculates
the dimensions of each mechanism in order to reproduce the natural trajectory
of the fingers with the minimum possible error. A single servomotor is in charge
of opening all the four long fingers at the same time by pulling a cable which has
two connection points on each mechanism. Closing gesture is passively allowed
releasing the same cable. Transmission system and finger mechanisms are placed
on the hand backside, not to influence objects handling. A taylor-made ther-
moformed splint is used as the interface between the device and the hand and
represents a stable kinematic coupling. Because of the stringent constraints that
this scenario imposes in terms of encumbrance and lightness, the electronics of
the system is reduced to the minimum necessary: a single magnetic encoder, two
EMG sensors and an Arduino Nano board are the only exploited components in
charge of triggering and controlling the exoskeleton.
3 Control Strategy and Preliminary Tests

The control strategy mainly focuses on managing trigger actions, by identifying
user’s intentions and translating them into appropriate control commands for the
actuation system. The outer control loop continuously checks that the system
does not overcome a fixed range of motion, while the inner one is in charge of
checking if an object is grasped while closing. Both loops rely on the information
collected by the encoder which is mounted on the exoskeleton in correspondence
442 N. Secciani et al.
of the metacarpophalangeal joint. Human hands are capable of performing lots

of different movements and the muscles that manage hand motion are many
and very close to each other. These are two of the main reasons that make the
classification of user’s intentions starting from sEMG signals a very complex
task. A common solution presented in literature [4,5], is the use of worksta-
tions which run heavy algorithms to precisely discriminate lots of possible hand
movements. The proposed strategy for EMG classification is, instead, thought to
be implemented on an embedded micro-controller board, which can be directly
mounted on the system. This point is particularly important since the aim of
the device is to give assistance, beside just rehabilitation, and therefore it has
to work as standalone, allowing the user to move freely not being constrained
in confined spaces close to a control station. Moreover, the decision to embed
everything in a single device, which differentiates this solution from others [5],
has been taken in order to pave the way to the development of a dock station to
allow the impaired user to autonomously wear the system, drastically enhancing
his own independence. Only hand opening, hand closing and hand resting have
then been considered as possible user’s intentions to be classified as a result of a
trade-off between the number of different movements that can be classified and
the available computational power. A custom graphical user interface has been
designed to provide a user-friendly tool to be used to straightforwardly tune the
Point-in-Polygon algorithm in charge of the classification. This algorithm is a
ray-casting to the right: it takes as inputs the number of the polygon vertices,
their coordinates and the coordinates of a test point; each iteration of the loop,
the line drawn rightwards from the test point is checked against one of the poly-
gon edges and the number of time this line crosses the edge is counted; once the
loop has ended, if the number of crosses is an odd number of times, then the
point is outside, if an even number, the point is inside. The collected EMG data
referring to different hand gestures are processed, extracting the envelope of the
RAW signal, and are displayed scattered on an interactive 2D Cartesian plane
whose axis report the signals from the two EMG sensors, which are respectively
placed on the forearm muscles bands responsible of fingers/wrist extension and
flexion. Manually drawing polygons, delimiting the clouds of points belonging to
the same gesture, it is possible to define different regions on the 2D plane which
can be associated to a precise movement, and which the classifier can act on.
One subject (male, aged 54, 1+ Ashworth Scale) has been enrolled to perform
two tests. This patient suffers since birth from Spinal Muscular Atrophy (SMA)
type II, which produced a selective damage to muscular extensors of both hands
causing a clenched fist deformity. The subject has been asked to grasp 10 differ-
ent objects of different sizes and shapes from those of daily use and to place them
on a standard shoebox. Five trials for each object have been conducted and the
average grasping time has been calculated. Test sessions have been accomplished
twice in a week, on Monday and on Friday.
Assessment of a Hand Exoskeleton Control Strategy 443
4 Results
Results, see Fig. 2, show that, as it was easy to expect, grasping time is, in gen-
eral, longer when the shape of the object gets more complicated. Lateral grasping
resulted to be more difficult than the vertical one, because both the weight of
the exoskeleton and of the object differently influenced the muscles activation.
Finally, even if the average times appear quite high compared to common stan-
dards for able-bodies, they have become remarkably lower (an average of about
30% less) in only one week of training.
Fig. 2. Results of the conducted trials.
5 Conclusion and Further Developments
The presented research work aimed to assess the usability of an assistive hand
exoskeleton controlled with a sEMG-based control strategy. The main goal was to
reach a satisfying solution in terms of wearability and ease of use. The proposed
system and control strategy, tested on a real patient, have highlighted promising
results and, in particular, that the use of a low-complexity system allows for
a steep learning curve (confirmed by a rapid reduction of the grasping times).
444 N. Secciani et al.
The authors are aware that many future improvements can be done and are
currently working on enhancing the performance of the classification phase (e.g.
increasing the number of different gestures). A detailed study of the classifier
accuracy has been planned and, in this perspective, the integration of inertial
measurements have been considered as a possible source of improvements.
References
1. Pervez, A.: Safe physical human robot interaction-past, present and future. J. Mech.
Sci. Technol. 22(3), 469–483 (2008)
2. Conti, R.: Kinematic synthesis and testing of a new portable hand exoskeleton.
Meccanica 52, 2873–2897 (2017)
3. Bianchi, M., et al.: Optimization-based scaling procedure for the design of fully
portable hand exoskeletons. Meccanica 53, 3157–3175 (2018)
4. Scheme, E., et al.: Motion normalized proportional control for improved pattern
recognition-based myoelectric control. In: 2017 IEEE/RSJ International Conference
on Intelligent Robots and Systems (IROS), Vancouver, BC, Canada, 24–28 Septem-
ber 2017
5. Ryser, F., et al.: Fully embedded myoelectric control for a wearable robotic hand
orthosis. In: 2017 International Conference on Rehabilitation Robotics (ICORR),
London, pp. 615-621 (2017)
Contribution of a Knee Orthosis
to Walking
O. Bordron(B) , C. Huneau, É. Le Carpentier, and Y. Aoustin
University of Nantes, Centrale Nantes, LS2N, UMR CNRS 6004, Nantes, France
olivier.bordron@ls2n.fr
Abstract. This study is devoted to investigate the influence of a knee

orthosis for human walking by using a mathematical model for a 7-links
planar biped - composed with two identical legs, two feet and one trunk
- with an orthosis attached to both thigh and calf, during walking. In the
first part, we design a cyclic walking gait in the sagittal plane for human
without the orthosis. The second part we consider that the human looses
his muscular possibilities in one of his knees. To overcome his handicap the
human is equipped of a knee orthosis. We analyse the positive effect of the
orthosis over the assisted knee to track the previous designed reference tra-
jectory. By using at each time an optimisation algorithm, we minimise the
torques provided by the human. The numerical tests confirm the possibil-
ity to reduce the torque produced in the disabled knee. The next step is
to take into account explicitly an information from EMG signal during a
walking to modulate the power of the orthosis.
1 Introduction
Active orthosis have been developed in medical environment. They can be used
for reducing chronic strains, compensating muscle impairments in case of reha-
bilitation or helping elderly people during walking (RobotKnee [1], EICoSI [2]).
Each assistive system has its own design, with its geometrical and mechanical
properties. These features have an impact over walking and over the operator [3].
Then, to estimate the impacts of an orthosis over walking gait before designing
the system is important.
Several studies using simulation and walking optimisation have already been
done [4,5]. Nevertheless, walking gait optimisation has never been done for opti-
mising the contribution of a knee orthosis. Depending on the reeducation stage
of the patient, we propose to optimise the torque profile supplied by an orthosis.
As we simply want to highlight the advantages and drawbacks of the proposed
method, the study is carried out with a 2D planar model.
The biped model, walking gait and optimal trajectories are defined Sect. 2.
Then, the dynamic model is completed with an orthosis on the left knee.
In Sect. 3, the knee orthosis contribution is optimised to carry out the same
walking.

https://doi.org/10.1007/978-3-030-01887-0_86
446 O. Bordron et al.

2.1 Model of the Biped
The model used to simulate a step is illustrated Fig. 1. This model is composed
of one trunk, two legs - each one is composed with both a thigh and calf - and
two feet. Model parameters of each link are those used in [8].
The left foot is considered flat on the ground (the left foot forms a complete
joint with the ground), that is to say for any time qp1 = 0 (see Fig. 1), and the
origin of the reference frame is taken at the left ankle. The created dynamic
model of the seven-link biped is detailed in [8].
Lp
Hf spy
q3 LT
Γ4 sT Lt Ld spx
Γ3 q4
q2 Γ5 st
sc
Γ2 q5
q1 Γ6
qp2 Lc
Γ1
(a) qp1 (b)
Fig. 1. Modelisation of the planed biped. (a) Parameterisation of the biped. Note that
angles are positive in counterclockwise. (b) Length segments and position of the centers
of mass.
T
t = t0 t= 2 t=T
q30 q3f
(xh , yh )
q10 q5f
x
Fig. 2. A step of the biped and conditions of cyclicality.
2.2 Definition of the Optimal Trajectories

The walking gait studied here is cyclic. Thus the walking gait can be defined by
designing over one step only (see Fig. 2). This step is defined by a phase of single
support and an impact with the ground. The impact model is the same used
Contribution of a Knee Orthosis to Walking 447
in [6]. We assume that trajectories describing the motion are defined by poly-
nomial functions of fourth order, depending on time. By using cyclic conditions,
we are able to describe the entire cyclic gait with these parameters:
(xhf , yhf , d, q3f , q̇f , qint ) (1)
(xh , yh ) is the configuration of the hip, d the step length, q3f the final position
of the trunk, q̇f the final velocity and qint the position at mid-cycle.
The design of the walking gait is made under the constraints: no take off and
no sliding of the stance foot during the single support, the center of pressure must
be inside the sole surface of the foot on the ground, and limits in magnitude of
the joint torques and joint velocities.
The cost function used for optimisation is the normalised dynamic effort [7].
From a data set of (1), the Matlab R
functions fmincon R
and MultiStart R
are
used to obtain the optimal trajectory in Fig. 4.
Fig. 3. Power of the knee at v = 1.0 m.s−1 and α = 0.4.
2.3 Contribution of the Orthosis

We add an orthosis to the biped model. It is modeled by one additional mass
with inertia to both calf and thigh. Its features are those identified for the EICoSI
orthosis [9]. Consequently, the dynamic model of the biped equipped with the
orthosis changes [8].
In this study, we assume that the orthosis is positioned on the knee of the
support leg. We are not interested here in the swing phase since the stance phase
requires much more torque to the knee.
The torque profile supplied by the orthosis is optimised by using the same
cost function as in Sect. 2.2, under the constraint Porth ≤ αPmax . α ∈ [0; 1] is a
reeducation coefficient reflecting the assistive level needed for a patient. It varies
448 O. Bordron et al.
throughout the reeducation. If α = 0, the knee is not assisted and if α = 1, the

knee is fully assisted. Pmax is the maximal power that must be provided by the
knee to carry out the trajectory with the orthosis.
3 Results
For example, at walking speed v = 1.0 m.s−1 , and for α = 0.4, the optimised
torque profile of the orthosis is given Fig. 3. The constraint Porth ≤ αPmax is
fulfilled. For t ∈ [0; 0.1], the patient has to provide power on the knee to carry
out the trajectory defined in Sect. 2. But for t ∈ [0.1; 0.2], the orthosis provides
all the power required to ensure the walking gait.
In the same way, we compute optimised torque profiles for different coefficient
α, at walking speed v = 1.0 m.s−1 . Energy distributions corresponding to these
results are given Fig. 5. They are normalised by the total energy spend at the
knee to carry out the step. It shows that the assistive factor α introduced in
Sect. 2 nearly corresponds to the normalised energy.
Fig. 4. An optimised step at walking Fig. 5. Optimised contribution of the

speed v = 1.0 m.s−1 . orthosis at v = 1.0 m.s−1 .
4 Conclusion
This paper has computed optimised torque profiles for the knee orthosis in order
to carry out a given trajectory (for instance the natural step of the patient), in
the case of a simple planar model. The numerical tests confirm the possibility to
reduce the torque produced in the disabled knee. Besides, we showed that the
assistive level can be modulated by the introduction of the assistive factor α.
Nevertheless, in the interests of realism, the model will be completed in a fur-
ther step by taking into account the impedance of the joints. Indeed, the impedance
of the tendon-muscle system can have a significant impact on the results.
Contribution of a Knee Orthosis to Walking 449
From this study, a possible extension is to extract information in time and

amplitude from the power profile in order to efficiently control the orthosis.
Another information could be extracted from the EMG signal to adjust the
desired trajectory.
References
1. Schmitt, C.: A study of a knee extension controlled by a closed loop functional
electrical stimulation. In: Proceedings of 9th Annual Conference of the International
FES Society, Bournemouth, pp. 135–137 (2004)
2. Mefoued, S.: A second order sliding mode control and a neural network to drive a
knee joint actuated orthosis. Neurocomputing 155, 71–79 (2015)
3. Huo, W., Mohammed, S., Moreno, J.C., Amirat, Y.: Lower limb wearable robots for
assistance and rehabilitation: a state of the art. IEEE Syst. J. 10, 1068–1081 (2016)
4. Tlalolini, D., Aoustin, Y., Chevallereau, C.: Design of a walking cyclic gait with
single support phases and impacts for the locomotor system of a thirteen-link 3D
biped using the parametric optimization. Multibody Syst. Dyn. 23, 33 (2009)
5. Aoustin, Y., Formalskii, A.M.: Strategy to lock the knee of exoskeleton stance leg:
study in the framework of ballistic walking model. In: Wenger, P., Chevallereau, C.,
Doina, D., Bleuler, H., Rodi, A. (eds.) New Trends in Medical and Service Robots,
vol. 3, pp. 179–195. Springer (2016)
6. Aoustin, Y., Formalskii, A.M.: Walking of biped with passive exoskeleton: evaluation
of energy consumption. Multibody Syst. Dyn. 43, 1–26 (2017)
7. Xiang, Y., Arora, J.S., Abdel-Malek, K.: Optimization-based prediction of asym-
metric human gait. J. Biomech. 44(4), 683–693 (2011)
8. Bordron, O., Le Carpentier, É., Huneau, C., Aoustin, Y.: Impact of a knee orthosis
over walking. In: Arakelian, V., Wenger, P. (ed.) ROMANSY 22 – Robot Design,
Dynamics and Control, pp. 466–473. Springer (2019)
9. Mefoued, S., Mohammed, S., Amirat, Y., Fried, G.: Sit-to-stand movement assis-
tance using an actuated knee joint orthosis. In: International Conference on Biomed-
ical Robotics and Biomechatronics, Rome, pp. 1753–1758 (2012)
Human Trunk Stabilization with Hip
Exoskeleton for Enhanced
Postural Control
Marko Jamšek1,2(B) and Jan Babič1

1
Laboratory for Neuromechanics and Biorobotics, Department of Automation,
Biocybernetics and Robotics, Jožef Stefan Institute, Ljubljana, Slovenia
marko.jamsek@ijs.si
2
Jožef Stefan International Postgraduate School, Jamova cesta 39,
1000 Ljubljana, Slovenia
Abstract. Tripping is a major cause of falls in elderly people. Consid-

ering that fall related injuries have severe consequences on their quality
of life, there is an urgent need to develop preventive solutions. One such
solution could be the use of assistive exoskeletons. In this work we inves-
tigated the effects of a hip exoskeleton on human posture under the influ-
ence of an external perturbation. During normal standing of a subject we
applied a forward pulling force at the chest and then enabled or disabled
the exoskeleton randomly. By analysing the kinematics of the human
body we compared responses to perturbations when the exoskeleton was
enabled or disabled. The results show that the exoskeleton efficiently
reduced the forward inclination of the trunk by 40%.
1 Introduction
Fall related injuries can have severe consequences on the quality of life of elderly
people. One of the main causes of falls is tripping [1,2]. It was shown that there
are predominantly two strategies for recovery after tripping [3]. In one of these
strategies it is the support limb that plays an important role in reducing for-
ward angular momentum during recovery [4,5]. Further work with elderly people
showed that some individuals cannot properly reduce the angular momentum
after a trip [6]. A possible solution for this would be the use of an exoskeleton to
provide additional torque to counteract this uncompensated forward momentum.
In this work we laid down the foundation for such a device. We investigated the
effects of a hip exoskeleton on human trunk stabilization during standing under
the influence of an outside perturbation.
This work was supported by the European Union’s Horizon 2020 through the
SPEXOR project (contract no. 687662); AnDy project (contract nr. 731540); and
by the Slovenian Research Agency (research core funding no. P2-0076).
https://doi.org/10.1007/978-3-030-01887-0_87
Human Trunk Stabilization with Hip Exoskeleton 451

For the experiment we used a bilateral powered exoskeleton (Fig. 1) comprised of
a belt, chest and leg straps, and the actuation mechanism. This mechanism con-
sists of a bidirectional pneumatic cylinder and a rotational joint on each side. The
applied pressure in the cylinder chambers produces torque at the hip to aid in
flexion or extension of the legs. Both active joints are equipped with a rotational
encoder to provide feedback about the current position of the exoskeleton. To
collect kinematic data of the subject we used the MVN Awinda motion capture
system (Xsens, Enschede, Netherlands).
Fig. 1. Photograph of the experimental setup.
The controller was developed in Simulink Real-Time (Mathworks, Natick,

MA, United States). The control schematic is presented in Fig. 2. The inputs
are the absolute trunk angle of the human and the encoder angles. The output
is a voltage signal that controls the pneumatic valves trough the pneumatics
controller. We designed the controller to maintain a constant angle of the trunk.
The angle was set to 0◦ which corresponds to a normal upright stance.
Fig. 2. Schematic representation of the exoskeleton control.

452 M. Jamšek and J. Babič
A subject was instructed to stand still while wearing the exoskeleton and the
motion capture sensors. To induce perturbations, a motorized pulley system was
used which was attached around the subject’s chest. To eliminate slack in the
cable, a pretension pulling force of 10 N was used. During the experiment the
pulley system induced a square signal perturbation of 100 N for 250 ms every
10 s. The time of 250 ms was arbitrarily chosen to mimic a short force impulse.
We then increased the perturbation force in 20 N increments to determine the
maximum pulling force the subject could resist, before making a step forward.
Each instance of the perturbation represents one trial. The exoskeleton was
enabled or disabled randomly at the beginning of every trial. This eliminated
the possibility of anticipating the assistance of the exoskeleton. A total of 26 trials
were carried out, 13 of which had the eksoskeleton enabled and 13 disabled.
The kinematics of the human motion was recorded using the MVN Analyze
(Xsens, Enschede, Netherlands) software. All other data was recorded in real
time within Simulink Real-Time (Mathworks, Natick, MA, United States). A
time-stamp of the kinematics recording was streamed to the controller and was
later used to synchronize all the data. Ad-hoc post-processing of the data was
performed in Matlab (Mathworks, Natick, MA, United States).
3 Results
To evaluate the effects of the exoskeleton on trunk stabilization we measured the
time response of the trunk angle with and without exoskeleton assistance. We
compared the response to the perturbation when the exoskeleton was enabled or
disabled (Fig. 3). The graph represents the mean and standard deviation of the
trunk-angle time-response of all trials. In the trials with the exoskeleton enabled,
the maximum angle of inclination was 40% lower than when the exoskeleton was
disabled. In addition, the time needed for the subject to stabilize (ts ) was shorter
(t1 and t2 ). Furthermore, after the subject regained normal stance, the variation
of the trunk angle was lower with the exoskeleton enabled.
Quantitative measures for these characteristics are presented in Table 1. Time
ts is defined as the time where the trunk angle falls between ±1 degree of the
final stabilized angle (from 2.5 s to 3 s after the start of the perturbation). When
the exoskeleton was enabled the time to return to a normal stance was 1.4 s and
1.75 s when the exoskeleton was disabled.
Table 1. Parameters of the trunk angle time response
Exoskeleton max(Angle(t)) [deg] ts [s] SD(Angle(3 s)) [deg]

Disabled 12.8 1.75 1.9
Enabled 7.8 1.4 0.5
Change [%] −40% −20% −74%
Human Trunk Stabilization with Hip Exoskeleton 453
Fig. 3. Mean and standard deviations of the trunk angle during the perturbation (grey
shaded area) while the exoskeleton was disabled (blue line) or enabled (red line).
t1 and t2 represent the time needed for the subject to stabilize with the exoskeleton
enabled or disabled respectively.
4 Discussion
From the analysis of the kinematics we observed that the activation of the
exoskeleton significantly decreases the forward inclination of the trunk. This
indicates that the exoskeleton is capable of producing torques to counteract a
portion of the forward momentum gained from the perturbing force. If used cor-
rectly this could compensate the insufficient torque produced by the elderly [6]
and possibly prevent falls when tripping. The activation of the exoskeleton also
ensures a shorter time to regain the vertical pose (t1 vs. t2 ). This is possibly due
to the limitation of the maximal forward inclination. Additionally, the variation
of the trunk angle after the subject regained normal stance is lower when the
exoskeleton was enabled. This indicates that the exoskeleton improved stability
of the subject even well after the perturbation.
5 Conclusion
In conclusion, we have seen positive effects of using an exoskeleton for trunk
stabilization in a standing scenario. The main limitations of this work were that
we measured only one subject and we acquired only kinematic data for a simple
standing scenario. We would like to extend this work to walking where one of
the main challenges will be to provide assistance in the appropriate phase of the
gait cycle. Additionally, we want to consider the human-exoskeleton interactions
also in terms of forces and torques. With the addition of measurements of foot
reaction forces, we could calculate the inverse dynamics of the human body
during such an experiment. We could then evaluate the effects of the exoskeleton
on the angular momentum of the human body. Furthermore, the additional
torque provided to the support limb should expand the range of achievable foot
454 M. Jamšek and J. Babič
placements during walking. This will potentially lead to a solution that will
provide assistance during tripping while walking.
References
1. Berg, W.P., Alessio, H.M., Mills, E.M., Tong, C.: Circumstances and consequences
of falls in independent community-dwelling older adults. Age Ageing 26(4), 261–268
(1997)
2. Roudsari, B.S., Ebel, B.E., Corso, P.S., Molinari, N.A.M., Koepsell, T.D.: The acute
medical care costs of fall-related injuries among the U.S. older adults. Injury 36(11),
1316–1322 (2005)
3. Eng, J.J., Winter, D.A., Patla, A.E.: Strategies for recovery from a trip in early and
late swing during human walking. Exp. Brain Res. 102(2), 339–349 (1994)
4. Pijnappels, M., Bobbert, M.F., Van Dieën, J.H.: Contribution of the support limb in
control of angular momentum after tripping. J. Biomech. 37(12), 1811–1818 (2004)
5. Pijnappels, M., Bobbert, M.F., Van Dieën, J.H.: How early reactions in the sup-
port limb contribute to balance recovery after tripping. J. Biomech. 38(3), 627–634
(2005)
6. Pijnappels, M., Bobbert, M.F., Van Dieën, J.H.: Push-off reactions in recovery after
tripping discriminate young subjects, older non-fallers and older fallers. Gait Posture
21(4), 388–394 (2005)
Development of a Wearable Haptic Feedback
System for Limb Movement
Symmetry Training
Amre Eizad1, Muhammad Raheel Afzal2, Hosu Lee2, Sung-Ki Lyu1,

and Jungwon Yoon2(&)
1
School of Mechanical and Aerospace Engineering,
Gyeongsang National University, Jinju, Republic of Korea
2
School of Integrated Technology,
Gwangju Institute of Science and Technology, Gwangju, Republic of Korea
jyoon@gist.ac.kr
Abstract. Arm swing asymmetry brought on by hemiparesis or hemi-neglect

due to stroke causes increased ground reaction moments during walking. Thus,
increasing loading asymmetry of the lower limbs and the metabolic load of
walking. The system presented here consists of wearable bracelets, designed to
help decrease arm swing asymmetry in people with stroke. Design of the bra-
celets and outcomes of pilot tests with a healthy subject with artificially induced
arm swing asymmetry are presented. The healthy subject was able to utilize the
vibration feedback to reduce arm swing asymmetry with involuntary increase in
arm swing magnitude. Thus, exploration of the effects of using this system with
stroke patients is warranted.
1 Introduction
Wearable devices have become a ubiquitous part of modern life. Such systems are also
being incorporated in physical rehabilitation. Previously, we have successfully
implemented wearable systems that provide kinesthetic [1] and tactile [2] feedback to
the user.
The system presented here has been developed to aid upper limb rehabilitation with
focus on improving upper limb swing symmetry during walking. People with hemi
paresis or hemi neglect due to stroke have asymmetric arm swing during walking [3].
Arm swing helps with walking by generating moment around the vertical axis, which
augments the ground reaction moment generated by the foot in stance phase [4].
Altering the arm swing causes increased ground reaction moments and increased
metabolic load [5]. Thus, asymmetric arm swing may cause asymmetric moment
The Ministry of Culture, Sports and Tourism (MCST) and Korea Creative Content Agency
(KOCCA) through the Culture Technology Research & Development Program 2018, Dual Use
Technology Program of Civil and Military, and Global University Project (GUP 2018) by
Gwangju Institute of Science and Technology supported this research work.

https://doi.org/10.1007/978-3-030-01887-0_88
456 A. Eizad et al.
generation, causing asymmetric loading of the legs during gait. Drawing attention to
the neglected arm may reduce the amount of difference in arm movements during
walking [3]. Feedback of arm activity level difference between healthy and neglected
arm has been shown to have a positive effect [6]. Vibration feedback aimed at
increasing arm swing magnitude has been shown to have promising outcomes in
subjects with Parkinson’s disease (PD) [7]. A feedback device may be beneficial for
use because a person cannot visually judge the range of arm swing happening behind
their back, and especially directing the eyes and mind towards judging the range of
their arm swing may actually increase asymmetry as increased cognitive load affects
the arm swing [8]. This paper presents design of the system components and results of
pilot tests with a healthy subject.
The system consists of two identical wearable devices worn one on each forearm. The
devices measure limb orientations and communicate them to a computer and can
generate vibration according to commands given by the computer. The wearable device
has been designed in the form of a bracelet with an adjustable diameter to allow ease of
fitting to extremities of different sizes.
The bracelet consists of one block containing battery, Inertial Measurement Unit
(IMU), microcontroller and Wi-Fi module, and four blocks containing one individually
addressable vibro-tactor each to enable generation of vibration with desired intensity
and duration in different patterns around the user’s limb. The blocks are joined using a
strap that has Velcro® fasteners and electrical linkage is maintained by connecting
wires. The IMU (MPU9250, Invensense, USA) and microprocessor (SAMD21, Atmel,
USA) are housed on one board (9DoF Razor IMU M0, Sparkfun, USA). Drive signals
to the vibrotactors (310-101, Precision Microdrives, UK) are sent via a driver
(ULN2803A, Toshiba, Japan). A Wi-Fi module (ESP8266, Espressif Systems, China)
enables wireless communication. Figure 1(a) provides block diagram of the bracelet
powered by a 3.7 V 850mAh Lithium ion battery. The bracelet hardware casings are
3D printed plastic (Fig. 1(b)). The system is extendable with the inclusion of more
bracelets. We will ultimately replace the computer with a smartphone.
Pilot testing of the system as a means of reducing arm swing asymmetry during gait
was done with a healthy subject at Gwangju Institute of Science and Technology,
Republic of Korea. The subject performed 3 treadmill walking trials of 1 min each at a
pace of 1 m/s. First was normal walk, then walk with perturbation of the arm swing to
generate asymmetry, last was walk with perturbation and feedback provided by the
bracelet to reduce arm swing asymmetry. A mass of 1.8 kg was strapped to the sub-
ject’s right wrist (Fig. 1(c)) to generate swing asymmetry [9]. Vibration cues were
provided to the weighted arm when its swing angle became equal to the peak angle of
the other arm during the last swing. The swing amplitude data taken form bracelets
worn on both arms was recorded.
Development of a Wearable Haptic Feedback System 457
Fig. 1. (a) Bracelet block diagram. (b) Bracelet prototype. (c) Two bracelets and weight worn
by participant. (d) Angle sign convention.
We tested the bracelets to validate their design and determine the practical functional
bounds. Each bracelet weighs 110 g. Based on its measured current draws, the bracelet
can run at nominal load for over four hours with 50% discharge of the onboard battery.
The bracelet has a measured communication range of above 20 m indoors.
Figure 2 shows representative arm swing cycles from each of the trials. Here, zero
reference represents the arms hanging vertically by the subject’s side. Positive angles
refer to movement to the front of the subject and negative angles represent movements
beyond the reference towards the back side of the subject (see Fig. 1(d)). It can be seen
in Fig. 2(a) that although swings of both the arms are not completely identical, they are
however quite close to each other in terms of peak magnitudes. Figure 2(b) shows that
addition of mass to the right arm reduced its positive peak while increasing the negative
one and increased the positive while reducing the negative peak of the left arm.
Figure 2(c) shows that with provision of feedback, subject was well able to reduce the
negative peak of right arm swing to match the left arm while increasing the positive
peak, but was not able to match the other arm as the swing magnitudes of the left arm
also increased in both directions. These changes in both arms indicate that swing
magnitudes may be inter-dependent, biomechanical implications of this need further
exploration. The subject was able to modify their arm swing based on the provided
458 A. Eizad et al.
feedback, showing that the system may be a feasible means of providing arm swing
asymmetry feedback.
Fig. 2. Subject arm angles in degrees (y-axis) plotted against time (x-axis). (a) Normal walk.
(b) Walk with perturbation on right arm. (c) Walk with perturbation on right arm and vibration
feedback.
The system was observed to be supportive in allowing a healthy person to reduce arm
swing asymmetry induced by artificial perturbation. Thus, we envisage use of this
system for ameliorating arm swing symmetry in stroke patients. We have previously
observed that vibration feedback about temporal stance asymmetry provided to subjects
with hemiparesis due to stroke has a significantly beneficial effect [10]. This device
may be able to provide similar benefits with regards to the upper extremities. Thus, we
intend to explore the effects of using this system with subjects with stroke in our future
studies. We also intend to evaluate the perception and effects of different haptic
biofeedback schemes with stroke patients.
REFERENCES
1. Afzal, M.R., Eizad, A., Palo Pena, C.E., Yoon, J.: Evaluating the effects of kinesthetic
biofeedback delivered using reaction wheels on standing balance. J. Healthcare Eng. (2018).
Article ID: 7892020
2. Afzal, M.R., Lee, H., Yoon, J., Oh, M.K., Lee, C.H.: Development of an augmented
feedback system for training of gait improvement using vibrotactile cues. In: 14th
International Conference on Ubiquitous Robots and Ambient Intelligence (URAI), pp. 818–
823. IEEE, June 2017
3. Punt, T.D., Riddoch, M.J.: Motor neglect: implications for movement and rehabilitation
following stroke. Disabil. Rehabil. 28, 857–864 (2006)
4. Park, J.: Synthesis of natural arm swing motion in human bipedal walking. J. Biomech. 41,
1417–1426 (2008)
5. Collins, S.H., Adamczyk, P.G., Kuo, A.D.: Dynamic arm swinging in human walking. Proc.
Royal Soc. London B: Biol. Sci. 276, 3679–3688 (2009)
Development of a Wearable Haptic Feedback System 459
6. Trejo-Gabriel-Galan, J.M., et al.: Rehabilitation of hemineglect of the left arm using

movement detection bracelets activating a visual and acoustic alarm. J. Neuroeng. Rehabil.
13, 79 (2016)
7. Thompson, E., Agada, P., Wright, W.G., Reimann, H., Jeka, J.: Spatiotemporal gait changes
with use of an arm swing cueing device in people with Parkinson’s disease. Gait posture 58,
46–51 (2017)
8. Killeen, T., et al.: Increasing cognitive load attenuates right arm swing in healthy human
walking. Royal Soc. Open Sci. 4, 160993 (2017)
9. Donker, S.F., Mulder, T., Nienhuis, B., Duysens, J.: Adaptations in arm movements for
added mass to wrist or ankle during walking. Exp. Brain Res. 146, 26–31 (2002)
10. Afzal, M.R., Oh, M.K., Lee, C.H., Park, Y.S., Yoon, J.: A portable gait asymmetry
rehabilitation system for individuals with stroke using a vibrotactile feedback. BioMed Res.
Int. (2015). Article ID: 375638
Failure Mode and Effect Analysis
(FMEA)-Driven Design of a Planetary Gearbox
for Active Wearable Robotics
Pablo López García(&), Stein Crispel, Tom Verstraten, Elias Saerens,

Bryan Convens, Bram Vanderborght, and Dirk Lefeber
Robotics and Multibody Mechanics Research Group (R&MM),

Faculty of Mechanical Engineering, Vrije Universiteit Brussel and Flanders
Make, Pleinlaan 2, 1050 Elsene, Belgium
plopezga@vub.be
Abstract. Conducting an FMEA for the design of a planetary gear transmission

for exoskeletons enables decision making based on the interdependence between
design parameters and the device requirements, as well as an early identification of
several functional risks. Therefore, the use of FMEAs in the design of wearable
robotic devices could contribute to higher design robustness, and ultimately result
in a broader acceptance of future active wearable robotic devices.
1 Introduction
The selection of a suitable actuating system for a given application is a common task in
machine engineering. In wearable robotics, actuating systems collaborate very closely
with the biomechanical actuators of the human body, to improve the performance of the
latter. This situation conditions their movement – to match the extraordinary versatile
mechanical characteristics of their muscle-based peers – challenging the selection of
suitable transmissions.
In this work, we propose to integrate Failure Mode and Effect Analysis (FMEA) in
the design process of robotic actuating systems to help manage this complexity. FMEA is
a step-by-step approach to identify and categorize all possible failures in a design or
manufacturing process [1]. Originally invented around 1950 by the US Army and used in
multiple NASA space programs, the automotive industry is reputed for having exploited
its full potential to (i) put the user’s need at the center of the complete product-design
process, (ii) test and improve the accuracy of an initial specification (set of requirements),
and (iii) identify interdependencies between design decisions and the requirements of the
specification, all these being also valuable elements for wearable robotics [2].
E. Saerens and B. Convens—SB PhD Fellows at the Research Foundation Flanders – Fonds voor
Wetenschappelijk Onderzoek (FWO). This work has been partially funded by the European
Commission ERC Starting grant SPEAR (no. 337596).
The authors would like to express our thanks to APIS Informationstechnologie GmbH (www.
apis-iq.com) for supplying the FMEA software used in this research project.

https://doi.org/10.1007/978-3-030-01887-0_89
Failure Mode and Effect Analysis (FMEA)-Driven Design 461
2 FMEA-Driven Transmission Design
In practice, transmissions are selected from usual technologies used in robotics –

Harmonic Drives, planetary gearheads or cycloidal drives among others – to shift the
torque-speed characteristic of the actuator and to cope with size and weight restrictions.
This choice tends to be strongly dependent on the previous experiences of the engineer.
To systematize this process and understand the potential of using customized
instead of standard gearheads, we at the R&MM group conducted a Design-FMEA
analysis of a planetary gear transmission for exoskeleton’s hip actuation.
2.1 Product Specifications

Putting the user’s needs at the center of the design is of fundamental importance for the
product acceptance and begins with the definition of a robust product specification.
Human actuators are not characterized by a very high efficiency or high specific
power characteristics [3–5]. However, they can provide impressive specific forces well
beyond the capabilities of our current actuators, explaining the need for transmissions.
And they are enormously versatile to assist the highly dynamic biomechanical actua-
tion, with fast and continuously changing speeds and fast variations of the mechanical
impedance (ratio between torque and speed) within a very broad range of values.
Finally, the narrow collaboration between robotic and biological actuators in
exoskeletons introduces as well unprecedented mechanical and ergonomic challenges
in terms of compliant mechanical interfacing, weight distribution and autonomy.
All these aspects are not yet sufficiently understood and stay in the focus of current
research activities [6–8]. To integrate them in a robust set of requirements, we collected
the input of experienced robotic engineers and completed it with further inputs from the
literature [5, 8].
2.2 System Structure and Functional Net

The second step of the FMEA consists in defining the assembly structure of the
components and subcomponents of the planetary gear transmission.
We then linked each of these components with the specifications through a
Functional Net, identifying for each component the internal functions which are
responsible for the fulfillment of each of the requirements at complete system (trans-
mission) level, see example in Fig. 1.
2.3 Failure Analysis: Consequences and Causes

Possible Malfunctions of each of the derived internal functions of the elements of the
planetary gear train, together with their Consequences on other internal functions and
on the overall product requirements, were then analyzed with the aid of the Functional
Net structure.
Additionally, all Potential Causes that could result in these malfunctions were
identified back to the lowest component level, and included, together with the previ-
ously found Consequences, in a Failure Tree (Ishikawa) structure (Fig. 1).
462 P. L. García et al.
Fig. 1. Extract of the Failure Tree showing part of the Functional Net of the planetary gear, and
a portion of the causal-interdependency for the Malfunction “meshing Contact Ratio insufficient
(0 < CR < 1)”.
2.4 Risk Assessment

Risks are combinations of a certain Potential Cause, a Malfunction and a Consequence.
To assess and categorize them, three main criteria are used: Occurrence (O), Detection
(D) and Severity (S). Occurrence refers to the probability of the Potential Cause to
ultimately occur and in our case, it is linked to the definition of suitable tolerances and
safety coefficients. Detection refers to the probability of being able to detect the
presence of a Malfunction during the validation (testing, simulation, etc.). Severity is
used to assess the criticality of Consequences and must be established at transmission
overall level (specifications).
For each of these criteria, standardized reference rating tables [1] were used to
assess the risks for each malfunction. An additional evaluation criterion (Risk Priority
Number - RPN) was generated multiplying these three criteria (Fig. 2).
Fig. 2. Assessments of several risks related to the Malfunction “Impossibility to transfer any
Torque”
2.5 Optimization
Finally, we reviewed Risks, Potential Causes and Malfunctions associated with the
highest values of O, D, S and RPN, to assess how design changes, additional testing or
simulation could improve the current design performance.
Failure Mode and Effect Analysis (FMEA)-Driven Design 463
3 Results
Our study allowed us to upgrade our initial design to exploit the potentials resulting
from (i) using Ferguson Paradox- planetary gear trains [9] to generate high gear ratios,
(ii) adapting the gear teeth shape to the asymmetric torque, back-drivability and
backlash demands, and (iii) selecting the diameter to width ratio to optimize ergonomic
footprint and minimum gear teeth size to bear the contact surface and bending loading
for a certain torque output.
Additionally, it also confirmed the important impact of current limitations of pro-
duct design in robotics already well identified in previous literature [5, 8, 10]. These
limitations result from (i) our limited ability to define a robust specification for actu-
ators due to the complexity of the mechanics and control of the human body, and from
(ii) the strongly personalized performance criteria due to the absence of generally
agreed validation criteria, adequately integrating inter-user and inter-task variability.
A possible solution to standardize performance evaluation, integrating complex
inter-user and inter-task variability and following a similar in approach to the Bench-
marking in Locomotion initiative [10] and to robot competitions like CYBATHLON,
could be based on the use of driving cycles. Driving cycles are successfully applied to
homologate vehicles and compare technologies in the automotive industry, where
performance depends strongly on the user’s driving style and usage conditions [11],
and they are a focus of future research in our group.
4 Conclusion
In conclusion, we believe that FMEAs can help make adequate decisions and identify
potentials in the design of actuating elements for wearable robotics, while putting the
user’s needs at the center of the design process.
Their use, combined with the application of usage-adapted driving cycles to vali-
date and compare the performances of different solutions, could contribute to
improving the acceptance of future active wearable robotic devices.
References
1. Potential Failure Mode and Effects Analysis FMEA Reference Manual, 4th (edn). ©GM
corp. (2008). ISBN #9781605341361
2. Mathijssen, G., et al.: From series elastic actuation to series-parallel elastic actuation. In:
Proceedings of International Conference on New Actuators (2014)
3. Beckerle, P.: Practical relevance of faults, diagnosis methods, and tolerance measures in
elastically actuated robots. Control Eng. Pract. 50, 95–100 (2016)
4. Madden, J., et al.: Artificial muscle technology: physical principles and naval prospects.
IEEE J. Oceanic Eng. 29(3), 706–728 (2004)
5. Veale, A.J., Xie, S.Q.: Towards compliant and wearable robotic orthoses: A review of
current and emerging actuator technologies. Med. Eng. Phys. 38, 317–325 (2016)
464 P. L. García et al.
6. Sugar, T.G., Holgate, M.: Compliant mechanisms for robotic ankles. In: ASME International
Design Engineering Technical and Computers and Information in Engineering Conferences,
vol. 6A, August 2013
7. Verstraten, T., et al.: Optimizing the power and energy consumption of powered prosthetic
ankles with series and parallel elasticity. Mech. Mach. Theory 116, 419–432 (2017)
8. Cenciarini, M., Dollar, A.M.: Biomechanical considerations in the design of lower limb
exoskeletons. In: IEEE International Conference on Rehabilitation Robotics, July 2011
9. Ferguson, J., Henderson, E.: Life of James Ferguson, F.R.S. Cambridge University Press
(1867)
10. Torricelli, D., et al.: Benchmarking bipedal locomotion: a unified scheme for humanoids,
wearable robots, and humans. Robot. Autom. Mag. 22(3), 103–115 (2015)
11. Barlow, T.J., Latham, S., McGrae, I.S., Boulter, P.G.: A reference book of driving cycles for
use in the measurement of road vehicle emissions. TRL Ltd. (2009)
Introducing Series Elastic Links
Andrea Calanca1(B) , Luca Bettinelli2 , Eldison Dimo1 , Rudy Vicario1 ,

Mauro Serpelloni2 , and Paolo Fiorini1
1
Department of Computer Science, University of Verona, Verona, Italy
andrea.calanca@gmail.com
2
Department of Information Engineering, University of Brescia, Brescia, Italy
Abstract. This paper introduces the concept of Series Elastic Link

which exploits the inherent elasticity of flexible links to implement com-
pliant actuation and control using lightweigt and low cost components.
1 Introduction
The use of robotics in contemporary society is rapidly increasing. Collabora-

tive robots, rehabilitation robots and assistive robots are emerging examples
of the upcoming robotic revolution. However, if we want to spread the robotic
technology among the population we must start accounting for affordability in
robot research and design. Even if the prices of robots is continuously decreas-
ing thanks to scale economies, further steps are required to target the disposable
income of the majority of population. Up to now, few research efforts are spent
to investigate the design paradigms that will guarantee both affordability and
the capabilities to cope, support and interact with humans. Robots are expected
to sense the physical interaction and to use force control technology to provide
a natural and safe interaction [1,2].
This paper proposes a fundamental building block in the direction of design-
ing affordable torque-controlled robots. Basing on the assumption that plastic
materials will substitute metal robot parts we introduce the concept of Series
Elastic Link (SEL) which exploits the inherent flexibility of plastic links to imple-
ment compliant actuation and control using inexpensive components. Differently
from the existing literature on flexible links, the concept of SEL introduces the
perspective of taking advantage of link compliance rather than counteracting it,
following the idea of series elastic actuators (SEA) [1]. However, with respect to
traditional SEA design, the SEL architecture substitutes two high-cost compo-
nents (the series compliance and the rigid link) with a single inexpensive plastic
link. In the SEL design the link itself becomes the force sensor leading to further
advantages with respect to SEA such as to mask the inherent link inertia and to
more accurately estimate interaction forces (because the force sensor is on the
link and not before the link as in SEA). Also, thanks to lightweight and elasticity
of plastic materials impact tolerance is improved.

https://doi.org/10.1007/978-3-030-01887-0_90
466 A. Calanca et al.
2 Series Elastic Link Modeling

An established model for series compliance is the one represented in Fig. 1 where
θ is the motor position, q is the environment position, τm is the motor input
torque (proportional to the current), τs is the spring torque and τe represents the
environment forces. The actuator parameters are the spring stiffness k and the
reflected motor-gear inertia Jm . For simplicity of representation Fig. 1 translates
angular variable (angles and torques) into linear equivalents (linear positions
and forces), nevertheless this work refers to angular quantities. The model in
Fig. 1 can be described as
Jm θ̈ = τm − τs (1)
where the spring torque is defined as τs = k(θ − q). The SEL dynamics can be
described by the same model (1) where the torque τs is defined as
τs = keq (θ − q) (2)
and keq represents an equivalent torsional stiffness given by the link. To define
such equivalent torsional stiffness, a cantilever beam model is considered for the
link as represented in Fig. 2a where L is the beam length and Δx represents the
end point displacement due to a perpendicular perturbing force F applied on
the link endpoint. In static conditions, the cantilever beam model leads to the
following relation
L3
Δx = − F (3)
3Iρ
where ρ is the Young modulus of the beam material and I represents the inertia
of the beam section. By translating linear quantities into angular equivalents and
by considering the force F as exerted by the environment (such that τe = F L)
the following equality can be computed
3Iρ
τs = −τe = (θ − q) = keq (θ − q) (4)
L2
which states that the stiffness keq is given by structural beam properties, such
as the mass distribution and the link length.
τs −τe
τm
Jm
Environment
k
θ q
Fig. 1. A wide used model for series compliance.

Introducing Series Elastic Links 467
−Fe
M Δx
R
Fig. 2. A flexible robot link modeled as a cantilever beam and deformed by a force F
applied on the link endpoint.
3 Series Elastic Link Implementation and Experimental

Results
In this work the SEL architecture is implemented using a polyethylene hollow

tube with diameter of 40 mm and a length of 200 mm mounted on a geared
DC motor. To sense the link deformation we use a couple of strain gauges and a
custom electronic board equipped with a Wheatstone bridge and an instrumental
amplifier. The relation between the amplifier output Vout and the link torque τs
has been identified using a commercial high-end force sensor which measures
the interaction forces with the environment, as represented in Fig. 3. The best
validation fitting has been found using a linear regression in the form τs = αVout
meaning that the strain gauges are working in the linear region. Force control
is implemented using a proportional feedback τm = Kp (τref − τs ), where τref is
the desired force or torque and Kp is a proportional gain. Force control is tested
using discontinuous and sinusoidal force references during interaction with a
rigid environment. Force tracking is shown in Fig. 4 where the τs is the torque
estimated using strain gauges and τe is the torque measured using a high-end
force sensor. It can be observed that:
(1) The step response is quite fast and accurate with a rise time of about 8 ms
and a maximum static error of 5%. An overshoot peak can be observed which
is due to the inherent link elasticity. The overshoot disappears for references
below 6 Hz and appears for frequencies around 8 ÷ 10 Hz which represent the
maximum force bandwidth.
(2) The torque estimated using stain gauges (τs , in red) is practically indistin-
guishable from the one measured by the high-end commercial sensor (τe , in
yellow).
(3) The inherent link compliance causes high strain gauge deformations lead-
ing to improve the signal to noise ratio. Figure 5 reports on the left plot a
magnification to graphically compare the noise on the signals τs and τe : the
noise variance in the signal τs is 11.0 times lower than in the signal τe
468 A. Calanca et al.
Fig. 3. The single d.o.f. Series Elastic Link testbed used during experiments.
1.5
0.5
5.8 6 6.2 6.4 6.6 6.8 7 7.2 7.4 7.6 7.8
0.8
0.6
0.4
0.2
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
0.5
0
1 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2
Fig. 4. Force tracking with discontinuous (upper plot) and sinusoidal force references
at 6 Hz (middle plot) and 8 Hz (lower plot) frequencies.
Impedance control has been implemented on the SEL prototype by adding a

collocated outer position feedback τref = −kdes θ as explained in [2,3]. The out-
come of the impedance controlled SEL when hand perturbed is reported in the
right plot of Fig. 5 in a force-position plot where the red line indicates the desired
force-position relation (corresponding to a pure stiffness kdes = 2.0 Nm/rad) and
the blue line reports the actual force-position relation of the SEL system.
Introducing Series Elastic Links 469
1.02
2
1
0.98
1
0.96
0.94 0
0.92
-1
0.9
0.88
-2
0.86
3.32 3.33 3.34 3.35 -1.5 -1 -0.5 0 0.5 1 1.5
Fig. 5. Left: magnification of a force tracking plot. Right: accuracy of the impedance
controlled SEL
References
1. Pratt, G.A., Williamson, M.: Series elastic actuators. In: International Conference
on Intelligent Robots and Systems, vol. 1, pp. 399–406. IEEE (1995)
2. Calanca, A., Muradore, R., Fiorini, P.: A review of algorithms for compliant control
of stiff and fixed compliance robots. IEEE Trans. Mechatron. 21(2), 613–624 (2016)
3. Calanca, A., Fiorini, P.: Impedance control of series elastic actuators: passivity and
acceleration-based control. Mechatronics 47, 37–48 (2017)
Polymer Optical Fiber Sensors Approaches
for Insole Instrumentation
Arnaldo G. Leal-Junior1, Antreas Theodosiou2, Anselmo Frizera1(&),

Maria F. Domingues4, Cátia Leitão4, Kyriacos Kalli2, Paulo André3,
Paulo Antunes4, Maria José Pontes1, and Carlos Marques4
1
Graduate Program of Electrical Engineering of Federal,
University of Espírito Santo, Vitória, Brazil
arnaldo.leal@aluno.ufes.br, frizera@ieee.org,
mjpontes@ele.ufes.br
2
Nanophotonics Research Laboratory,
Cyprus University of Technology, 3036 Limassol, Cyprus
3
Instituto de Telecomunicações and Department of Electrical
and Computer Engineering, Instituto Superior Técnico, University of Lisbon,
1049-001 Lisbon, Portugal
4
Instituto de Telecomunicações and Physics Department and I3N,
Campus Universitário de Santiago, Aveiro, Portugal
{fatima.domingues,catia.leitao,carlos.marques}@ua.pt,
pantunes@av.it.pt
Abstract. Advantages like electromagnetic field immunity, fracture toughness,

high strain limits, flexibility in bending and impact resistance of polymer optical
fibers (POFs) are beneficial for applications that involve embedment in flexible
structures. Since insoles are one of these flexible structures that may be used in
different wearable applications, POFs can be applied and this paper proposes the
application of POF sensors in insole instrumentation with two different
approaches: intensity variation-based and polymer optical fiber Bragg gratings
(POFBGs). Results show that both approaches present low errors with root
mean squared errors (RMSEs) of 45.17 kPa for the plantar pressure monitoring
with the POFBG-based insole and 5.30 N for the ground reaction force mea-
surement with the intensity variation sensors. These results demonstrate the
feasibility of POF sensors applications in flexible structures and in wearable
applications such as insoles and soft robotics instrumentation.
1 Introduction
One of the key parameters to monitor during gait is the foot plantar pressure, which is
an important indicator of the foot health condition and gait pattern [1]. Information
about the spinal cord condition and the evolution of foot ulcerations in diabetes patients
This research is financed by CAPES (88887.095626/2015-01), FAPES (72982608), CNPq

(304192/2016-3 and 310310/2015-6), FCT by the National Funds through the FCT/MEC and the
ERDF under the PT2020 Partnership Agreement (SFRH/BPD/109458/2015,
UID/EEA/50008/2013).

https://doi.org/10.1007/978-3-030-01887-0_91
Polymer Optical Fiber Sensors Approaches for Insole Instrumentation 471
can be inferred from the plantar pressure distribution map [2]. Another important
information that can be inferred by the foot plantar pressure is the monitorization of
ground interaction forces broadly used in gait analysis. Gait cycle is defined as the
interval between two consecutive and identical events during human gait, which is
divided in two main phases: stance and swing [3]. On the stance phase, an equal and
opposite direction ground reaction force (GRF) changes in direction and magnitude as
the body center of mass moves forward in a gait cycle with the foot in contact with the
ground [4]. Since the GRF changes with the stance phase of the gait cycle [5], it is
possible to evaluate the stance phase through the GRF assessment. Besides the pos-
sibility of distinguishing the normal from pathological gait, the stance phase evaluation
can also be applied in the control strategies of exoskeletons [6]. Although force plat-
forms and electronic insoles have been applied in the plantar pressure monitoring,
studies show that they may lack in portability or reliability [7].
Polymer optical fiber (POFs) sensors present electromagnetic field immunity,
multiplexing capabilities, fracture toughness, high strain limits and impact resistance
[8]. Such advantages enable their application in the instrumentation of flexible struc-
tures such as an insole. For these reasons, this paper proposes the application of POF
sensors for plantar pressure and GRF monitoring with two different optical fiber sen-
sors approaches. One with intensity variation-based sensors, which is a low-cost
solution, but lack in multiplexing capabilities. The other approach is based on polymer
optical fiber Bragg gratings (POFBGs) that presents great multiplexing capabilities, but
with higher cost when compared with intensity variation sensors.
Figure 1 presents both insoles applied in this work, where Fig. 1(a) shows the intensity
variation-based insole for GRF monitoring. In this approach, the optical power atten-
uation is analyzed with respect to the GRF. The input of the sensor is connected to light
emitting diode (LED) IF-E97 (Industrial Fiber Optics, USA) and the output is con-
nected to a photodiode IF-D91 (Industrial Fiber Optics, USA). The regions referred as
Sensors 1-4 are the regions in which the fiber has a lateral section with controlled depth
and length, creating a region with higher sensitivity [9]. Therefore, during the gait,
there is a resultant optical power attenuation in all 4 sensors, which is proportional to
the GRF.
Figure 1(b) shows the POFBG-based insole, where the sensing element, i.e., the
fiber Bragg grating is a periodic modulation of the fiber refractive index, which, in this
case, is accomplished with direct incidence of a femtosecond laser pulse with high
energy density [10]. When the fiber is illuminated in a certain wavelength range, this
refractive index modulation creates a peak in the POF reflection spectrum that is related
to the period of the refractive index modulation, which is sensitive to mechanical
deformations. Thus, it is possible to create different reflection peaks in the same fiber
Fig. 1. POF-based instrumented insoles employed. (a) Intensity variation-based and

(b) POFBG-based, inset presents the reflection spectrum.
with the variation of the refractive index modulation period. In this case, there are 5
peaks that present an independent shift in their wavelengths with the plantar pressure
variation. Therefore, there are 5 plantar pressure measurement points.
The root mean squared errors (RMSE) obtained for each POF insole are compared.
The RMSE of the POFBG-based insole is the mean value between the 5 sensors, which
is 45.17 kPa, where the lowest RMSE was obtained in POFBG 5 (29.22 kPa).
Regarding the intensity variation-based POF insole, the RMSE between the POF sensor
and the force platform is 5.30 N. It is worth to mention that the units between the two
POF insoles are different, since the POFBG-based measures plantar pressure at each
point and the intensity variation-based measures the GRF. After the insoles charac-
terization, four gait cycles are made with each POF sensor. The results are presented in
Fig. 2(a) and (b) for the POFBG and intensity variation-based, respectively. The pre-
sented results show high repeatability of both sensor approaches. In addition, both
sensors presented the M-shape pattern of the GRF during the stance phase and the
sensors presented a resolution of about 10 N. However, the proposed sensors present
the limitations acquiring the force at each pressure point for the case of the intensity
variation-based insole (Fig. 2(b)) and, for the POFBG-based sensor, the hardware for
the sensor interrogation is less compact, which reduces its portability.
Polymer Optical Fiber Sensors Approaches for Insole Instrumentation 473
Fig. 2. Results of the POF instrumented insoles proposed (a) POFBG-based insole. (b) Intensity
variation-based insole.
4 Conclusion
The manuscript introduces the application of POF sensors in insole instrumentation is

presented with two different approaches: intensity variation and POFBGs. Each
approach has its own advantages and drawbacks. While the intensity variation
approach presents low cost and lacks in multiplexing capabilities, the POFBG approach
presents great multiplexing capabilities, but with much higher cost on the development
of the interrogation optoelectronics. Nevertheless, both approaches presented low rel-
ative errors (45.17 kPa and 5.30 N for the POFBGs and intensity variation sensors,
respectively) and high repeatability. Therefore, there is a tradeoff between low cost and
multiplexing capabilities that must be addressed in each application.
References
1. Hadi, A., et al.: Foot plantar pressure measurement system: a review. Sensors 12(7), 9884–
9912 (2012)
2. Morag, E., Cavanagh, P.R.: Structural and functional predictors of regional peak pressures
under the foot during walking. J. Biomech. 32(4), 359–370 (1999)
3. Taborri, J., Palermo, E., Rossi, S., Cappa, P.: Gait partitioning methods: a systematic review.
Sensors 16(1), 66 (2016)
4. Abboud, R.J.: Relevant foot biomechanics. Current Orthop. 16(3), 165–179 (2002)
5. Kirtley, C.: Clinical Gait Analysis: Theory and Practice. Elsevier, Philadelphia (2006)
6. Villa-Parra, A., et al.: Knee impedance modulation to control an active orthosis using insole
sensors. sensors 17(12), 2751 (2017)
7. Leal-Junior, A.G., et al.: Polymer optical fiber for in-shoe monitoring of ground reaction
forces during the gait. IEEE Sens. J. 18(6), 1558–1748 (2018)
8. Peters, K.: Polymer optical fiber sensors—a review. Smart Mater. Struct. 20(1), 13002
(2011)
9. Leal-Junior, A.G., Frizera, A., José Pontes, M.: Sensitive zone parameters and curvature
radius evaluation for polymer optical fiber curvature sensors. Opt. Laser Technol. 100, 272–
281 (2018)
10. Vilarinho, D., et al.: POFBG-embedded cork insole for plantar pressure monitoring. Sensors
17(12), 2924 (2017)
Pushing the Limits: A Novel Tape Spring
Pushing Mechanism to be Used
in a Hand Orthosis
Claudia J. W. Haarman1,2(B) , Edsko E. G. Hekman1 , Hans S. Rietman1,3 ,

and Herman van der Kooij1
1
Department of Biomechanical Engineering, University of Twente, 7522 Enschede,
NB, The Netherlands
c.j.w.haarman@utwente.nl
2
Hankamp Rehab B.V., Enschede, The Netherlands
3
Roessingh Research and Development, Enschede, The Netherlands
Abstract. A device that supports hand function may significantly

improve the quality of life of patients with muscular weakness. Since
tight constraints such as size and weight are placed upon the device, com-
plexity of the hardware and functional performance should be carefully
balanced. A novel force transmission mechanism based on tape springs
is presented for use in a hand orthosis. The actuator force is transmitted
to the finger by a system consisting of a tape spring, two slider blocks
and an end stop per finger. The tape spring allows for bending in one
direction, and resists bending in the other direction. A prototype with
the new mechanism is constructed. The low profile together with the
ability to transmit large forces makes this mechanism suitable for hand
orthoses.
1 Introduction
Support of the fingers during activities of daily living (ADL) may be necessary as
a result of muscle weakness caused by neuromuscular diseases such as Duchenne
muscular dystrophy. A hand orthosis that supports grasping movements while
being as unobtrusive and lightweight as possible may significantly improve the
quality of life of individuals with weakness of the finger flexors and extensors.
Commonly used force transmission mechanisms in hand orthoses consist of
underactuated glove-like designs using cables with remote actuation [1,2], soft
pneumatics [3,4] or a three-layered sliding leaf spring [5,6]. Their advantages
compared to traditional exoskeletons with many rigid links are numerous: low
profile, increased compliance (beneficial for human-robot interaction safety), a
larger number of DoF (although not actively controlled) and lower material and
This work is part of the research programme Symbionics with project number 13525,
which is (partly) financed by the Netherlands Organisation for Scientific Research
(NWO).
https://doi.org/10.1007/978-3-030-01887-0_92
476 C. J. W. Haarman et al.
manufacturing costs. However, the amount of force that can be applied is often
limited, fingertip sensation is impeded or the range of motion is restricted.
In this paper a novel force transmission mechanism for a hand orthosis is
presented to actively flex the finger joints. The underactuated mechanism uses
tape springs to transmit forces to the finger.
Fig. 1. Geometry of an undeformed tape spring
2 Method
2.1 Tape Springs
Tape springs (Fig. 1A) are thin metallic strips of uniform thickness t with a
transverse curvature of radius R, spanning an angle α. The transverse curvature
gives the tape spring a longitudinal structural stiffness. Bending of the tape
spring imposes a longitudinal curvature.
Two types of bending are distinguished:
• Opposite-sense bending (bending moment M > 0, as shown in Fig. 1B) occurs

if the longitudinal and transverse curvatures are in the opposite direction.
• Equal-sense bending (M < 0) produces longitudinal and transverse curvatures
that have the same direction.
In Fig. 2 the moment-angle relationship of a tape spring is shown. It can be

max
seen that the peak opposite-sense bending moment (M+ ) is larger than the
max
peak equal-sense bending moment (M− ). This feature will be deployed in the
force transmission mechanism of the hand orthosis.
2.2 Force Transmission Conceptual Design

The tape spring, located on the dorsal side of the hand, is pushed distally with
actuation force Fact at a height hm and hp above the metacarpophalangeal
(MCP) respectively proximal interphalangeal (PIP) joint, see also Fig. 3A. This
will create a flexion moment around the finger joints. Buckling is expected at
locations X and Y of Fig. 3A in a regular (flat) leaf spring. However, the large
max
peak opposite-sense bending moment (M+ ) will prevent buckling of the tape
spring.
Pushing the Limits: A Novel Tape Spring Pushing Mechanism 477
Fig. 2. Moment-angle relationship of tape spring under opposite-sense (bending angle

θ > 0) and equal-sense (θ < 0) bending, based on deployment dynamics of tape springs
described in [7].
Fig. 3. (A) Force transmission concept showing the finger and tape spring mechanism.
The tape spring (in red) is pushed with actuation force Fact . X and Y indicate loca-
tions where buckling is expected. (B) Cross-sectional view of undeformed tape spring.
(C) Cross-sectional view of flattened tape spring.
2.3 Flattening of the Tape Spring
During flexion movements equal-sense bending is deployed, which is difficult to

control as the moment is non-linear when moving from a neutral position to a
flexed position. By flattening the tape spring roughly at the location of the MCP
and PIP joint, the peaks of the bending moment are removed locally. Figure 3B
shows the undeformed cross section of the tape spring. In Fig. 3C the tape spring
is flattened by applying forces at three points. The tape spring is forced to follow
one direction of bending which is equal-sense bending, because the tape spring
is attached to a jointed structure (the finger).
478 C. J. W. Haarman et al.
3 Results
Figure 4 shows the resulting prototype of the hand orthosis. The underactuated
mechanism is able to flex two fingers using the novel tape spring mechanism (1)
operated by remotely placed actuators connected through Bowden cables (2).
The cables are connected to the proximal end of the tape spring and routed in
the direction of the fingers. There the cable is rotated 180◦ to the distal end of
the hand slider block. This will cause a pulling force on the tape spring causing
the tape spring to slide outward. The force transmission mechanism is mounted
on the dorsal side of the hand, so that it does not obstruct object manipula-
tion while enabling finger tip sensation. A tape spring (width = 12.2 mm, R
= 13.6 mm, α = 50◦ ) is chosen for application in the hand orthosis. With this
configuration sufficient flexion moments around the finger joints can be created
to allow execution of ADL tasks. Two sliders (3) and (4), connected to the hand
and the proximal phalanx, serve as a guide for the tape spring. A constant force
spring in the hand slider pulls the tape spring back during finger extension.
The tape spring is rigidly connected to the end block (5) which is attached to
the middle phalanx. The tape spring is flattened above the finger joints (6). The
extrusion of the end block prevents hyperextension of the distal interphalangeal
joint (DIP).
Fig. 4. Prototype of the hand orthosis. (1) Tape spring (2) Bowden cable (3) Hand
slider (4) Finger slider (5) End block (6) Tape flattener
4 Conclusion
The design shows promising results for application of the new tape spring mech-
anism in a hand orthosis for patients with muscular weakness. The low profile in
combination with a high force output makes this design capable of transmitting
high forces to the finger. The transmitted forces are only limited by the onset of
Pushing the Limits: A Novel Tape Spring Pushing Mechanism 479
buckling. Further research is required to find the optimal tape spring configura-
tion to limit buckling and prevent plastic deformation. Future work will include
technical tests to measure the grip force, range of motion, and usability testing.
References
1. Nilsson, M., Ingvast, J., Wikander, J., Von Holst, H.: The soft extra muscle system
for improving the grasping capability in neurological rehabilitation. In: IEEE-EMBS
Conference on Biomedical Engineering and Sciences, IECBES 2012, pp. 412–417,
December 2012
2. In, H., Lee, H., Jeong, U., Kang, B.B., Cho, K.-J.: Feasibility study of a slack
enabling actuator for actuating tendon-driven soft wearable robot without preten-
sion. In: ICRA, pp. 1229–1234 (2015)
3. Polygerinos, P., Wang, Z., Galloway, K.C., Wood, R.J., Walsh, C.J.: Soft robotic
glove for combined assistance and at-home rehabilitation. Robot. Auton. Syst. 73,
135–143 (2015)
4. Yap, H.K., Lim, J.H., Nasrallah, F., Goh, J.C.H., Yeow, R.C.H.: A soft exoskeleton
for hand assistive and rehabilitation application using pneumatic actuators with
variable stiffness. In: Proceedings - IEEE International Conference on Robotics and
Automation, pp. 4967–4972, June 2015
5. Arata, J., Ohmoto, K., Gassert, R., Lambercy, O., Fujimoto, H., Wada, I.: A new
hand exoskeleton device for rehabilitation using a three-layered sliding spring mech-
anism. In : Proceedings - IEEE International Conference on Robotics and Automa-
tion, pp. 3902–3907 (2013)
6. Nycz, C.J., Butzer, T., Lambercy, O., Arata, J., Fischer, G.S., Gassert, R.: Design
and characterization of a lightweight and fully portable remote actuation system for
use with a hand exoskeleton. IEEE Robot. Autom. Lett. 1(2), 976–983 (2016)
7. Seffen, K.A., Pellegrino, S.: Deployment dynamics of tape springs. Proc. R. Soc.
A: Math., Phys. Eng. Sci. 455(1983), 1003–1048 (1999)
Design and Preliminary Validation
of a Smart Personal Flotation Device
Julian Fraize, Mirjam Furth, and Damiano Zanotto(B)
Stevens Institute of Technology, Hoboken, NJ 07030, USA

dzanotto@stevens.edu
Abstract. This work introduces the SmartPFD, a Personal Flotation

Device (PFD)/Life vest featuring independently inflatable air compart-
ments. Unlike off-the-shelf inflatable PFDs, the SmartPFD is designed to
modulate the inflation of its compartments based on the wearer’s orien-
tation and depth, with the aim of optimizing the device’s righting ability.
Preliminary results suggest that by properly sequencing the activation
of the compartments, the time required to turn the wearer face up can
be reduced.
1 Introduction
Drowning is the third leading cause of unintentional injury death worldwide [1].
In 2016 alone, the US Coast Guard (USCG) reported 4463 recreational boating
related accidents, marking a 11% increase from 2015 data [2]. These accidents
led to 701 deaths, 73% of which were due to drowning.
PFDs can considerably prolong life expectancy of a person that has gone
overboard. The International Standards Organization provides guidance for the
design on PFDs through the ISO Standard 12402 [3]. This standard classifies
PFDs into lifejackets, which help keep the user face up in the water, and buoy-
ancy aids, which require the user to move themselves in order to float face up.
In the United States, the USCG is responsible for testing and approving PFDs.
The USCG classifies PFDs into five different types, with Type I having the high-
est standards for flotation and righting of the user. Type I PFDs are geared for
situations where rescue is not immediate, but they are bulky and uncomfortable,
and therefore they are not often worn by recreational users [4].
Inflatable PFDs feature a slimmer profile that makes them more comfortable.
The goal of this study is to develop and test an improved inflatable Type I PFD,
which uses embedded sensors and control to modulate the inflation of multiple
compartments with the aim of turning unconscious wearers face-up in the water
in the shortest time possible.
This abstract presents the design of the first prototype of the SmartPFD,
which features independent anterior (A) and posterior (P) compartments, along
with preliminary results demonstrating the effects of staggered inflation sequenc-
ing on the righting ability of the device.

https://doi.org/10.1007/978-3-030-01887-0_93
Design and Preliminary Validation of a Smart Personal Flotation Device 481
Fig. 1. The SmartPFD features anterior/posterior inflatable compartments and

embedded electronics to modulate the inflation of the compartments in response to
the wearer’s depth and orientation.
2.1 Design
The SmartPFD, shown in Fig. 1, consists of a horseshoe vest with two sepa-
rate inflatable A/P compartments attached to a harness. The battery pack and
electronics are housed in a waterproof box attached to the user’s waist. Both
of the two inlets of the box are connected to a 16g CO2 canister through a
manually adjustable pressure regulator, which also integrates a pre-puncturing
mechanism.
The system is powered by a Li-Po battery pack through step-up and step-
down voltage regulators that serve the valves (24 V) and the logic circuit (5 V),
respectively. The battery pack is oversized (3000 mAh, 11.4 V) for testing pur-
poses. A 32-Bit ARM Cortex-M4 microcontroller samples data at 300 Hz from an
analog pressure sensor (TE connectivity) and from a 9-degree-of-freedom inertial
measurement unit (Yost Labs), which track the wearer’s depth and orientation,
respectively. The microcontroller also triggers two fast-response 3/2 NC pneu-
matic valves (Festo MHE series). The outlet of each valve is connected to one
inflatable compartment through polyurethane tubing, while the NC port is con-
nected to one of the pressure regulators. The NO ports are connected together to
allow for air pressure distribution between the two compartments after inflation
is complete.
Data is logged to an on-board mini-SD card and also sent to a host computer
using a Wi-Fi module. A graphical user interface (GUI) running on the host
computer allows the experimenter to visualize data in real time and reset the
device remotely.
482 J. Fraize et al.
2.2 Experimental Protocol and Data Analysis
Multiple tests were conducted in a 1.97 m deep water tank to demonstrate the
effects of staggered inflation on the righting ability of the device. We hypoth-
esized that staggered inflation would create a larger moment arm about the
longitudinal (L) axis of the mannequin, thus resulting in a faster rolling of the
body compared to the simultaneous activation.
Before testing, the vest was fitted to a rigid mannequin used for life guard
training, and the electronics box was secured to the posteroinferior part of the
same mannequin (Fig. 1). The experiment was conducted by ballasting the man-
nequin to sink to the bottom of the tank with a face-down orientation. With
the mannequin resting at the bottom of the tank, the staggering of inflation was
accomplished by introducing a time delay Δt between the activations of the A
and P valves. We tested 6 values of Δt (0 to 5 s, in one-second increments), and
conducted 6 repetitions for each Δt.
Time to Surface (TTS) and Time to Inclination (TTI) were selected as the
evaluation metrics. The former indicates the time required for the mannequin’s
mouth to reach the surface, starting from the time of activation of the A valve.
The latter is the time required for the mannequin’s L axis to achieve an inclina-
tion of at least −20◦ from the vertical axis. This metric was assigned according
to the International Life-Saving Appliance guidelines [5], however, the require-
ment was made more restrictive by imposing a minimum angle of 15◦ between
the mannequin’s anteroposterior (AP) axis and the surface.
Kruskal-Wallis H tests were applied to TTS and TTI data to check for sig-
nificant (α = .05) effects of Δt on these metrics. If a significant effect was found,
post-hoc analysis was conducted using Mann-Whitney U tests, with Bonferroni
correction for multiple comparisons.
TTS increased significantly compared to the baseline case (Δt = 0) for all the
tested values of Δt, Fig. 2(a). Indeed, ascent rates became smaller and smaller for
increasing Δt, due to a reduction in the buoyancy available at each time instant,
Fig. 2(c). On the other hand, TTI was significantly reduced when Δt = 3 s and
Δt = 4 s, Fig. 2(b). Inspection of the orientation trajectories, shown in Fig. 2(d),
suggests that the mannequin primarily rotated about its mediolateral (ML) axis
when Δt = (0, 1) s. Conversely, for larger Δts, an initial rotation about the L axis
was followed by a rotation about the ML axis. By appropriately delaying the ML
rotation in favor of the L rotation, the target inclination could be reached more
quickly. Further increasing Δt to 5 s, however, did not reduce TTI, possibly a
consequence of the reduced buoyancy when the mannequin was still underwater.
Thus, for these testing conditions, a delay between 3 s and 4 s might be optimal.
Design and Preliminary Validation of a Smart Personal Flotation Device 483
(a) (b)
500 UP
0
Depth [mm]
-500
-1000
75
T= 0s
T= 1s 65
-1500 T= 2s
55
T= 3s
T= 4s 45
T= 5s
-2000 30
0 5 10 0
t [s] DOWN
(c) (d)
Fig. 2. TTS (a) and TTI (b) for different Δt. Error bars indicate ±1SE; * indicates
a significant difference (p = .01) w.r.t. the baseline. Average trajectories for depth
(c) and orientation (d). In (c), t = 0 indicates the time instant when the A valve was
activated. The colored region in (d) indicates the target inclination, whereas each line
represents the projection of the mannequin’s AP axis onto a vertical plane passing
through its L axis.
4 Conclusion
This work presented the design and preliminary validation of a novel PFD instru-
mented with sensors and two inflatable compartments. The novelty of the pro-
posed device lies in the possibility to independently control the inflation of each
compartment in response to the wearer’s depth and orientation, to optimize the
righting ability of the PFD. Results confirmed that simultaneous activation of
the compartments, which is common in all existing inflatable PFDs, might not
be the best strategy when the goal is to turn a wearer face-up in the shortest time
possible, and that staggered inflation can positively affect the righting ability of
the PFD.
484 J. Fraize et al.
Future work will include the development of numerical models to study the
best activation strategy in response to the user’s states. This strategy will then
be experimentally validated on a new version of the SmartPFD, which features
multiple inflatable air compartments.
Acknowledgments. The authors would like to acknowledge the assistance of all

the students involved in this project, especially, Kevin Raleigh, Juan Sanchez,
Stephanie Mallon, and Luke Pacchiana.
References
1. World Health Organization: Drowning, January 2018. http://www.who.int/en/
news-room/fact-sheets/detail/drowning
2. 2016 recreational boating statistics: U.S. Department of Homeland Security, U.S.
Coast Guard, Office of Auxiliary and Boating Safety, Washington, DC, COMDT-
PUB P16754.30, May 2017
3. Personal flotation devices - part 5: Buoyancy aids (level 50) - safety requirements.
International Organization for Standardization, Standard, September 2006
4. Quistberg, D.A., Quan, L., Ebel, B.E., Bennett, E.E., Mueller, B.A.: Barriers to life
jacket use among adult recreational boaters. Inj. Prev. 20(4), 244–250 (2014)
5. International Maritime Organization: Life-saving appliances (inc. LSA Code),
2017th ed. IMO Publishing, March 2017
Introducing Compound Planetary Gears
(C-PGTs): A Compact Way to Achieve
High Gear Ratios for Wearable Robots
Stein Crispel(B) , Pablo López Garcı́a, Tom Verstraten, Bryan Convens,

Elias Saerens, Bram Vanderborght, and Dirk Lefeber
Robotics & Multibody Mechanics Research Group (R&MM), Faculty of Mechanical

Engineering, Vrije Universiteit Brussel and Flanders Make,
Pleinlaan 2, 1050 Elsene, Belgium
Stein.crispel@vub.be
Abstract. In the field of wearable robots, high power density and highly
efficient actuators are required to handle the high-power motion without
becoming heavy and bulky and hence hamper their mobility. Typically,
electrical motors are used in combination with high gear ratio gearheads
or lever arms in order to achieve the required torques. These gearboxes
consist mainly out of several stages of simple Planetary Gear Trains
(PGTs). However, this approach leads to big and heavy gearboxes when
high torque is needed. An alternative, more compact, design to obtain
the required torque increase can be achieved using Compound Planetary
Gears (C-PGTs). It is shown that the latter mechanism can obtain gear
ratios up to 1:600 while withstanding an output torque of 100 Nm.
1 Introduction
Research has shown that during a gait cycle of a healthy person, the ankle needs
to inject up to five time more energy than it can recuperate. Moreover, it pro-
duces as much as 150 Nm peak torque for a 86 kg person [1], which requires high
torque motors when they are used as direct drives. Although electrical motors
are the most popular type of actuators, they tend to become big and heavy for
high torques. Therefore, most drivetrains combine a fast running electrical motor
with a transmission mechanism to trade speed for torque. In the following para-
graphs the most widely used gearboxes are discussed and an alternative design
is introduced and discussed: the C-PGT.
Bryan Convens and Elias Saerens are both SB PhD Fellows at the Research Foun-
dation Flanders-Fonds voor Wetenschappelijk Onderzoek (FWO). Part of this work
was funded by the European Commission ERC Starting grant SPEAR (no. 337596)
https://doi.org/10.1007/978-3-030-01887-0_94
486 S. Crispel et al.
2 Common Gearboxes for Wearable Robots

2.1 Series Configuration of PGTs
Probably the most used option to increase the torque of a motor is the imple-
mentation of one or more successive PGT stages, see Fig. 1. PGTs are preferred
over a conventional gear train since they are more compact for the same reduc-
tion. Indeed, for the same size, the gear ratio, i, of a PGT is more than twice
as high as the one of the latter. The gear ratio of a PGT, for which the ring is
grounded, is given by:
ωs Zp
i= =2 +1 (1)
ωc Zs
Where ωc and ωs represent the speeds of the carrier (output) and the sun (input),
respectively, and Zp , Zs the number of teeth of, respectively, the planet- and sun
gear.
Fig. 1. Two stages of PGTs where the carrier of the first stage is connected to the sun
of the second stage
The reduction that can be obtained for a single stage PGT is limited to 1:10,
due to size restrictions of the sun gear [2]. Larger gear ratios can be achieved by
combining multiple stages.
The total efficiency of this type of configuration is consequently the product
of the efficiencies of all individual stages, which can be as high as 97%. Although
the transmission ratio increases faster than for a conventional gear pair, the
amount of stages necessary to obtain a transmission ratio suitable for the ankle
actuator, such as in [3], is three or four. This makes the overall system big and
heavy, which conflicts with the initial idea of creating a compact and lightweight
actuator.
2.2 Harmonic Drive

This type of speed reducer is often used in industrial robotic applications due
to its high transmission ratio, which is limited to 1:320 [4], and its near to zero
backlash. However, the input speed should be restricted to around 2000rpm,
which makes the use of compact, light and high speed electric motors impossible.
Introducing Compound Planetary Gears (C-PGTs) 487
The three major components of a Harmonic Drive are the wavegenerator,

flexspline and the circular spline. The high gear ratio is obtained due to the
small difference in the number of teeth between these two splines.
Although Harmonic Drives have a very high efficiency at nominal conditions,
it drops rapidly with increasing input speed, decreasing temperature and -output
torque [5].
3 Compound Planetary Gear Train

Due to the versatility of planetary gears, other configurations can be found where
very high gear ratios can be obtained in a very compact way. For example, by
combining the carrier and planets of two stages of PGTs, instead of connecting
the carrier to the ring gear as in Sect. 2.1, gear ratios up to 1:1000 and higher
can be obtained. Three possible arrangements are shown in Fig. 2 [6].
These types of C-PGTs have found their use in aerospace drives, but are
not investigated for wearable robotic devices yet, to the best of the authors’
knowledge.
Fig. 2. Three possible C-PGT configurations: (a.) Sun and output ring -, (b.) sun and
fixed ring -, (c.) both the output and fixed ring are meshing with the same planet
gear [6].
A very interesting layout is given in Fig. 2-b. This configuration is discussed

below.
3.1 Gear Ratio

The gear ratio, i, for the configuration depicted in Fig. 2-b. is given by:
ωs 2rp1 rr2
i= = (2)
ω r2 rs (rp1 − rp2 )
With ωs and ωr2 the speeds of the sun- and second ring gear, respectively. The
working pitch radii of respectively the sun, the first and second planet and the
ring gear are given by rs , rp1 , rp2 and rr2 . It can be seen that when the radii of
the two planets are chosen close together, the gear ratio can become very high.
488 S. Crispel et al.
3.2 Efficiency
The main drawback of this type of gearbox is the possible re-circulation of virtual
power in the system. Due to this phenomenon the total efficiency tends to drop
when the gear ratio increases [7]. In order to increase the efficiency as much as
possible, the meshing losses of the individual teeth contacts should be minimal.
To do so, profile shift and addendum modification, among others, can be used
to minimise the sliding area of the teeth.
The efficiency of this arrangement (Fig. 2-b.) is calculated using the method
given in [7] and given by:
Zr2 Zr 1
Zs + ηs ηrk1 Zr1 Zp2 − Zp1
η= · Z
(3)
Zs + Zr1 Zr2
− ηrk1 ηrk2 Zpr1
Zp2 1
Where Zx , with x = {s, p1 , p2 , r1 , r2 }, represents the number of teeth of respec-

tively the sun, the first and second planet, and the first and second ring. The
meshing efficiency between the sun and first planet is characterised by ηs ,
between the first planet and first ring by ηr1 and between the second planet
and second ring by ηr2 . Depending on the sign of expression (4) the exponent k
is plus or minus one.

1 if Zr1 Zp2 − Zr2 Zp1 < 0
k= (4)
−1 if Zr1 Zp2 − Zr2 Zp1 > 0
3.3 Size
The state of the art reference in size for high gear ratio transmissions is the
very compact Harmonic drive. Stress simulations for a 100 Nm output, Fig. 3,
show that C-PGTs can achieve gear ratios as high as 1:600 while remaining very
compact.
Fig. 3. Gear ratio which can be achieved in function of the exterior diameter for an
output torque of 100 Nm (when η = 100%). The exterior diameter is given by the
maximum diameter of the two rings.
Introducing Compound Planetary Gears (C-PGTs) 489

In this abstract a transmission system is proposed that has the potential to
compete with the Harmonic Drive in both size and gear ratio. The authors believe
that the presented layout can be integrated in multiple robotic applications and
increase the mobility of wearable robots.
In view of applying C-PGTs to wearable robots, a more extended efficiency
model will be developed, including different techniques to minimise the power
losses. Moreover, other C-PGT configurations will be investigated and compared
in order to investigate the effect of virtual power re-circulation, their dynamic
behaviour and compactness.
References
1. Winter, D.A.: Energy generation and absorption at the ankle and knee during fast,
natural, and slow cadences. Clin. Orthop. Relat. Res. 175, 147–154 (1983)
2. Pawar, P.V., Kulkarni, P.R.: Design of two stage planetary gear train for high reduc-
tion ratio. Int. J. Res. Eng. Technol. 4(06), 150–157 (2015)
3. Verstraten, T., Geeroms, J., Mathijssen, G., Convens, B., Vanderborght, B., Lefeber,
D.: Optimizing the power and energy consumption of powered prosthetic ankles with
series and parallel elasticity. Mech. Mach. Theory 116, 419–432 (2017)
4. Bailak, G.V., Rubinger, B., Jang, M., Dawson, F.: Advanced robotics mechatronics
system: emerging technologies for interplanetary robotics. In: Canadian Conference
on Electrical and Computer Engineering, vol. 4, pp. 2025–2028. IEEE, May 2004
5. Tuttle, T.D.: Understanding and modeling the behavior of a harmonic drive gear
transmission (No. AI-TR-1365). Massachusetts Inst of Tech Cambridge Artificial
Intelligence Lab (1992)
6. Kapelevich, A., AKGears, L.L.C.: High gear ratio epicyclic drives analysis. Ratio 3,
10 (2013)
7. Salgado, D.R., Del Castillo, J.M.: Analysis of the transmission ratio and efficiency
ranges of the four-, five-, and six-link planetary gear trains. Mech. Mach. Theory
73, 218–243 (2014)
Model-Based Approach in Developing
a Hand Exoskeleton for Children:
A Preliminary Study
Matteo Bianchi1(B) , Nicola Secciani1 , Alessandro Ridolfi1 , Federica Vannetti2 ,

and Guido Pasquini2
1
Department of Industrial Engineering (DIEF), University of Florence,
Florence, Italy
matteo.bianchi@unifi.it
2
Don Carlo Gnocchi Foundation, Florence, Italy
Abstract. Developing wearable robotic systems for children’s hand

rehabilitation highlights several issues during the design phase due to the
difficult interactions with the patients and the high adaptability these
devices require to fit fingers always growing. In this research work, the
authors propose a model-based approach which exploits only a photo-
graph of the hand to automatically generate a tailor-made device capable
of replicating hand trajectories. A real device has been developed and
tested on a patient after three growing steps in order to assess its kine-
matic compliance with fingers.
1 Introduction
A wide variety of pathologies affects upper and lower limbs, deeply undermin-
ing subjects autonomy. In this context, robotic rehabilitation systems, capable
of delivering effective high-dosage therapies, are rapidly growing. A thorough
review of such devices can be find in [1]. The wearability of such devices plays a
key role as they are conceived to be worn for a long time. Among these systems,
hand exoskeletons are deeply affected by wearability issues because of lack of
physical room and the required compatibility with different hands morphology.
Designing a fully wearable device is then particularly challenging when it has to
be worn by a child and only few studies have hence investigated such assistive
systems in children with Celebral Palsy (CP) [2]. The difficulty in interacting
with the specific users may impede an accurate analysis and their fast grow steps
demand an always different but comfortable device.
The authors have developed an easy but effective procedure, which, basing
on a model of the patient’s hand, allows to rapidly design a taylor-made device
(Fig. 1) specific for the user’s hand. The overall procedure is visually summarized
through the flowchart of Fig. 2.
This work has been supported by the HOLD Project, funded by the University of
Florence. The authors would also like to thank the Don Carlo Gnocchi Foundation
for the help.
https://doi.org/10.1007/978-3-030-01887-0_95
Model-Based Approach in Developing a Hand Exoskeleton for Children 491
Fig. 1. The hand exoskeleton of the University of Florence: overall design (on the left)
and testing phase on a child (on the right).
2 Hand Model
Starting from the hand overall dimensions, a kinematic model of a child’s hand
has been developed in order to study hand movements without facing the user
although. This procedure allowed to generate the hand trajectories the exoskele-
ton has to replicate and to develop a hand 3D model which has been exploited
to virtually test the ergonomics.
Hand Kinematic Model

In the presented kinematic model long fingers present 4 DOFs each: one
flexion-extension movement per joint, and one MetaCarpoPhalangeal (MCP)
ab/adduction. In the model though, ab/adduction has not been considered, as
the exoskeleton presents a passive DOF, which pursues this movement, as dis-
cussed in Sect. 3. Thus, each long finger behaves as a three-link planar arm
manipulator, presenting three cylindrical joints with parallel axes and moving
on the plane perpendicular to the aforementioned axes where the finger longitu-
dinal dimension lies.
The procedure followed to obtain the complete hand movement uses the
Denavit-Hartenberg (D-H) convention to identify joints positions and orienta-
tions with respect to a wrist coordinate system. D-H parameters have been
defined by exploiting ImageJ open source image processing software which
allowed, by a photograph of the hand, to calculate the phalanges dimensions.
Once the positions of the fingers joints in the extension configuration have
been calculated, flexion trajectories have been obtained by moving the phalanges
from the extension to the complete closure configuration. The results obtained
with this technique have been processed in MATLAB R environment calculating
all the hand joints trajectories.
Hand CAD Model

A 3D model of the hand has been developed for hand-exoskeleton coupling tests.
Fingers have been modeled as chains of rigid bodies connected by rotational
joints (the kind of joints discussed in Sect. 2 which provide the movement); finger
492 M. Bianchi et al.
phalanges have then embodied the links modeled by cylinders (with the same
radii of the real fingers).
3 Exoskeleton
This version of the exoskeleton originates from the one presented in [3]. Even
if that device represented a fully portable, wearable and highly customizable
system that could be used both as an assistive hand exoskeleton and as a reha-
bilitative one, this new prototype for children has required a specific re-design,
aiming to comply with strict constraints of size and weight.
Fig. 2. Model-based design procedure flowchart.
Exoskeleton Kinematic Model
The accurate development of a novel mechanism, characterized by a single DOF

per finger, allowed to precisely and comfortably reproduce the complex hand
kinematics without being forced to use an equally complex robotic device. This
assumption proved to be fundamental in designing a device as efficient as simple,
conceived to be employed on a child. A detailed kinematic synthesis of the single-
DOF finger mechanism is given in [3].
An optimization-based strategy has then been exploited to adapt the
exoskeleton to the child fingers kinematics: a Nelder-Mead based optimization
algorithm has been used [4] achieving a straightforward adaptability to the user.
Taking the hand kinematics (calculated by the aforementioned hand model) and
the one of the mechanism as inputs, the implemented algorithm provides the
customized geometry specific for the patient.
Model-Based Approach in Developing a Hand Exoskeleton for Children 493
Exoskeleton CAD Model and Mechanical Design

All the mechanical parts have been 3D-printed in ABS thermoplastic polymer.
This choice allowed to change the mechanism when necessary. In fact, due to
the fast growth of the child, the proposed procedure needs to be performed
periodically in order to generate new mechanisms properly fitting the new fingers
sizes.
Then, an additional passive DOF was added upstream of each finger mech-
anism to allow for the natural ab/adduction. The introduction of the rotational
passive joint let to act only on the finger flex/extension plane, passively following
the finger ab/adduction gestures.
The system is actuated by a single servomotor and a specific cable driven
transmission system has been developed to open all the four long fingers together
at the same time. Closing gestures are instead controlled by the user himself,
while the exoskeleton only manages range of movements and closure velocity.
Different mechanisms velocities are obtained thanks to different pulleys diame-
ters, which are calculated depending on users’ fingers dimensions.
4 Results
The kinematic interaction between the hand and the exoskeleton has been eval-
uated by the implementation of virtual ergonomic tests. In particular, exploit-
ing the SolidWorks Motion Simulation tools, the actual (kinematic) interaction
between the 1-DOF mechanism and the finger has been assessed, avoiding inter-
penetrations and verifying mutual coupling during the whole range of motion.
Once the effectiveness of the device has been tested, a real prototype has been
built to carry out the wearability assessment (Fig. 1). The kinematic impact
on a single user has been evaluate after three consecutive growing step. The
whole described procedure has been exploited each time, generating different
fingers mechanisms geometries. Qualitative tests have confirmed a satisfying
adaptability of the device to the child’s hand both in fitting size and in following
finger trajectories. A physiotherapist of the Don Carlo Gnocchi Foundation has
been involved in this phase of the study to evaluate if, when the exoskeleton is
worn by the child, any hindrances during flexion/extension movements occur.
5 Conclusion and Further Developments

Designing wearable device for children involves several technical difficulties low-
ering the effective outcome of these systems themselves. In this work, a model-
based approach allowing to design a device with few measures on the user’s hand
has been presented. The exploitation of a simple but rapid strategy led to over-
come difficulties in adapting the device during the fast grow steps of the user
yielding a complete patient-tailored robotic aid.
Although the presented work represents a preliminary study, it opens sev-
eral ways to further developments. A specific control and actuation strategies
494 M. Bianchi et al.
have to be developed taking into account the particular application on a child.

In addition, the device testing phase arose the necessity to reduce the overall
dimensions of the actuation system not to result an obstacle during wrist exten-
sion. At the time of writing, experimental tests to evaluate the usability of the
whole mechatronic system are planned.
Acknowledgment. A special thank goes to Chiara Brogi that has collaborated to

the design of the prototype during her Bachelor Thesis.
References
1. Troncossi, M., Mozaffari-Foumashi, M., Parenti-Castelli, V.: An original classifica-
tion of rehabilitation hand exoskeletons. J. Robot. Mech. Eng. Res. 1(4), 17–29
(2016)
2. Keller, J.W., van Hedel, H.J.: Weight-supported training of the upper extremity in
children with cerebral palsy: a motor learning study. J. Neuro Eng. Rehabil. 14(1),
87 (2017)
3. Conti, R., et al.: Kinematic synthesis and testing of a new portable hand exoskeleton.
Meccanica 52(11–12), 2873–2897 (2017)
4. Bianchi, M., et al.: Optimization-based scaling procedure for the design of fully
portable hand exoskeletons. Meccanica 53(11–12), 3157–3175 (2018). https://doi.
org/10.1007/s11012-018-0858-7
Design of Bio-joint Shaped Knee Exoskeleton
Assisting for Walking and Sit-to-Stance
Mehmet F. Kapci1 and Ramazan Unal2(&)

1
Mechanical Engineering Department,
Abdullah Gul University, Kayseri, Turkey
mehmetfazil.kapci@agu.edu.tr
2
Ozyegin University, Istanbul, Turkey
ramazan.unal@ozyegin.edu.tr
Abstract. In this study, a bio-joint shaped knee joint exoskeleton is presented.

This design is meant for avoiding misalignment of the exoskeleton joint with the
biological knee joint. For this purpose a cam mechanism has been designed to
prevent the misalignment from translation of the femur on tibia. Additionally,
walking and sit-to-stance is passively assisted with a spring element that is
activated with the heel contact. A single spring is used for both walking and sit-
to-stance, due to the similar characteristics of the gait cycle and initial phases of
the sit-to-stance in joint stiffness.
1 Introduction
For the past few decades, new design solutions of knee joint have been proposed in the
exoskeleton design field. Most of them are mainly classified into three categories w.r.t. their
purpose [1], i.e., rehabilitation, assistance and strength augmentation like load carrying.
Rapid increase of the exoskeleton research also brings some important challenges.
One of them is, exoskeleton alters the biomechanics of the human movement and also
misalignments occur between the exoskeleton and human joint during the movement.
Misalignment at the knee causes significant changes at the human exoskeleton inter-
action forces especially at the thigh interface [2]. In most of the lower limb exoskeleton
design, knee joints are modeled as simplified engineering joints such as pin joint which
have fixed rotation axis although the knee joint has a non-uniform geometry with
varying surface and rotation axis [3]. Unlike a pin joint, tibia and femur condyles make
both a rotation and sliding movement on each other during the flexion and extension of
the knee as illustrated in Fig. 1 [4]. This rolling and sliding at the contact point on a
non-uniform surface causes a non-constant axis rotation center and also causes dis-
placement of the bio-joint from initial rotation axis that most of the exoskeletons’
rotation center is located.
Besides, there are several design solutions that addresses the issue. For example a
polycentric four bar mechanism [4] is used to eliminate the misalignment problem.
However, as shown in Fig. 1 there is a non-constant ratio between the rolling and
sliding motions. Another possible solution is to use a cam mechanism to accomplish
Research supported by TUBITAK under the project number: 1109B321600215.

https://doi.org/10.1007/978-3-030-01887-0_96
496 M. F. Kapci and R. Unal
Fig. 1. Sliding and rolling demonstration of the contact point at the knee joint (a) initial contact
point (b) instantaneous contact point [4]
distance increase between the rotation axis and lower link center of mass. In an
experimental setup [3] internal forces and torques are minimized significantly by using
an adaptive knee joint that is used cam mechanism for misalignment. Additionally, in a
body weight supporting lower limb exoskeleton [5], decrease in the plantar forces at the
knee is accomplished by an adaptive knee joint and it also shows several advantages
compared with some other weight supporting exoskeletons.
In this study, design of a knee joint exoskeleton with a bio-joint shaped cam
mechanism is proposed that aims to overcome the misalignment problem at the knee
with gait cycle and sit to stance passive support.
2 Data Analysis
Biomechanics of the knee joint is analyzed to assist the knee during walking and sit-to-
stance. During weight acceptance which starts with heel contact, torque on the knee
increase up to 45 Nm for a 75 kg person and starts to decrease with the transition of
flexion to extension. Afterwards, in swing phase compliant behavior in the knee is
observed. In the proposed design, gait cycle is passively supported by a spring which
stores the energy in the weight acceptance and releases after transition. In order to
avoid interference, the spring is deactivated at heel-off before the swing.
Unlike walking, sit-to-stance is not a cyclic motion and consists of 2 phases which
are loading phase from start to the peak torque and rising phase from peak torque to the
end [6]. A linear spring is used to support the rising phase that stores energy during
loading phase.
3 Design Parameters
In the conceptual design, in order to eliminate the misalignment of the bio-joint rotation
center and increasing length of the shank with respect to fixed rotation center at the
femur, the rotation center of the exoskeleton at the upper link is placed within a linear
slot at the lower link. The path for the cam slot is designed with considering the sagittal
shape of lateral compartment of the femur that roll and slide on the tibia [7]. A com-
pression spring is placed at the upper link and connected to cam roller fixed at the lower
Design of Bio-joint Shaped Knee Exoskeleton Assisting for Walking and Sit-to-Stance 497
link and rolls in the cam slot at the upper link. Spring deflects by horizontal and vertical
translation of the roller inside the cam slot. However, for activating the spring by heel
contact and deactivating by the heel off passively, spring is relocated to the lower link
and the cam slot is reproduced at the lower link as symmetric of the previous design
w.r.t. rotation center in transverse plane, which provides to increase of the upper link
length instead of lower link in order to eliminate the misalignment caused by tibio-
femoral translation. Revised design of the exoskeleton is shown in the Fig. 2.
Fig. 2. (a) Revised design (b) knee mech. (c) spring de/activation mech.
Length of the linear slot (Fig. 2b) is determined by distance change between the
exoskeleton rotation center and cam rollers that reaches 15 mm in the full flexion and
10 mm at the 90° which corresponds to posterior translation of the femur on tibia [8].
At the weight acceptance of the gait cycle, from 7° to 20° flexion, cam roller deflects
the spring by 10 mm. Maximum flexion at the cycle occurs at the end of backward
swing phase with about 65° flexion and a gear link mechanism (Fig. 2c) is used to
deactivate the spring when the heel is not in contact with the ground up to 70° in order
to prevent activation during the swing.
Spring force for both walking and sit to stance is shown in Fig. 3 and calculated as;
s
F¼ ð1Þ
cosa sinb r
F ¼kx ð2Þ
s: torque at the knee during gait and sit to stance

a: angle between roller link and lower link
b: angle between roller link and upper link
r:length between cam roller and rotation center
k: spring constant
x: deflection of the spring
498 M. F. Kapci and R. Unal
Fig. 3. Spring force analysis of the mechanism
To obtain the spring constant, several spring forces are calculated for both walking
and sit to stance from Eq. (1) for corresponding spring deflections (Fig. 4). Forces in
gait cycle are iterated above the 0 for matching with the sit to stance data. From the
regression lines it is observed that spring constant for walking is approximately
40 N/mm and it shows a quite close result for sit to stance data up to 20° flexion. Thus,
single spring is used in the design for both walking and sit to stance that is stiffer
between 7°–20° flexion.
Fig. 4. Spring forces for several flexion angles vs spring deflections.
4 Conclusion
In this study, a bio-joint shaped knee joint exoskeleton is proposed in order to over-
come misalignment of the exoskeleton joint at the knee. Upper link of the exoskeleton
is increased in length as flexion angle increase by using a cam mechanism that is placed
at the lower link in order to prevent the misalignment from translation of the femur on
tibia. Additionally, gait cycle and sit to stance is passively supported by a spring
mechanism that is activated with the heel contact. A single spring is used for both
walking and sit-to-stance, since the similar results are obtained between the gait cycle
and initial phases of the sit-to-stance.
Design of Bio-joint Shaped Knee Exoskeleton Assisting for Walking and Sit-to-Stance 499
References
1. Chen, B., et al.: Recent developments and challenges of lower extremity exoskeletons.
J. Orthop. Transl. 5, 26–37 (2016)
2. Zanotto, D., et al.: Knee joint misalignment in exoskeletons for the lower extremities. IEEE
Trans. Robot. 31(4), 978–987 (2015)
3. Wang, D., et al.: Adaptive knee joint exoskeleton based on biological geometries. IEEE
Trans. Mech. 19(4), 1268–1278 (2014)
4. Lee, K.M., Guo, J.: Kinematic and dynamic analysis of an anatomically based knee joint.
J. Biomech. 43(7), 1231–1236 (2010)
5. Wang, D., Lee, K.M., Ji, J.: A passive gait-based weight-support lower extremity exoskeleton
with compliant joints. IEEE Trans. Robot. 32(4), 933–942 (2016)
6. Wu, M., Haque, M.R., Shen, X.: Obtaining natural sit-to-stand motion with a biomimetic
controller for powered knee prostheses. J. Healthc. Eng. 2017, 3850351 (2017)
7. Iwaki, H., et al.: Tibiofemoral movement 1: the shapes and relative movements of the femur
and tibia in the unloaded cadaver knee. J. Bone Jt. Surg. 82(8), 1189–1195 (2000)
8. Hamai, S., et al.: In vivo healthy knee kinematics during dynamic full flexion. Biomed. Res.
Int., ID 717546 (2013)
ANT-M: Design of Passive Lower-Limb
Exoskeleton for Weight-Bearing
Assistance in Industry
Berkay Guncan and Ramazan Unal(&)
Mechanical Engineering Department, Abdullah Gul University, Kayseri, Turkey

{berkay.guncan,ramazan.unal}@agu.edu.tr
Abstract. This study describes the optimized design of a passive lower limb
exoskeleton for workers in the industry. The exoskeleton is aimed at helping
workers who carry heavy loads, by supporting their posture and reducing stress
in their knees which would prevent future injuries. However, most of the pre-
vious passive designs are insufficient in a way that they are bulky. Therefore,
this study is focused on achieving lightweight passive exoskeleton. Topology
optimization has been carried out to reach this goal. The results are validated
using finite elements methods, in ANSYS environment.
1 Introduction
Numerous challenges have been encountered in the exoskeleton field since the
beginning of these emerging devices. Although electro-mechanical structures, i.e.
actuators, and control strategies are very crucial to overcome some of these challenges,
the mechanical design of an exoskeleton is also important in order to conform with
biomechanics and improve power efficiency both mechanically and metabolically [1].
In specific, designing lower-limb exoskeleton have its own critical attributes that
decides the function of the exoskeleton which will allow its interaction with the
humans. Keeping that in mind, most of the exoskeletons are designed in a way that it
covers the body and has anthropomorphic designs to reduce both weight and volume
and improve compatibility with the kinematics of the body.
For knee exoskeletons single axis joints would often be used [2], yet the polycentric
behavior of the knee yields these types of design less useful in terms of kinematics.
Therefore, to reduce the misalignment and increase the wearing comfort, single knee
axis should be avoided.
This paper introduces optimum topology for the knee exoskeleton called ANT-M:
lightweight, fully passive knee exoskeleton. It provides weight-bearing assistance to
the workers in the factory to effectively lift the weights by supporting the knee and
providing appropriate posture. For full solution, including walking, running etc. pre-
viously developed lower-limb exoskeletons could be investigated, such as HERCULE
that can reach 5 km/h walking speed [3], HUMA, a weight-bearing assistance
exoskeleton, which has polycentric knee design [4]. More impressing data is provided
by HULC exoskeleton [5], with maximum speed of 11 km/h for the military purposes,
while MIT exoskeleton has only 3.3 km/h in the test phase [6]. However, the HULC is

https://doi.org/10.1007/978-3-030-01887-0_97
ANT-M: Design of Passive Lower-Limb Exoskeleton 501
powered hydraulically, hence its main downside is agility, since hydraulic systems are
bulky and cannot respond as quickly as desired. While the passive exoskeleton ANT-M
is not designated for walking or running, it is agile enough for the specific task of
lifting weight and expected to reduce the risk of joint injuries by providing correct
posture and reducing the stress in the knee joints.
The ANT-M is consisted of passive 1-DoF ankle, and passive 1-DoF spring
assisted knee joints to help the workers. The design is fully passive to overcome the
battery, weight, safety and control problems.
2 Weight Lifting
In this section the Olympic weight lifting biomechanics is analyzed to further gain
insight for the design of ANT-M weight bearing exoskeleton.
Maximum velocity of the weight is an important factor to determine the
exoskeleton parameters. The multiplication of the maximum velocity and weight stands
for the external physical component of the lifting, and the parameter is called speed-
strength power. Following that barbell mass = 100 kg, max velocity = 2 m/s, and
vertical lift path 1.25 m:
m g vmax ¼ Pss
m m
kg 2 ¼ W
s s
Kinetic energy:
100 kg 22 m
¼ 200 Nm
2 s
Potential energy:
9:81 m
100 kg 1:25 m ¼ 1226:3 Nm
s2
From the calculations, we can conclude that the lift component (potential energy) is
six times greater than the acceleration component (kinetic energy).
3 Conceptual Design
Schematic representation of ANT-M is given in Fig. 1, where springs are employed in

order to support weight-bearing. On top of springs, a linear actuator is added to the
design to actively assist weight bearing of the workers as a future reference.
502 B. Guncan and R. Unal
Fig. 1. Design specifications of the proposed model
The link lengths are defined by using the mean segment lengths of a human body.
For this purpose, it is found that the thigh is corresponding to 23.2% and the shank is
corresponding to 24.7% of the total body height of a 176 cm male human according to
Leva [7].
For the spring constant, from sit to stand we see that knee joint is requiring 1.2
Nm/kg. The world record weight lifter requires 15 Nm/kg. We interpolated this to an
80 kg person, lifting 30 kg weight and obtain the required force data [1].
We obtained the spring length data by applying the kinematics of sit to stand
data [1].
Then elastic constant of the spring is approximated by force-elongation plot as it is
given in Fig. 2.
Fig. 2. Elastic constant of spring
From the slope of linear fit to force-elongation curve the spring constant value is
found as k = 10.74 N/mm.
ANT-M: Design of Passive Lower-Limb Exoskeleton 503
4 Design Optimization
4.1 Model Preparation for Optimization
Before the optimization, the model is kept as simple as possible, almost as an ingot, to
let the algorithm find the optimum design.
For this process material selection is done by considering accessibility, manufac-
turability, price and strength values and as a result Al-7075 is selected as material due
to its low density, high strength, and low price.
4.2 Topology Optimization

The results of the topology optimization have been depicted in Figs. 3 and 4. In these
figures left side shows the end result of the initial ingot model by keeping stress
concentration and weight as first priority.
Fig. 3. Optimized design of lower (left) and upper (right) links
Initial and final designs of the exoskeleton frame have been presented in Fig. 4.
Fig. 4. Initial and final design of the exoskeleton

504 B. Guncan and R. Unal
5 Conclusion
In this study, ANT-M, the weight-bearing assistance exoskeleton design is proposed to

help the workers to lower their fatigue, increase working performance, and correcting
their posture. The exoskeleton initially has only consisting of passive components, i.e.
springs. Spring constant has been determined according to weight lifting biomechanics
and then the design of exoskeleton has been analyzed in environment ANSYS for
topology optimization to minimize weight w.r.t. sufficient strength.
References
1. Ferris, D.P., et al.: A physiologist’s perspective on robotic exoskeletons for human
locomotion. Int. J. Humanoid Rob. 4(03), 507–528 (2007)
2. Banala, S.K., et al.: Robot assisted gait training with active leg exoskeleton (ALEX). IEEE
Trans. Neural Syst. Rehabil. Eng. 17(1), 2–8 (2009)
3. RB3D: Exoskeleton Hercule. http://www.rb3d.com/wp-content/uploads/2015/06/RB3D_
BrochureEXO_HV3_EN_L.pdf
4. Hyun, D.J., et al.: Biomechanical design of an agile, electricity-powered lower-limb
exoskeleton for weight-bearing assistance. Robot. Autonom. Syst. 95, 181–195 (2017)
5. Martin, L.: Human Universal Load Carrier (HULC), United States of America. http://www.
army-technology.com/projects/human-universal-load-carrier-hulc/
6. Walsh, C.J., et al.: Aquasi-passive leg exoskeleton for load-carrying augmentation. Int.
J. Humanoid Rob. 4(03), 487–506 (2007)
7. Leva, P.D.: Adjustments to Zatsiorsky-Seluyanovs segment inertia parameters. J. Biomech.
29(9), 1223–1230 (1996)
Effects of an Inclination-Controlled Active
Spinal Exoskeleton on Spinal Compression
Forces
A. S. Koopman1(&), S. Toxiri2, M. P. de Looze1, I. Kingma1,

and J. H. van Dieën1
1
Department of Human Movement Sciences,
Faculty of Behavioural and Movement Sciences, Vrije Universiteit Amsterdam,
Amsterdam Movement Sciences, Amsterdam, The Netherlands
a.s.koopman@vu.nl
2
Abstract. Mechanical loading of the spine is a known risk factor for the
development of low-back pain. The objective of this study was to assess the
effect of an inclination-controlled exoskeleton on spinal compression forces
during lifting with various techniques. Peak compression decreased on average
by around 20%, and this was largely independent of lifting technique.
1 Introduction
Low-back pain (LBP) is the number one cause of disability in the world [1], with a
lifetime prevalence between 75–84% [2]. Mechanical loading of the spine has been
shown to be an important risk factor for the development of LBP [3]. Although
physically demanding tasks are gradually taken over by robots and cranes, still many
workers have to repeatedly lift heavy loads in jobs [4] that depend on the versatile
capabilities of the human.
Therefore, assistive devices have been developed that aim to support the trunk
during forward bending and lifting tasks. One goal of such devices is to reduce the
spinal compression forces, by taking over a part of the required muscular moment.
Active devices are stronger and more versatile compared to passive devices, and allow
for controlling the torque applied by the exoskeleton. However, it is still unclear how
the generated torque should be controlled, and how this interacts with human behavior.
The objective of this study was to assess the effect of an inclination-controlled
exoskeleton on spinal compression forces during lifting with various techniques.
This work was supported in part by the European Union’s Seventh Framework Programme under
grant agreement Robo-Mate no. 608979, in part by the European Union’s Horizon 2020 through
the SPEXOR project, contract no. 687662, in part by the People Programme (Marie Curie Actions)
under REA grant agreement SMART-E no. 608022 and in part by the Italian Workers’
Compensation Authority (INAIL).

https://doi.org/10.1007/978-3-030-01887-0_98
506 A. S. Koopman et al.
Secondly, it was investigated how well the inclination-controlled torque matched the
users need in terms of timing of the support torque relative to the required torque.
2 Methods
2.1 Exoskeleton
The device used in this study was developed as part of the EU-funded project
Robo-Mate, a revised second version (Mk2B) was used in this experiment (Fig. 1).
Details of the device (the EXO) can be found in Toxiri et al. (2018) [5]. Two actuators,
approximately aligned with hip flexion-extension axis, could generate a maximum
torque of 20Nm each. The controller provided the support as a sine function of the
inclination of the thorax.
Fig. 1. The experimental setup including; force plate, optotrak camera, marker clusters, EMG
and the EXO.
2.2 Participants & Experimental Procedure

Eleven healthy young males participated in the experiment. None of the participants
had a history of low-back pain. After signing an informed consent, each subject was
instructed to complete a lifting task with three different techniques; FREE, SQUAT and
Effects of an Inclination-Controlled Active Spinal Exoskeleton 507
STOOP, once WITH and once WITHOUT the EXO. Participants had to grasp an
object of 15 kg from mid-shin height and return to upright stance and subsequently
place the object back and return to upright stance once more. Full-body 3-D kine-
matics, ground reaction forces and back and abdominal muscle activity were measured.
2.3 Data Analysis

A dynamic bottom-up 3-D linked segment model [6] was used to calculate the net
moment (ML5S1) and reaction force (FL5S1) at the L5S1 intervertebral disc. Linear
envelopes of the EMG signals were normalized to maximum voluntary contractions
and used as an input to an EMG driven trunk muscle model [7, 8]. For each participant,
a best fit between ML5S1 and the muscle moment was obtained by optimizing three
parameters (overall muscle maximum stress, rest length, and passive muscle force
scaling factor) over all lifts performed by a participant in the WITHOUT condition.
Compression forces at the L5S1 intervertebral disc (Fcomp_L5S1) were finally obtained
by summing the muscle forces and net reaction forces and projecting them on the
coordinate system of the L5S1 disc.
2.4 Statistics
Statistical differences in compression forces were tested along the complete time series
of outcome values using one-dimensional statistical parametric mapping (SPM1D).
A SPM1D two-way repeated measures ANOVA with the factor EXO (WITHOUT &
WITH) and factor technique (FREE, SQUAT & STOOP) was conducted.
A significant main effect of EXO was found in phases of forward bending (Fig. 2).
Although some short significant episodes were found, no relevant main effects of or
interaction with technique were found. Furthermore, the instant of peak support of
Mrobo neither coincided with the instant of peak inclination, nor with the instant of
peak ML5S1 (Fig. 3). The misalignment of the Mrobo peak and the inclination peak was
due to performance limitations of the actuators. The difference between the inclination
peak and loading peak is an inherent feature of lifting.
508 A. S. Koopman et al.
Fig. 2. SPM 1D output, showing the main effect of EXO, Technique and their interaction on
spinal compression forces. Significant intervals are indicated with a grey area. Lower panel
shows the mean spinal compression time series averaged over subjects for all conditions.
Peak loading
Peak inclination
Peak Mrobo
Fig. 3. Timing differences between the peak of the subject generated moment (Mpp) and trunk
flexion for a FREE lift without and with (INC) EXO. For the WITH condition, also the peak of
the support torque (Mrobo) and the inclination angle (Robo inclination) are shown.
4 Conclusion
The EXO with inclination-controlled torques substantially reduced compression forces

at L5S1 during phases of forward bending. Peak compression decreased on average by
around 20%, and this was largely independent of lifting technique. Apart from
Effects of an Inclination-Controlled Active Spinal Exoskeleton 509
generating larger moments, support could be improved by an improved control method

that generates peak support at the instant of peak loading.
REFERENCES
1. Hartvigsen, J., et al.: What low back pain is and why we need to pay attention. The Lancet
(2018)
2. Thiese, M.S., et al.: Prevalence of low back pain by anatomic location and intensity in an
occupational population. BMC Musculoskelet Disord 15(1), 283 (2014)
3. Coenen, P., et al.: The effect of lifting during work on low back pain: a health impact
assessment based on a meta-analysis. Occup. Environ. Med. 71(12), 871–877 (2014)
4. Eurofound, Trends in job quality in Europe. Publications Office of the European Union,
Luxembourg (2012)
5. Toxiri, S., et al., Rationale, Implementation and Evaluation of Assistive Strategies for an
Active Back-Support Exoskeleton. Frontiers in Robotics and AI, 2018. 5(53): p
6. Kingma, I., et al.: Validation of a full body 3-D dynamic linked segment model. Hum. Mov.
Sci. 15(6), 833–860 (1996)
7. van Dieën, J.H.: Are recruitment patterns of the trunk musculature compatible with a synergy
based on the maximization of endurance? J. Biomech. 30(11–12), 1095–1100 (1997)
8. van Dieen, J.H., Kingma, I.: Effects of antagonistic co-contraction on differences between
electromyography based and optimization based estimates of spinal forces. Ergonomics 48(4),
411–426 (2005)
Novel Mechanism of Upper Limb Exoskeleton
for Weight Support
Daegeun Park(&), Jesus Ortiz, and Darwin G. Caldwell

daegeun.park@iit.it
Abstract. Industrial workers suffer from musculoskeletal disorders. Especially,

the shoulder disorder affects many working movements. To reduce it, various
assistive devices have been developed. However, because of the complex
shoulder, there are several issues to apply the devices to the actual workspace,
such as bulky size or heavyweight. In this paper, a novel mechanism consisting
of three sub-mechanisms is suggested. It is focused on the weight support as the
main function. This target-oriented approach can reduce the complexity of the
mechanism and can lead to a light and compact structure. The mechanism is
designed to compensate three main issues between the exoskeleton and shoulder
joint: various arm raising direction at the initial posture, protraction/retraction,
and scapulohumeral rhythm. The design parameters were optimized by using the
center of the rotation of the shoulder, and the whole structure was designed with
a trajectory error of 5 mm or less.
1 Introduction
Industrial workers frequently suffer from musculoskeletal disorders. Among these, the
shoulder disorder is one of the main cause that reduces the workers’ productivity (about
18% of total musculoskeletal disorder patients in Italy Industry, 2016) [1]. It is usually
caused when muscle fatigue is accumulated or heavy loads are applied to the shoulder.
To reduce these disorders, there are several exoskeletal robots by applying the addi-
tional force to support the shoulder movement [2–5]. Because of the complicated
structure of the shoulder, the devices tend to be bulky and heavy (about 10*12 kg on
average for the portable devices).
In this research, we aim to support the arm weight to reduce muscle fatigue while
working. Through this target-oriented approach, the complexity of the mechanism can
be reduced. Then a less complex mechanism can increase the lightness and com-
pactness of the device. Moreover, the mechanism was developed based on the tendon-
driven mechanism to reduce the inertia of the distal linkage. This structure with lower
inertia can be actuated by a smaller actuator. Also, by using a single actuator to support
one directional force, gravity, the total system weight, and size can decrease.
This work has been funded by the Italian Workers’ Compensation Authority (INAIL).

https://doi.org/10.1007/978-3-030-01887-0_99
Novel Mechanism of Upper Limb Exoskeleton for Weight Support 511
In this paper, we suggest a novel mechanism of the upper limb exoskeleton. This
mechanism can be divided into three mechanisms for three main issues: various arm
raising direction at the initial posture, shoulder protraction/retraction, and scapulo-
humeral rhythm.
The first mechanism is suggested to change the direction of the assistive linkage
passively when the upper arm is parallel to gravity at the initial posture. The mecha-
nism consists of an active joint rotated against gravity, a passive rotational joint rotated
on the horizontal plane, an elastomer, and a passive prismatic joint on a cuff (Fig. 1 –
(a)). When the upper arm is parallel to gravity, the rotational axis of an assistive linkage
linked to the active joint is located with a certain angle with respect to a base plate.
When the wearer raises the arm in any direction, the assistive linkage is rotated on the
horizontal plane at first to parallelize the rotational axis of the assistive linkage to the
rotational axis of the shoulder. After this alignment, the assistive linkage rotates
according to the arm angle (Fig. 1 – (b)). When the wearer lowers the arm, the assistive
linkage goes back the initial posture because of the elastomer. The misalignment
between the assistive linkage and the human is compensated through the passive
prismatic joint.
Fig. 1. A conceptual design of the mechanism for the right arm and conceptual operation of the
first mechanism.
In the second mechanism, a ball-socket joint is used to compensate the

protraction/retraction of the shoulder (Fig. 2).
Fig. 2. A conceptual design of the second mechanism.

512 D. Park et al.
The third mechanism consists of four linkages, two constrained joints, one active
joint, and two tendons (Fig. 3). This mechanism works not only for scapulohumeral
rhythm (SHR) but also for transmitting the assistive force. For compensating the SHR,
tendon 1 is attached to the base plate and linkage 2 passing through two pulleys at the
constrained joints. At both pulleys, the tendon is wrapped in opposite direction at each
pulley. For example, at the right-side module, the tendon is wrapped in the counter-
clockwise direction at pulley 1 and wrapped in the clockwise direction at pulley 2.
When the actuator rotates linkage 1 in the counterclockwise direction with respect to
the base plate, tendon 1 is pulled. Then tendon 1 pulls linkage 2 by rotating it in the
counterclockwise direction. Because the active joint is located at the end of linkage 2, a
center of rotation (CoR) of the exoskeleton can change according to the human
shoulder CoR. For force transmission, tendon 2 is attached to linkage 1 and the
assistive linkage in the same way as tendon 1. When tendon 1 pulls linkage 2 with
respect to linkage 1, tendon 2 is pulled and the assistive linkage rotates since tendon 2
pulls the assistive linkage with respect to linkage 2. Finally, the assistive force is
transmitted from the base plate to the assistive linkage.
Fig. 3. Schematics of the third mechanism with parameters and the conceptual operation of the
third mechanism
3 Design Methodology
3.1 Scapulohumeral Rhythm Compensation
The parameters of the mechanism for SHR are shown in Fig. 3. The shoulder CoR is
derived by (1) and (2).

xa ¼ l1 sin h1 þ l2 sinðh1 þ h2 Þ
ð1Þ
ya ¼ l1 cos h1 þ l2 cosðh1 þ h2 Þ þ y0
Dd1 ¼ r1 Dh1 ¼ r2 Dh2 ð2Þ
where ðxa ; ya Þ is the shoulder CoR. y0 is the y-position of the joint between the center
of rotation of the scapular (Fig. 4) and the center of pulley 1. li ði ¼ 1; 2Þ is the length of
linkage 1 and 2. Ddi ði ¼ 1; 2Þ is the pulled length of tendon 1 and 2. Dhi ði ¼ 1; 2Þ is
Novel Mechanism of Upper Limb Exoskeleton for Weight Support 513
the angle of the joint 1 and 2. Based on the shoulder CoR data derived by (3), all design
parameters (Fig. 4) are calculated as the example.

xCoR ¼ lCoR cos hCoR
ðl ¼ 0:15m; hCoR ¼ 0 60 Þ ð3Þ
yCoR ¼ lCoR sin hCoR CoR
where lCoR is the length between the body center and the shoulder CoR. hCoR is the
angle of the shoulder CoR.
The simulation result is shown in Fig. 4. By tracking the SHR trajectory, the
misalignment between the shoulder and exoskeleton decreases under 5 mm.
Fig. 4. The simulation result of CoR trajectory with error comparing the human and
exoskeleton.
3.2 Force Transmission

The assistive force is derived by (4), and (5).
sac ¼ T1 r1 ; T1 ¼ T2 ; T2 ras ¼ sas ð4Þ
sas ¼ ðras =r1 Þ sac ð5Þ
where sac is the actuation torque, and sas is the assistive torque. Ti ði ¼ 1; 2Þ is the
tension of the tendon 1 and 2. ras is the radius of the pulley of the active joint.
4 Conclusion and Discussion
The novel mechanisms for the upper limb exoskeleton are suggested to reduce the
structural complexity. This mechanism is focused on the arm weight support and
consists of three mechanisms controlled by one actuator with the tendons. Although
these mechanisms transmit the assistive force from the actuator to the upper arm by
compensating the misalignment between the exoskeleton and human, there are some
514 D. Park et al.
remaining issues. In the first mechanism, mechanical delay can occur while rotating the
assistive linkage on the horizontal plane. This delay can disturb the wearer’s initial
movement. The second issue is how to fit the third mechanism for each wearer easily.
The final structure of the third mechanism is fixed based on the CoR trajectory, which
can vary among individuals. Therefore, to use the device universally, a method to
adjust the device to the wearer should be suggested as the future works.
References
1. OPEN DATA INAIL: https://dati.inail.it
2. Maciejasz, P., Eschweiler, J., Gerlach-Hahn, K., Jansen-Troy, A., Leonhardt, S.: A survey on
robotic devices for upper limb rehabilitation. J. Neuro Eng. Rehabil. 11, 3 (2014)
3. Lo, H.S., Xie, S.Q.: Exoskeleton robots for upper-limb rehabilitation: State of the art and
future prospects. Med. Eng. Phys. 34(3), 261–268 (2012)
4. Niyetkaliyev, A.S., Hussain, S., Ghayesh, M.H., Alici, G.: Review on design and control
aspects of robotic shoulder rehabilitation orthoses. IEEE Trans. Hum.-Mach. Syst. 47(6),
1134–1145 (2017)
5. Gopura, R.A.R.C., Bandara, D.S.V., Kiguchi, K., Mann, G.K.I.: Developments in hardware
systems of active upper-limb exoskeleton robots: a review. Robot. Auton. Syst. 75, 203–220
(2016)
Human-Centered Design of an
Upper-Limb Exoskeleton for Tedious
Maintenance Tasks
Andrea Blanco(B) , Jorge A. Dı́ez, David López, José V. Garcı́a,

José M. Catalán, and Nicolás Garcı́a-Aracil
Biomedical Neuroengineering Research Group, Miguel Hernandez University,

Elche, Spain
ablanco@umh.es
Abstract. In this paper, the design of a new exoskeleton to enhance

and support the human abilities in industrial maintenance environments
is presented. The motivation to design this device arises from the neces-
sity to reduce or eliminate musculoskeletal disorders caused by manual
movement of heavy loads, prolonged raised arm working postures and
repetition of movements associated with the installation and mainte-
nance of industrial facilities.
1 Introduction
In the last European Working Conditions Survey [1], three main physical risks at
work were studied and analyzed: (i) Posture-related (ergonomic) risks; (ii) Ambi-
ent risks; and (iii) Biological and chemical risks. Specifically, posture-related risk
measures exposure to vibrations, tiring positions, lifting people, carrying heavy
loads and repetitive movements at work. These are the most prevalent risks
in Europe – in particular, repetitive hand and arm movements (62% of workers
report this) – and include the risks that can play a role in the common workplace
complaint, musculoskeletal disorders.
Musculoskeletal injuries are a set of inflammatory or degenerative lesions of
muscles, tendons, ligaments, etc., ranging from discomfort, nuisance or pain, to
more severe medical conditions that require sick leave and even receiving medical
treatment. Moreover, they are one of the most common work-related complaints,
affecting millions of workers and costing billions of euros to employers [1].
This work has been supported by the European Commission through the project
AIDE: Adaptive Multimodal Interfaces to Assist Disabled People in Daily Activities
(Grant agreement no. 645322); by the AURORA project (DPI2015-70415-C2-2-R),
which is funded by the Spanish Ministry of Economy and Competitiveness and by
the European Union through the European Regional Development Fund (ERDF),
“A way to build Europe” and by and by Conselleria d’Educació, Cultura i Esport of
Generalitat Valenciana through the grants ACIF2016/216 and APOTIP/2017/001.
https://doi.org/10.1007/978-3-030-01887-0_100
516 A. Blanco et al.
A good review of assistive exoskeletons that have specifically been developed

for industrial purposes and to assess the potential effect of these exoskeletons on
reduction of physical loading on the body can be found in [2]. Despite the high
interest for exoskeletons with an industrial application purpose, a large-scale
implementation of exoskeletons in industry has still a long way to go.
The ExIF project (Intelligent Robotic Exoskeleton and Advanced Interface
Systems Man Machine for maintenance tasks in the Industries of the Future)
arises from the necessity to reduce or eliminate musculoskeletal disorders asso-
ciated with the installation and maintenance of industrial facilities. This is a
collaborative research project developed by MovilFrio, Miguel Hernandez Uni-
versity and Polytechnic University of Valencia, and funded by Centre for the
Development of Industrial Technology (CDTI) of Ministry of Economy and
Competitiveness, which fosters the technological development and innovation
of Spanish companies.
The ExIF project proposes the development of a robotic upper-limb exoskele-
ton that will be supported by a passive exoskeleton structure for the lower-limb
that transmits the supported loads to the ground, as can be seen in Fig. 1. The
project also proposes the development of an advanced human-machine inter-
face system based on Augmented Reality techniques linked to a comprehensive
maintenance computer system. The human-machine interface will have cognitive
capabilities that allow it to be proactive based on the analysis of the context
and the environment.
Fig. 1. Main components of the ExIF project.
2 Exoskeleton Design
2.1 Concept of the Robotic System
In the first place, it is necessary to know the movements of the area that is
intended to support, the upper-limb. Specifically, the proposed exoskeleton will
have 5 degrees of freedom, 3 belonging to the shoulder, one to the elbow and
one corresponding to the movements of pronation and supination of the wrist.
The maximum ranges of each of these movements have been defined based on
the information collected in [3]. Oversizing the design of the mechanism to these
Human-Centered Design of an Upper-Limb Exoskeleton 517
maximum ranges, it will be positioned on the side of safety, since the operator
will not reach in any case these mechanical limits when performing the tasks of
their job.
Taking into account these movements, a first concept of the system is designed
that serves as a structure to use it as the basis of the design in the subsequent pro-
totype. This way, the kinematic compatibility of the mechanism can be studied
to decide the optimal position of the actuators. In addition, a pronosupination
mechanism designed by the nBio group within the framework of the European
project AIDE [4] has been incorporated.
2.2 Selection of Actuators
To determine the actuators that the device must have, it is necessary to know
the efforts that appear in the joints of the user’s arm when performing the tasks
of his/her job. With this objective, a simulation has been carried out with the
AnybodyTM program, for which a series of trajectories of the user’s upper-limb
have been recorded during the accomplishment of a target tasks in his/her work
environment in which the device is intended to be incorporated.
By means of this information it is possible to estimate what torque each actu-
ator must perform to compensate the load that the worker must bear without
the help of the exoskeleton. The values extracted from the simulation correspond
to the efforts of the operator when he/she performs the task with the passive
device, so it will be necessary to recalculate those values when the worker per-
forms the task with the definitive exoskeleton, since the weight, both of the
structure and of the motors, has not been considered in the current analysis. To
be on the security side, the values extracted from the simulation have tripled,
assuming that, in no case, the exoskeleton will exceed that weight.
2.3 Exoskeleton Model
Taking into account all the described requirements, it has been designed the
exoskeleton shown in Fig. 2.
This exoskeleton is able to reproduce the movements of the human body,
as well as reduce the efforts borne by the user. It is an ergonomic device that
can be adapted to different arm dimensions, so that a single exoskeleton can
be used by different users, regardless of their complexion. This adaptability is
achieved with the incorporation of linear guides located in the area of the back,
which allow positioning correctly the motors associated with the shoulder. The
anthropometric dimensions that have been taken as reference for the design of
the mechanism are included in [5].
518 A. Blanco et al.
Fig. 2. Exoskeleton
3 Validation of the Robotic System

To demonstrate the benefits of using an upper-limb exoskeleton when it is per-
formed the assembly tasks that have been defined in the project, the inverse
dynamic analysis for the musculoskeletal model of the human body has been
carried out using the AnybodyTM program, comparing the efforts withstanded
by the user when performing the tasks with and without the device.
The results obtained from this simulation are collected in Table 1.
Table 1. Simulation results.
Operator without Assisted arm Active full

exoskeleton exoskeleton exoskeleton
sA/A [Nm] 2.27 1.26 1.20
sF/E [Nm] 12.83 4.93 4.63
sI/E [Nm] 6.12 0.12 0.01
eF/E [Nm] 3.32 2.66 2.69
Max. Torso muscle 18.61 44.00 2.37
activity [%]
It has also been analyzed whether the incorporation of a passive exoskeleton

of a lower-limb implies any improvement for the user, and after observing the
results obtained, we can affirm that the proposed solution is valid.
4 Conclusion
This paper describes the design process of an 5 DOF upper-limb exoskeleton that
aims to support the operator during the performance of industrial maintenance
work, proposing a solution to the problem of musculoskeletal disorders existing
in the industry due to overstress and tiring postures taken by operators in their
jobs.
The ExIF project proposes, in addition to the manufacture of the upper-limb
exoskeleton, its anchorage to a passive exoskeleton structure for the lower-limbs,
that allows to transmit the weight of the device to the ground.
Human-Centered Design of an Upper-Limb Exoskeleton 519
After analyzing the data we extracted from the simulation, we can say that
using an upper-limb exoskeleton in the tasks defined in the project results in
a decrease in the efforts supported by the operator. In addition, the benefit of
anchoring this robotic system to a lower-limb exoskeleton that transmits the
weight and effort supported to the ground has been proven.
References
1. Eurofound: Sixth European Working Conditions Survey - Overview report. Publi-
cations Office of the European Union (2016)
2. de Looze, M.P., Bosch, T., Krause, F., Stadler, K., OSullivan, L.: Exoskeletons for
industrial application and their potential effects on physical work load. Ergonomics
59(5), 671–681 (2015)
3. Norkin, C.C., White, D.J.: Measurement of Joint Motion: A Guide to Goniometry.
FA Davis, Philadelphia (2016)
4. Dı́ez, J.A., Blanco, A., Cataln, J.M., Badesa, F.J., Sabater, J.M., Garcia-Aracil, N.:
Design of a prono-supination mechanism for activities of daily living. In: Converging
Clinical and Engineering Research on Neurorehabilitation II, vol. 15, pp. 531–535
(2017)
5. Benjumea, A.C.: Datos antropométricos de la población laboral española. Pre-
vención, trabajo y salud: Revista del Instituto Nacional de Seguridad e Higiene
en el Trabajo 14, 22–30 (2001)
A Supernumerary Soft Robotic
Hand-Arm System for Improving Worker
Ergonomics
Andrea S. Ciullo1,2(B) , Manuel G. Catalano1 , Antonio Bicchi1,2 ,

and Arash Ajoudani1
1
Istituto Italiano di Tecnologia, Genova, Italy
andrea.ciullo@iit.it
2
Bioengineering and Robotics Research Center of University of Pisa, Pisa, Italy
Abstract. Long exposure to overload and vibration transmission on the

upper limb are among the high risk injury factors in industrial environ-
ments. They contribute to the development of musculoskeletal disorders,
which can lead to economic and social setbacks. To address this issue,
robotic systems have been developed that act either as an autonomous
system or in collaboration with the workers. In this direction, and with
the aim to develop a system that contributes to a simultaneous reduction
of the overloading and vibration transmission, we present a novel wear-
able soft robotic hand-arm system. Preliminary experimental results in
a vibrational tool use are reported to shown the potential of the system
in improving worker ergonomics.
1 Introduction
Several industrial job duties require physical demands that may result in exces-
sive internal forces and vibration transmission, increasing the risk for work-
related musculoskeletal disorders (WMSD). In the last decades, numerous safety
policies were defined and applied to reduce the workers’ exposure to such risks.
A parallel line of work deals with the use of robotic solutions for improving both
the safety and the comfort of the industrial workers. Such systems can act either
autonomously, to perform the tasks that are potentially risky for human work-
ers, or in collaboration with them (e.g. collaborative robots or exoskeletons), so
that a certain level of dexterity can be added to the industrial processes [1].
More recently, a new robotic paradigm has taken place in this field: the
Supernumerary Robotic Limbs (SRL). Differently from an exoskeleton, a SRL
does not constraint the human joint to follow a specific trajectory. During the
task execution, the SRL can augment the worker’s ability, e.g. by providing
additional robotic arms and/or legs. This can help the workers to carry objects
Funded, in part, by the European Project H2020 SoftPro, Grant Agreement No.
688857.
https://doi.org/10.1007/978-3-030-01887-0_101
A Supernumerary Soft Robotic Hand-Arm System 521
while using a tool with their own hand [2], or to augment the body stability in
uncomfortable working positions (e.g. crawling-like) [3].
Noteworthy, most of these devices only reduce the payload on the arm joints,
leaving the fingers still overloaded for grasping. In addition, this could be painful
especially using vibrational tool (e.g. drill or polisher). In fact, vibration trans-
mission on the hand/arm have been correlated with chronic nerve and tendon
disorders [4]. Some industrial solutions (e.g. www.deltaregis.com/Tool Supports)
already deal with this problem, but they are usually grounded, limiting the nat-
ural workspace. To address these challenges, we present a new wearable and
under-actuated soft robotic platform. The system is composed of the Pisa/IIT
SoftHand [5] mounted on a commercial steady-cam (Armor Man 2.0, Tilta), with
four passive dampers in between. The whole system reduces the load on both the
arm and finger joints during tasks. In addition, since the user does not hold the
tool directly with his/her own hand, most of the vibration are dissipated by the
system, highly reducing the injury risk on the nerves and tendons. A detailed
description of the system, and its potential to improve human ergonomics is
provided below.
Fig. 1. The proposed supernumerary soft robotic hand-arm system.
2 System Description
Developed with the aim at improving the ergonomics of workers in industrial
environments, it is a wearable system, composed of two passive gravity com-
pensator arms (the Armor Man 2.0 suite), integrated with two robotic hands.
The suite can be worn as a backpack, while the robotic hands are integrated
thanks to a custom mechanical wrist interface (see Fig. 1). This interface allows
passively the prono-supination rotation of the robotic hand.
522 A. S. Ciullo et al.
In this first prototype, a handle attached to the wrist interface was used to
proportionally control the opening/closure of the SoftHand (see Fig. 1).
3 Improving Worker Ergonomics
To improve worker ergonomics in industrial environments, the system is meant

to reduce two of the main injury factors: vibration transmission and overloading.
Vibration Suppression: Several studies showed a direct correlation between
the exposure of hand-transmitted-vibration (HTV) and symptoms and signs of
disorders in the vascular, neurological and osteoarticular systems of the upper
limb [4]. Thanks to this system, an important suppression of these vibrations on
the worker’s upper limb is carried out. This suppression is due to mainly to the
four dumper by which the robotic hand is attached on the mechanical interface
and to the padding of both the robotic hand palm and the suite. A simplified
mass-spring-damper schema of a user wearing the system is shown in Fig. 2.
The overall impedance, composed by the four dampers, with intrinsic damping
that can be chosen based on the vibration suppression cutoff frequency, and the
Pisa/IIT SoftHand, is shown under SOFTHAND block. This block will isolate
the vibration from a tool to the human hand. The residual vibration is also
reduced by the Armor Man’s intrinsic mass-spring-damper system. The effect of
such design parameters in reducing the level of vibration from the tool to the
person’s hand can be calculated from:
Fe (s)
Xb (s) =
(s2 m SH + sbSH + kSH )
−1
(sbSH + kSH )2
× 1−
(s2 M + sB + K)(s2 mSH + sbSH + kSH )

Xb (s)(sbSH + kSH ) (sbSH + kSH )2
Xa (s) = 2 1− 2
(s mSH + sbSH + kSH ) [s M + sB + K]
where Xa (s) and Xb (s) are the Laplace transform of the motion equations of
the point a and b. KSH , bSH and mSH are the spring coefficient, the viscous
damping coefficient and the mass of SoftHand. While, M is the sum of the mass
of the human arm/hand and the Armor Man (mHAH and mAM ), and B and
K are the sum of all the viscous damping coefficient and the spring coefficient
respectively.
Load Reduction: The system helps the worker to reduce the load not only
on the arm joints (mainly elbow and shoulder), but also on the finger ones. In
fact, the user does not need to carry the tool directly with his/her own hand. In
this way, during the working phase the fingers are not over-loaded. In addition,
thanks to this system, the worker can hold an object while using other tools (e.g.
screw driver, see Fig. 3).
A Supernumerary Soft Robotic Hand-Arm System 523
Fig. 2. Mass Spring Damper model of a user wearing the system, where Ki , bi and
mi are the spring, and viscous damping coefficients and mass respectively. fe is the
external force applied to the whole system.
Fig. 3. The system can hold a part for the worker while using two hands to manipulate
it.
Fig. 4. Top: The envelope (RMS window of 150 samples) of the EMG activity signal of
the biceps muscle. Bottom: Acceleration on the hand (removed mean value). In both:
orange with the system, blue without.
524 A. S. Ciullo et al.
4 Preliminary Results
To evaluate both the vibration and the load reduction on human arm, an exper-
imental task was designed, simulating the usage of a vibrating tool (mass: 1 kg).
The EMG activity and the acceleration on the user’s arm, while holding a rotat-
ing polisher for 10 s (no contact with any surface), were acquired using eight
wireless EMG electrodes the Trigno EMG system (Delsys Inc.). Sensors were
attached to the right arm as shown in Fig. 1: two on the dorsal side of the hand;
two on the forearm (one on the extensor muscle and one on the flexor muscle of
the wrist); two on the arm (one on biceps and one on the triceps); two on the
shoulder deltoids anterior, and posterior (EMG rate: 2.0 kHz; acc. rate: 150 Hz).
Figure 4 illustrates the preliminary results of this experiment. On the top plot,
the activation of the biceps muscle, as the major muscle in holding the object,
with (orange line) and without (blue line) the use of the system demonstrates a
significant reduction of the loading on the right arm. While, in bottom plot, the
vibration suppression is shown by the different amplitude of the two signal.
References
1. Ajoudani, A., Zanchettin, A.M., Ivaldi, S., Albu-Schäffer, A., Kosuge, K., Khatib,
O.: Progress and prospects of the human-robot collaboration. In: Autonomous
Robots, pp. 1–19 (2017)
2. Parietti, F., Asada, H.H.: Supernumerary Robotic Limbs for aircraft fuselage assem-
bly: body stabilization and guidance by bracing. In: IEEE International Conference
on Robotics and Automation (ICRA) (2014)
3. Kurek, D.A., Asada, H.H.: The MantisBot: design and impedance control of super-
numerary robotic limbs for near-ground work. In: IEEE International Conference
on Robotics and Automation (ICRA) (2017)
4. Bovenzi, M.: Health effects of mechanical vibration. G Ital. Med. Lav. Ergon. 27(1),
58–64 (2005)
5. Catalano, M.G., Grioli, G., Farnioli, E., Serio, A., Piazza, C., Bicchi, A.: Adaptive
synergies for the design and control of the Pisa/IIT SoftHand. Int. J. Robot. Res.
33, 768–782 (2014)
An Optimization Approach to Design
Control Strategies for Soft Wearable
Passive Exoskeletons
Andres F. Hidalgo Romero1(&), Eveline Graf2, and Eduardo Rocon1

1
Centro de Automática y Robótica (CSIC), Madrid, Spain
{af.hidalgo,e.rocon}@csic.es
2
Institute of Physiotherapy,
ZHAW Zurich University of Applied Sciences, Winterthur, Switzerland
grav@zhaw.ch
Abstract. Soft assistive devices constitute a promising alternative to help people

with mobility impairments. Nevertheless, some issues as the control of these
systems preclude from their generalized usage in common daily activities. The
objective of this paper is to obtain the activation profile for controlling a clutched
spring to store and release energy in a way that helps the subject to achieve a
specific movement target. To do this, a parameter and partially constrained
optimization method has been implemented. The results obtained showed a
clutch activation profile which is synchronized with a reduction of the hip flexion
torque exerted by the subject. Additionally, significant computational times
savings have been obtained due to important reductions of the size of the opti-
mization problem introduced by a partitioning of the state and control vectors.
1 Introduction
Soft devices constitute a promising alternative to help people with mobility impair-
ments, but there are still some issues to generalize their usage due to limitations related
to energy harvesting, design, weight and control. To tackle the three first problems, soft
passive assistive devices have been proposed as a plausible solution [1, 2], but the
control of this kind of systems is far from being a solved matter; the optimization
methods [3] being in this regard a suitable option.
A scenario of particular difficulty is the case of soft wearable passive robotic
exoskeletons, where the device works in parallel with the human but not independently
from him. This apparently simple dynamic process involves some challenges. From an
optimal control point of view, the difficulty of this situation relies on two aspects. One
is related to the oscillatory nature of the gait, while the other is related to the deter-
mination of weights that measure the influence that each of the forces (the one exerted
by the human and the one exerted by the device) has on the movement. The issue of the
oscillatory movement is that it changes the sense of the optimization objective function,
while the system of parallel forces introduces dynamic indeterminacies.
In order to address this problem, in this article we propose a parameterized con-
strained optimization approach to control the activation of a clutched elastic element

https://doi.org/10.1007/978-3-030-01887-0_102
526 A. F. Hidalgo Romero et al.
that helps to increase the hip flexion movement in a partially prescribed way. To this
end, the state and control vectors are partitioned between unknown, known and par-
tially known variables, something that considerably reduces the size of the optimization
problem and, at the same time, provides a means to solve the indeterminacies.
2 Methods
2.1 Musculoskeletal and Soft Wearable Device Models
An OpenSim [4] model (a 68-year-old stroke patient, mass 72 kg and height 1.70 m,)
was built based on walking experiments carried out with a motion capture system and
force plates. In order to reduce the size of the problem, virtual rotational actuators have
been considered at each joint instead of muscles. The joint trajectories and torque
excitation profiles of the virtual actuators were obtained by means of the IK and CMC
OpenSim tools.
As solely the excitation of an elastic element to modify the hip flexion movement is
going to be computed, the musculoskeletal model has been further reduced to one leg
only. Thus, instead of focusing on the movement of the complete musculoskeletal
system with respect to a stationary ground, the problem has been inverted by con-
sidering a moving ground with respect to a five-degree-of-freedom-DOF pelvis-leg
system, where the pelvis is fixed to the inertial system, the five DOF corresponding to
the three hip, knee and ankle rotations.
The location of the wearable device (the anchor points of hip elastic element) has
been determined in order to maximize the energy stored (elastic band elongation)
during the hip extension movement. This has been done by considering the elastic band
elongation as a function of the hip flexion angle and a coefficient gamma that express
the rate between the horizontal distance of the hip and thigh anchor points of the elastic
band to the center of rotation of the pelvis measured in the sagittal plane. This relation
is illustrated in Fig. 2, where both surfaces correspond to knee anchor points located in
the back part of the thigh. As it can be seen, the rearest the point, the better.
2.2 Optimization
The optimization problem is based on a Direct Collocation [5] approach originally
published by [6]. In this article, the proposed optimization algorithm simultaneously
obtains a partial prediction of hhf ; ehf and ec ; the hip flexion trajectory movement, the
excitation of the new hip flexion torque and that necessary for engaging the clutch of
the elastic element that minimizes the sum of the squared joint torques excitations, as
the following objective function indicates
Xnj Ztf
J¼ i¼1
e2i ðtÞdt ð1Þ
t¼0
An Optimization Approach to Design Control Strategies 527
and constrained by the dynamics of the system which is represented by the differential
equation of motion
x_ ¼ f ðx; uÞ ð2Þ
as well as partial constraints to the states and torques. These partial constraints are
discretely specified at each direct collocation node as shown in Fig. 1 (left). The
magenta points indicate the hip flexion joint trajectory constrains, while green and
black points show the clutch and hip torque constraints.
Trajectory constraint collocation points

0.8 0.1
HJT
HJC 0.08
0.6 TE
CEC
HTC 0.06
0.4
0.04
Torque excitation
Angle (rad)
0.2
0.02
0
0
-0.2
-0.02
-0.4 -0.04
-0.6 -0.06
5.7 5.8 5.9 6 6.1 6.2 6.3 6.4 6.5 6.6
Time (s)
Fig. 1. Hip flexion and torque target points (left). OpenSim model with the clutched elastic band
in blue (right).
Fig. 2. The elastic band elongation as a function of the knee and hip anchor points.
3 Results
Figure 3 (left) shows that the predicted hip flexion trajectory reaches the target points.
The right side of Fig. 3 depicts the change in the hip flexion excitation due to the
activation of the clutch and the contribution to the movement done by the elastic
element.
528 A. F. Hidalgo Romero et al.
0.8
0.6
0.4
Angle (rad)
0.2
Optimized predicted trajectory

-0.2 Actual trajectory
Trajectory constraints
-0.4
-0.6
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (s)
Fig. 3. Hip joint trajectory (left) and torques excitations (right) before (blue) and after (red) the
optimization, i.e., without and with the clutched elastic element activated by the computed
optimal control.
At the same time, Fig. 4 shows that only the initial guess for the clutch activation
and the original torque exerted by the subject without wearing the exoskeleton device
has changed. All the other torques remain unchanged since they were supposed to be
completely known and they were not included in the optimization problem.
Fig. 4. Joint torques excitations before (blue) and after (red) the optimization, i.e., without and
with the clutched elastic element.
By using 51 collocation points the problem was solved in 196 s. An optimization

problem of this size was solved in [6], where the authors reported resolution times in
the order of hours, even by using the IPOPT optimization library.
The proposed partially constrained optimization method has proven to be a suitable

tool to design biomimetic control strategies of wearable devices. The method has also
proven to be useful to eliminate dynamic indeterminacies and for reducing computa-
tional times.
An Optimization Approach to Design Control Strategies 529
Acknowledgment. This work has received funding from the European Union’s Horizon 2020
research and innovation programme under grant agreement No. 688175 (XoSoft).
References
1. Wehner, M. et al.: A lightweight soft exosuit for gait assistance. In: Proceedings of IEEE
International Conference on Robotics and Automation, pp. 3362–3369 (2013)
using an unpowered exoskeleton. Nature 522(7555), 212–215 (2015)
3. Anderson, F.C., Pandy, M.G.: Dynamic optimization of human walking. J. Biomech. Eng.
123(5), 381–390 (2001)
4. Delp, S.L., et al.: OpenSim: open-source software to create and analyze dynamic simulations
of movement. IEEE Trans. Biomed. Eng. 54(11), 1940–1950 (2007)
5. Ackermann, M., van den Bogert, A.J.: Optimality principles for model-based prediction of
human gait. J. Biomech. 43(6), 1055–1060 (2010)
6. Lee, L.-F., Umberger, B.R.: Generating optimal control simulations of musculoskeletal
movement using OpenSim and MATLAB. PeerJ 4, e1638 (2016)
Actuator Optimization
for a Back-Support Exoskeleton:
The Influence of the Objective Function
Tommaso Poliero1,2(B) , Stefano Toxiri1 , Darwin G. Caldwell1 , and Jesús Ortiz1

1
via Morego, 30, 16163 Genoa, Italy
tommaso.poliero@iit.it
2
Department of Informatics, Bioengineering, Robotics and Systems Engineering,
University of Genoa, 16145 Genoa, Italy
Abstract. Exoskeletons have been interest of researchers and develop-

ers spanning a wide range of applications. Active and passive or quasi-
passive exoskeletons are being developed. For this latter category, the
actuator parameter tuning is generally based on optimization studies.
This paper studies how the definition of the objective function affects the
passive or quasi-passive exoskeleton actuators configuration and param-
eters. It also provides indications on how focusing solely on energy opti-
mization might result in an assistance that could alter the user behaviour.
Therefore, usability of the exoskeleton and user comfort could be nega-
tively affected. Finally, as a result of the optimization output analysis,
we discuss about the advantages of designing a quasi-passive exoskeleton.
1 Introduction
Exoskeletons can be categorized according to the actuation relying on active or

passive elements. Active actuators (e.g., electrical motors, pneumatic muscles)
introduce energy into the system and, thus, can be used to provide assistance
to the user, according to some control strategy. Usually, active exoskeletons are
selectively required to deliver-assistance and to be transparent when no assis-
tance is desired [1]. On the other hand, passive actuators (springs, dampers) are
potentially lighter, cheaper and safer since they do not introduce new energy
into the system but store and release that coming from the user. As a drawback,
they have a predetermined assistance that cannot be automatically excluded
from the system, if needed [2]. Recently, elements as clutches are being stud-
ied for their capability of selectively enabling the flow of energy inside passive
elements. Clutches are active elements that can exploit different actuation princi-
ples such as layer jamming, electro-adhesion or electro-magnetism [3,4]. Defining
the combination of a clutch and a passive element as quasi-passive actuation,
This work has been founded by the Italian Workers’ Compensation Authority
(INAIL).
https://doi.org/10.1007/978-3-030-01887-0_103
Actuator Optimization for a Back-Support Exoskeleton 531
this latter presents the advantages of passive actuation and, moreover, it allows
to make the exoskeleton transparent if needed. As a consequence, the choice of
the passive or quasi-passive elements characteristics is extremely important since
they have a strong connection with the assistance and the considered task. In a
previous work [5], we introduced a model for the exoskeleton actuator configura-
tion optimization based on the minimization of the mechanical energy required
to perform a straight walking task.
In this paper, we extend the model to three subjects performing manual
material lifting of two different objects. The idea is to analyse the (quasi-)passive
actuators configuration and the optimal parameters of the exoskeleton taking
into consideration different cost functions.

Three users were required to lift rectangular boxes and, successively, the exten-
sion torque acting between the L5-S1 vertebrae (τbio ) was estimated in an inverse
dynamics fashion [1]. τbio can be reduced with the contribution of an exoskele-
ton (τexo ). Therefore, the new torque requirement (τnew ) for completing the task
results being:
τnew = τbio − τexo (1)

Details on the optimization algorithm formulation are reported in [5]. This
work presents a study on how the objective function J affects the optimal con-
figuration. In particular, we define the following functions to be minimized:
N
J1 = N1 i=1 | τbio (i) − τexo (i) |
N (2)
J2 = N1 i=1 (1 + α(i)) | τbio (i) − τexo (i) |
where N represents the total amount of samples. J1 tries to minimize the abso-
lute torque difference between the user requirements and that offered by the
exoskeleton. This cost function is not considering the human reaction to such
an assistance. Indeed, if Pnew (i) and Pbio (i)1 have different signs, this implies
that the user has to dissipate energy instead of providing it, or vice versa. We
define this as a change in the lifting behaviour. To penalize those topologies in
which Pnew and Pbio have different signs, we introduce in J2 a weight (α(i)) that
is null if Pnew (i) · Pbio (i) ≥ 0, otherwise it is set to a constant positive value.
The higher this value, the more the penalty is increased. Finally, the elements
included in this analysis are: damper, unidirectional spring and clutch.
3 Results
Table 1 summarizes the energy reduction for the three considered subjects and
for two different objects (7.5 kg and 15 kg). Column mix refers to subjects first
1
P stands for power and can be derived from (1) multiplying by the speed.
532 T. Poliero et al.
lifting the 7.5 kg and, then, the 15 kg boxes, successively. Figure 1 displays Pnew
associated with J1 (blue dotted line) and J2 (green solid line). It is possible to
see inverse behaviour in all the regions where Pbio (black dashed line) and Pnew
have opposite signs. As expected, J1 has significant inverse behaviour. Taking
into account the 9 tests used for the optimization, in 55% of the cases, the
optimal configuration for the exoskeleton is a parallel of damper and spring2
(K = 61 ± 9.90 [Nm/rad] and C = 9.24 ± 0.96 [Nms/rad]) and in the 22.5%
the optimal configuration is the parallel of two springs (K1 = 7.5 ± 4.94 and
K2 = 46 ± 0.2 [Nm/rad]) if considering J1 . Instead, analyzing J2 , in 88% of
the cases, the optimal configuration for the exoskeleton is a spring (stiffness
47.47 ± 5.49 [Nm/rad]) and in the 12% the optimal configuration results being
the series of two springs (K1 = 40 and K2 = 379 [Nm/rad]).
Table 1. Energy reduction percentage
J1 J2
7.5 Kg 15 Kg Mix 7.5 Kg 15 Kg Mix
Sub1 79.56% 67.77% 73.25% 66.77% 59.92% 62.13%
Sub2 80.91% 70.23% 74.76% 75.43% 70.10% 69.64%
Sub3 78.50% 67.67% 72.62% 60.75% 51.37% 61.56%
Fig. 1. The figure reports how the biological power requirements are reduced for Sub1
when lifting 7.5 kg (0–6 s) and 15 kg (6–12 s) simulating the assistance provided by the
optimal exoskeleton configurations for J1 and J2 (α = 15 if Pnew · Pbio < 0).
2
From now on, spring refers to a unidirectional spring coherent with the bending
angle between thigh and trunk.
Actuator Optimization for a Back-Support Exoskeleton 533
4 Discussion
Considering Table 1, it is clear that the optimization returning the best perfor-
mance in terms of energy is J1 . However, reducing energy to this extent, the
exoskeleton has to modify the user behaviour. As a consequence, this would
probably compromise user acceptance and make it extremely difficult to reach
such values of energy reduction without a training for using the exoskeleton and
adapting to the change in the muscle behaviour. On the other hand, the outcome
of J2 is an exoskeleton configuration that has a lower energy reduction but is
more coherent with the user biological behaviour. Indeed, the inverse behaviour
region is drastically reduced. Still focusing on J2 , the strong linearity between
τbio and the bending angle sounds as a reasonable explanation of the optimal
configuration being constituted only by a spring without a clutch for switch-
ing off assistance at specific lifting phases, as opposed to [5]. A further reason
why clutches are not part of the optimal configuration is that the considered
objective function tries to find the global optima for the specific lifting-objects
task. Thus, it might be of interest expanding the model so that it would find the
optima not only for a given task but also for others, as this is actually the main
limitation of passive exoskeletons [6]. Moreover, considering the stiffness of the
spring opens a discussion about designing an optimal exoskeleton for every user
or an exoskeleton that can adapt to the users’ needs, within a predefined range.
All things considered, we propose a solution based on a quasi-passive actua-
tion. Indeed, introducing controllable elements, the amount of assistance could
be controlled according to some feedback from the user, such as the muscle activ-
ity. Selecting the stiffness of the spring to a value within the range identified in
Sect. 3, by simply deciding when to engage the clutch it is possible to control the
spring elongation and, therefore, the amount of energy. Instead, controlling the
dis-engage of the quasi-passive element would make the exoskeleton transparent
when no assistance is required (e.g. switching between tasks). Hence, the use of
the clutch would allow to design an exoskeleton for many different users that not
only could be adjusted and controlled on-line, but also could prevent obstructing
the user in tasks different from material lifting.
5 Conclusion
This work shows that in the design of a (quasi-)passive exoskeleton, the choice of
the elements characteristics may result in an uncomfortable device usage for the
user. This can be prevented by accounting for the user’s behaviour in the selec-
tion of the components, e.g. in the optimization process. Defining such objective
functions, considering the specific task and given the strong linearity between
τbio and bending angle, the optimal actuators configuration is that constituted
by a spring. Based on this, the design of a quasi-passive exoskeleton could allow
on-line adjustments, reducing assistance or making the exoskeleton transparent
when needed.
534 T. Poliero et al.
References
Caldwell, D.G.: Rationale, implementation and evaluation of assistive strategies for
an active back-support exoskeleton. Front. Robot. AI 5, 53 (2018)
2. de Looze, M.P., Bosch, T., Krause, F., Stadler, K.S., OSullivan, L.W.: Exoskeletons
for industrial application and their potential effects on physical work load. Ergon.
59(5), 671–681 (2016)
3. Choi, I., Corson, N., Peiros, L., Hawkes, E.W., Keller, S., Follmer, S.: A soft, control-
lable, high force density linear brake utilizing layer jamming. IEEE Robot. Autom.
Lett. 3(1), 450–457 (2018)
4. Diller, S., Majidi, C., Collins, S.H.: A lightweight, low-power electroadhesive clutch
and spring for exoskeleton actuation. In: IEEE International Conference on Robotics
and Automation (ICRA), IEEE, pp. 682-689, May 2016
6. Baltrusch, S.J., van Dien, J.H., van Bennekom, C.A.M., Houdijk, H.: The effect of
a passive trunk exoskeleton on functional performance in healthy individuals. Appl.
Ergon. 72, 94–106 (2018)
Design of MobIle Digit Assistive System
(MIDAS): A Passive Hand Extension
Exoskeleton for Post Stroke Rehabilitation
Titus S. Hansen, Chris K. Bitikofer, Bahram E. Sobbi,

and Joel C. Perry(&)
Mechanical Engineering Department, University of Idaho,

Moscow, ID 83843, USA
{hans2993,biti6400,eila8850}@vandals.uidaho.edu,
jperry@uidaho.edu
Abstract. Stroke often causes flexor hypertonia as well as weakness of finger

extension. This limits functionality of the hand degrading independent ability to
perform upper limb activities of daily living (ADL’s). Hand rehabilitation post
stroke is vital to regaining functionality in the affected limb, leading to improved
independence and quality of living. In this paper the development of DigEx and
MIDAS passive arm orthoses are detailed. A quick-change cam system is
implemented featuring one-handed cam swapping. This provides the ability to
vary assistance levels to improve usability and independence for the user.
Pulleys and bearings are added to reduce friction caused by mechanical contacts
and material failure. Initial tests with the prototype are promising.
1 Introduction
Each year, approximately 795,000 people suffer a stroke. [1]. More than 70% of stroke
survivors are reported to experience limited functional recovery of their upper limb.
This diminishes quality of life and limits patients’ ability to perform activities of daily
living (ADL’s) [2]. Grasping and object manipulation are particularly important to
ADL’s, but flexor hypertonia as well as weakness of finger extension are common in
stroke patients, and impede the extension of the fingers and thumb during ADL tasks.
Incorporating an affected limb into daily living activities is an effective way to improve
motor function [3], but creating enough functionality of the limb for spontaneous use is
impractical in a clinical format due to inherent time constraints and accessibility issues.
A take-home orthosis could both expand access and improve patient outcomes by
incorporating an affected hand’s grasping functions into the completion of common
ADL tasks.
This work is supported by the University of Idaho and the Eunice Kennedy Shiver National
Institute of Child Health and Human Development of the National Institutes of Health under
Award Number K12HD073945. The content is solely the responsibility of the authors and does
not necessarily represent the official views of the University of Idaho or the NIH.

https://doi.org/10.1007/978-3-030-01887-0_104
536 T. S. Hansen et al.
The SaeboFlex is a commercially-available dynamic custom-fabricated wrist/hand

orthosis that assists stroke survivors in incorporating the involved hand functionally in
therapy. SaeboFlex utilizes extension springs attached to monofilaments to provide
extension assistance to patients’ fingers and thumb while supporting the wrist and
hand [4].
SaeboFlex has demonstrated rehabilitative success for numerous stroke survivors,
and its design can be improved to further increase effectiveness, comfort, and utility.
The SaeboFlex springs provide the most assistance when the fingers are flexed and the
least assistance when the fingers are extended. This assistance profile is not optimal
considering extension weakness rooted in an abnormal synergy is the primary cause of
poor grasping function of the hand. The SaeboFlex’s range of motion limits its ability
to assist in tasks that require small grips, such as using utensils, tools, and manipulating
small objects (Fig. 1). To further improve the utility of SaeboFlex for a wider range of
ADL’s, two successive redesigns termed the Digit Extension System (DigEx) and
MobIle Digit Assistive System (MI.D.A.S.), with a greater range of motion have been
implemented.
Short filaments
limiting ROM
Proportional spring
extension assistance
Fig. 1. The SaeboFlex Orthosis.
2 Methods
2.1 DigEx Redesign

DigEx focused on delivering a desirable non-linear force assistance profile and
increasing finger flexion range of motion (ROM) via the addition of a series wrapping
cam mechanism [5]. The mechanism’s geometry was synthesised programmatically
according to methods from [6]. It used a set of rubber springs, attached to a dual
wrapping cam that allows assistance force to be reduced as the springs extend and the
fingers flex.
Design concerns were made apparent by constructing and testing DigEx. The lack
of a bearings on the cam post and the monofilament/truss interface produced significant
friction. The extension of finger flexion ROM necessitated longer filaments and led to
acute bending angles at the contact point between the filament and the device. Extreme
bending angles resulted in the formation of cracks on the exterior surface of the
filaments which in turn increased sliding friction (Fig. 2).
Design of MobIle Digit Assistive System (MIDAS) 537
Finger Truss Monofilament/truss

acute bending point
Finger cap
in flexion
Finger cap Surface cracks

in extension
Fig. 2. Monofilament bending around the finger truss guide hole (left) and exterior surface
cracking (right) as a result of the tight bending during finger flexion.
2.2 MIDAS Redesign

MIDAS seeks to address remaining issues in the DigEx system while expanding its
capabilities. The device aims to achieve smooth motion via friction reduction. To
provide adjustable therapy, variable force profiles are achieved by applying a cam
quick-change system. MIDAS must be easily operable with one hand for the conve-
nience of patients. The physical profile of the device should be reduced by keeping
cables close to the device body.
2.3 Friction, Bending, and Profile Reduction

Friction is reduced in the finger trusses by using a pulley/anti-pinch guide truss to lower
resistance during finger flexion/extension. The trusses include snap-in ball-bearing
pulleys that reduce friction and provide strain relief by limiting filament bending to the
pulley radius. MIDAS lowers the profile of cables and spring attachments connecting
the cam system by introducing low friction PTFE tubes. These route filaments along
the forearm toward finger-trusses through a guide loom. A dual-bearing cam post
attachment transmits torsion with minimal shaft deflection while limiting contact
friction (Fig. 3, top).
2.4 Cam Quick-Change System

MIDAS adds a quick cam change system (Fig. 3, bottom) allowing the user to adjust
the assistance force profile. Cams may be switched to provide more or less assistance,
potentially increasing the therapy effectiveness.
3 Results
The MIDAS redesign (Fig. 4, top) has been implemented as a functioning passive
prototype that has potential to allow patients to perform ADLs that involve flexion and
extension of the fingers. In static weighted loading tests, MIDAS allows assistive
opening force to reduce with increased finger flexion up to a maximum distance,
d (Fig. 4, top), of about 86 mm. In comparison, the SaeboFlex’s opening force in-
creases with increased flexion to a maximum distance of about 53 mm.
538 T. S. Hansen et al.
Thumb Screw Locking
Keyed Shaft/Cam
Flanged, Shielded, Twin
Fig. 3. Key redesign components of the MIDAS system: (top, left to right) filament truss, PTFE
guide loom, keyed shaft, thumb screw locking cap, and bottom view of custom cam with square
keyway; (bottom) exploded view of cam quick-change system. The system enables one-handed
cam swapping using a slip-fit square key with a thumb screw cap.
Fig. 4. Complete MIDAS redesign (top) and comparison between assistive opening forces as
fingers are flexed using the MIDAS and SaeboFlex (bottom).
The MIDAS design based on the SaeboFlex orthosis has been implemented. It provides
a larger finger extension/flexion ROM while providing a desirable force assistance
profile. Pulley-bearing finger-trusses and PTFE tubing reduce friction in transmission
cabling. The PTFE guide also lowered the height profile near the wrist as compared to
DigEx and the original SaeboFlex. The cam quick change system performs to speci-
fication and can be changed out in less than one minute. Initial tests of the device are
promising with friction being noticeably lower than in previous versions. To meet
Design of MobIle Digit Assistive System (MIDAS) 539
target specification, the team wishes to do formal testing and final design modifications.
The design for the cam needs to be finalized in Solidworks and 3D printed for use.
Additionally, the quick release for the cam and filament need to be adjusted to relieve
tension from the cam and fingers.
Acknowledgments. Special thanks to Jeremiah Schroeder for developing the cam synthesis
scripts, the DigEx design team, Bridger Hopkins and Jeremiah Schroeder, for their initial
redesign efforts, and Saebo Inc. for the donation of the SaeboFlex.
References
1. ONU: World population, ageing, Suggest. Cit. United Nations, Dep. Econ. Soc. Aff. Popul.
Div. (2015). World Popul. Ageing, vol. United Nat, no. (ST/ESA/SER.A/390), p. 164 (2015)
2. CDC: Stroke Facts | cdc.gov. https://www.cdc.gov/stroke/facts.htm. Accessed 15 Dec 2017
3. Dobkin, B.H.: Rehabilitation after stroke. N. Engl. J. Med. 352(16), 1677–1684 (2005)
4. Saebo: SaeboFlex / SaeboReach Details | Saebo. https://www.saebo.com/saeboflex-
saeboreach-details/. Accessed 15 Dec 2017
5. Schroeder, J.S., Perry, J.C.: Development of a series wrapping cam mechanism for energy
transfer in wearable arm support applications. In: 2017 International Conference on
Rehabilitation Robotics (ICORR), 17 July 2017, pp. 585–590. IEEE
6. Tidwell, P.H., Bandukwala, N.N., Dhande, S.G., Reinholtz, C.F., Webb, G.G.: Synthesis of
Wrapping Cams. ASME J. Mech. Des. 116(2), 634–638 (1994). https://doi.org/10.1115/1.
2919425
Author Index
A Bertomeu-Motos, Arturo, 386

Abdikadirova, Banu, 18 Bessler, Jule, 123
Acer, Merve, 8 Bettinelli, Luca, 465
Afzal, Muhammad Raheel, 455 Bianchi, Matteo, 370, 440, 490
Aguirre-Ollinger, Gabriel, 294 Bicchi, Antonio, 370, 520
Ajoudani, Arash, 281, 520 Bidard, Catherine, 123
Allotta, Benedetto, 440 Bitikofer, Chris K., 535
Alonso, Francisco J., 257 Bizovičar, Nataša, 91
Alonso-Ramos, Víctor, 234 Blanco, Andrea, 386, 515
André, Paulo, 470 Bleuler, H., 75
Ang Jr., Marcelo H., 401 Bonnet, Vincent, 65
Antunes, Paulo, 470 Bonomo, Fabio, 370
Aoustin, Y., 445 Bordron, O., 445
Aprigliano, Federica, 110 Bos, Danny Plass-Oude, 391
Arami, Arash, 361 Bottenberg, E., 53
Arquilla, Matteo, 304, 334, 361 Bou, Julia, 234
Asselin, Pierre K., 309, 314 Bouri, M., 75, 85
Augustine, Jonathan, 309, 314 Brinks, G. J., 53
Azorín, Jose M., 206 Brug, T. J. H., 334
Azorín, José M., 289 Bruijn, S. M., 229
Burdet, Etienne, 361
B Buurke, Jaap H., 123
Babič, Jan, 244, 450
Bae, Juwhan, 430 C
Baeyens, Jean-Pierre, 96 Caicedo, Pablo E., 201
Bai, Shaoping, 3, 180 Caimmi, M., 356
Ballen, Felipe, 160 Calanca, Andrea, 381, 465
Baltrusch, S. J., 229 Caldwell, Darwin G., 170, 219, 351, 365, 381,
Barberi, Federica, 110 435, 510, 530
Barios, Juan, 386 Carbonaro, Nicola, 13
Barrance, Peter, 309 Carpentier, É. Le, 445
Bastos, T., 196 Carpinella, I., 356
Baten, Chris T. M., 391 Casas, Diego, 160
Baud, R., 75 Catalán, José M., 386, 515
Beccai, L., 53 Catalano, Manuel G., 370, 520

https://doi.org/10.1007/978-3-030-01887-0
542 Author Index
Cattaneo, D., 356 F

Cesini, Ilaria, 105, 115 Falisse, Antoine, 267
Cevzar, Mišel, 244 Feng, Hui, 401
Chakarov, Dimitar, 410 Ferguson, Peter Walker, 276
Chen, Baojun, 224 Ferrarin, M., 356
Cheng, Hsiao-Ju, 294 Filosa, Mariangela, 105
Chiaradia, Domenico, 35, 415 Fiorini, Paolo, 465
Christensen, S., 180 Fontano, M. Isabel, 234
Cifuentes, Carlos A., 160 Font-Llagunes, Josep M., 257
Cirnigliaro, Christopher, 314 Forrest, Gail F., 309, 314
Ciullo, Andrea S., 520 Fraize, Julian, 480
Conti, R., 132, 147 Franklin, David, 262
Convens, Bryan, 460, 485 Frisoli, Antonio, 35, 415
Crea, Simona, 105, 115, 224 Frizera, Anselmo, 155, 206, 470
Crispel, Stein, 165, 460, 485 Frizera-Neto, A., 196
Croci, Eleonora, 100 Furnémont, Raphaël, 165
Cuadrado, Javier, 257 Furth, Mirjam, 480
D G
Dacal-Nieto, Angel, 234 Garbarini, Erica, 309, 314
Davalli, Angelo, 110 García, José V., 515
de Eyto, Adam, 44 García, Pablo López, 460, 485
De Groote, Friedl, 267 García-Aracil, Nicolás, 386, 515
de Jonge, Benjamin, 370 Gassert, Roger, 100, 370
De Keersmaecker, Emma, 96 Gil-Agudo, Ángel, 206
de Looze, M. P., 505 Giovacchini, F., 132
De Momi, Elena, 170, 281, 435 Goh, Aaron Jing Yuan, 401
de Vries, Wiebe, 391 Goljar, Nika, 91
del-Ama, Antonio J., 206 Gómez, Daniel, 160
Delisle-Rodriguez, D., 196 Gonçalves, Sérgio B., 299
Dell’Agnello, Filippo, 115 Graf, Eveline, 525
Desloovere, Kaat, 267 Grazi, Lorenzo, 224
Di Giovanni, R., 356 Gregoor, Wouter, 361
Di Natali, Christian, 351 Groeneveld, R., 53
Di Pardo, Massimo, 219 Gruppioni, Emanuele, 110
Díez, Jorge A., 386, 515 Guncan, Berkay, 500
Dimapasoc, Brando, 276 Gunz, Daniel, 80
Dimo, Eldison, 465 Gutiérrez, D., 187
Ding, Ye, 142 Gutierrez-Martínez, Josefina, 192
Disselhorst-Klug, Catherine, 192
Domingues, Maria F., 470 H
dos Santos, Wilian M., 175 Haarman, Claudia J. W., 475
Duarte, Jaime, 80 Haddadin, Sami, 370
Durandau, Guillaume, 325 Hamker, Fred H., 396, 420
Dzeladini, F., 75 Hansen, Titus S., 535
Dzeladini, Florin Florin, 361 Harant, Monika, 249
Hartigan, Bernard, 44
E Hekman, Edsko E. G., 475
Eizad, Amre, 455 Henderix, Stieven, 96
Ellena, D., 356 Henze, Bernd, 344
Emmens, A., 334 Herzog, Walter, 262
Emmens, Amber, 361 Hidalgo Romero, Andres F., 525
Erkens, L., 53 Hocaoglu, Elif, 28
Ezquerro, Santiago, 386 Hong, Sin Wai, 406
Author Index 543
Houdijk, H., 229 Lefeber, Dirk, 165, 239, 460, 485

Hu, Xinyao, 329 Lefeber, Nina, 96
Huang, Hsien-Yung, 361 Leitão, Cátia, 470
Huneau, C., 445 Li, Miao, 401
Husain, Syed R., 309, 314 Lichard, P., 75
Hwang, Seokjin, 430 Lima, J., 196
Linssen, Jeroen, 391
I Lippi, V., 321
Iáñez, Eduardo, 289 Liu, Charles, 142
Ibarra Zannatha, Juan M., 192 Logar, Grega, 23
Ibarra-Zannatha, J. M., 187, 210 López, David, 515
Ijspeert, A. J., 304 López-García, Pablo, 165
Ijspeert, Auke, 75, 361 Lorenzini, Marta, 281
Islam, M., 180 Low, Jin Huat, 401
Islam, Muhammad R. U., 3 Luft, Andreas R., 370
Ivanic, Zoran, 23 Lugris, Urbano, 257
Luo, Chuang, 329
J Lyu, Sung-Ki, 455
Jakubowitz, Eike, 370
Jamšek, Marko, 244, 450 M
Jonkers, Ilse, 267 Mallat, Randa, 65
Just, Fabian, 80 Manchola, Miguel, 160
Mancini, Lorenza, 219
K Marques, Carlos, 155, 470
Kainz, Hans, 267 Martini, Elena, 105, 115
Kalli, Kyriacos, 470 Masciullo, Marcella, 304, 334, 361
Kansizoglu, Ahmet Talha, 70 Masia, Lorenzo, 35, 415
Kapci, Mehmet F., 495 Masood, Jawad, 234
Karavas, Nikolaos, 142 Masud, N., 180
Kawamura, Kazuya, 425 Matjačić, Zlatko, 91
Kerckhofs, Eric, 96 Mattsson, P., 180
Khalil, Mohamad, 65 Mazzolai, Barbara, 49
Kim, Gyoosuk, 430 Mazzoni, Alberto, 110, 115
Kim, Jinsoo, 142 Meijneke, Cor, 304, 361
Kim, Myunghee, 142 Mergner, T., 321
Kim, Wansoo, 281 Micera, Silvestro, 110
Kingma, I., 505 Michielsen, Marc, 96
Knezevic, Steven, 309, 314 Millard, Matthew, 249, 262
Konstantinos, Kostas, 18 Miller-Jackson, Tiana M., 406
Koopman, Axel S., 229, 239, 505 Mohammed, Samer, 65
Kuindersma, Scott, 142 Molinari, Marco, 304, 334, 361
Kwon, Chilyong, 430 Mombaur, Katja, 249
Momeni, Kamyar, 309, 314
L Mondini, Alessio, 49, 53
Lambercy, Olivier, 370 Montini, Valeria, 127
Lancini, Matteo, 127 Moreno, Juan C., 206
Lanotte, Francesco, 224 Mouzo, Francisco, 257
Lassen, Aske E. B., 123 Munera, Marcela, 160
Lazzaroni, Maria, 219, 435 Munih, Marko, 23
Leal-Junior, Arnaldo G., 155, 470
Lee, Hosu, 455 N
Lee, Jongwon, 430 Näf, Matthias B., 239
Lee, Sangjun, 142 Narayan, Ashwin, 294
544 Author Index
Nassour, John, 396, 420 Rietman, Hans S., 475

Natividad, Rainier F., 406 Roa, Javier O., 206
Nielsen, Kurt, 123 Roa, Maximo A., 344
Rocon, Eduardo, 391, 525
O Rodrigues, Thomaz, 206
O’Sullivan, L., 180 Rodríguez, Luís E., 201
O’Sullivan, Leonard W., 44 Rodriguez-Guerrero, Carlos, 239
Oddo, Calogero Maria, 115 Rodríguez-Ugarte, Marisol, 289
Oddo, Calogero, 105 Romero-Avila, Elisa, 192
Olenšek, Andrej, 91 Ronsse, Renaud, 271
Orozco-Soto, Santos M., 187, 192, 210 Rosen, Jacob, 276
Ortiz, Jesús, 44, 170, 219, 351, 365, 381, 435, Russo, Stefania, 13
510, 530
Ortiz, Mario, 289 S
Ortlieb, A., 75 Saccares, L., 132
Ott, Christian, 344 Sacchetti, Rinaldo, 110
Sadeghi, Ali, 49
P Saenz, José, 123
Papageorgiou, Eirini, 267 Saerens, Elias, 460, 485
Park, Daegeun, 510 Sanchez, Andres Hidalgo, 391
Pasinetti, Simone, 127 Sánchez-Villamañán, M. C., 137
Pasquini, Guido, 490 Sansoni, Giovanna, 127
Peñaloza, Alicia Meneses, 192 Santello, Marco, 370
Peng, Dongsheng, 329 Šarabon, Nejc, 249
Pérez-Sanpablo, A. I., 210 Sartori, Massimo, 325
Perez-Sanpablo, Alberto I., 192 Schaake, Leendert, 123, 391
Perry, Joel C., 535 Scherly, Daniel, 391
Petrič, Tadej, 244 Schumacher, Christian, 339
Piazza, Cristina, 370 Secciani, Nicola, 440, 490
Pinto-Fernandez, D., 147 Sekine, Masashi, 425
Pisotta, I., 304, 334 Sendur, Polat, 70
Pisotta, Iolanda, 361 Serpelloni, Mauro, 465
Pitto, Lorenzo, 267 Serrano, Daya, 160
Poliero, Tommaso, 381, 530 Seyfarth, Andre, 339
Pons, José L., 132, 137, 147, 206, 375 Sharbafi, Maziar A., 339
Pontes, Maria José, 155, 470 Shen, Yang, 276
Power, Valerie, 44, 180 Shirota, Camila, 100
Prasanna, Sahana, 115 Sierra, Wilson A., 201
Prattichizzo, Domenico, 370 Silva, Miguel T., 299
Simonetti, Davide, 80
Q Siqueira, Adriano A. G., 175
Qu, Xingda, 329 Sluiter, Victor, 361
Quinto, Luís P., 299 Sobbi, Bahram E., 535
Solaro, C., 356
R Solazzi, Massimiliano, 415
Rafique, S., 180 Spigler, Giacomo, 105, 115
Ramanujam, Arvind, 309, 314 Sposito, Matteo, 170, 219
Ramírez-Moreno, M. A., 187 Spungen, Ann M., 309, 314
Rauter, Georg, 80 Sreenivasa, Manish, 249
Rengifo, Carlos F., 201 Stadler, Konrad, 391
Ricciardi, Emiliano, 370 Staman, Kyrian, 39
Ridolfi, Alessandro, 440, 490 Sun, Yi, 401
Riener, Robert, 80 Swinnen, Eva, 96
Author Index 545
T Vera-Bustamante, P., 210

Tagliamonte, Nevio Luigi, 304, 334, 361 Verstraten, Tom, 165, 460, 485
Tamburella, Federica, 304, 334, 361 Vicario, Rudy, 465
Taxis, Domitilla, 115 Vicentini, Federico, 123
Theodosiou, Antreas, 470 Villa-Parra, A. C., 196
Thorsteinsson, Freygardur, 361 Vitiello, Nicola, 105, 115, 132, 147, 224
Tognetti, Alessandro, 13 Voilqué, Anthony, 234
Tonis, Frederick, 370
Torricelli, D., 132, 137, 147, 375 W
Tosatti, L. Molinari, 356 Walsh, Conor, 58, 142
Totaro, M., 53 Ward, Tomas, 370
Toxiri, Stefano, 170, 219, 365, 381, 435, 505, Wesseling, Mariska, 267
530 Witteveen, Heide, 361
Tsveov, Michail, 410 Witteveen, Juryt, 391
Wu, A. R., 304
U Wu, Amy, 361
Ugurlu, Barkan, 70
Unal, Ramazan, 495, 500
X
Xiloyannis, Michele, 35, 415
V
Xu, Kun, 3
Vaghani, Sidhdharthkumar, 420
Xydas, Evagoras, 18
Valette, R., 334
van Asseldonk, E. H. F., 334
van Asseldonk, Edwin, 361 Y
van Bennekom, C. A. M., 229 Yeow, Chen-Hua, 406
van der Kooij, H., 304, 334 Yeow, Raye Chen Hua, 401
van der Kooij, Herman, 325, 361 Yildirim, Mehmet C., 70
Van Der Kooij, Herman, 370 Yıldız, Adnan Furkan, 8
van der Kooij, Herman, 39, 475 Yoon, Jungwon, 455
van Dieën, J. H., 229, 505 Yu, Haoyong, 294
van Oort, Gijs, 361 Yu, Wenwei, 425
Van Rossom, Sam, 267
Vanderborght, Bram, 165, 239, 460, 485 Z
Vannetti, Federica, 440, 490 Zadravec, Matjaž, 91
Veale, Allan J., 39 Zanotto, Damiano, 480
Venev, Pavel, 410 Zhao, Guoping, 339
Veneva, Ivanka, 410 Zhao, Kristin D., 370

2019 Book WearableRoboticsChallengesAndT PDF

Uploaded by

Copyright:

Available Formats

2019 Book WearableRoboticsChallengesAndT PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2019 Book WearableRoboticsChallengesAndT PDF

Uploaded by

Copyright:

Available Formats

Biosystems & Biorobotics

About the Cover

More information about this series at http://www.springer.com/series/10421

ISSN 2195-3562 ISSN 2195-3570 (electronic)

Library of Congress Control Number: 2018956723

© Springer Nature Switzerland AG 2019

Wearable Sensors for Robotic Exoskeletons

Soft Wearable Robots

Application of a User-Centered Design Approach to the Development

Subject-Centered Based Approaches for Controlling Wearable Robots

Robotic and Neuroprosthetic Balance Management Approaches

Benchmarking Wearable Robots

What Do People Expect from Benchmarking of Bipedal Robots?

Flexible and Transparent Technologies for Innovative

Wearable Robotics for Rehabilitation and Assistance

Wearable Robotic Solutions for Factories of the Future

Industrial Wearable Exoskeletons and Exosuits Assessment Process . . . 234

Human Modeling and Simulation for Neurorehabilitation Engineering

Smart Human-Machine Systems for Lower-Limb Assistance

Biorobotics Approaches to Understand and Restore Human Balance

Training Balance Recovery in People with Incomplete SCI Wearing

Exoskeleton Research in Europe

EUROBENCH: Preparing Robots for the Real World . . . . . . . . . . . . . . 375

Optimizing Body Thickness of Watchband-Type Soft Pneumatic

Pushing the Limits: A Novel Tape Spring Pushing Mechanism

Actuator Optimization for a Back-Support Exoskeleton:

Muhammad R. U. Islam1(&), Kun Xu2, and Shaoping Bai1

Abstract. Human intention decoding is a primary requirement to control an

© Springer Nature Switzerland AG 2019

3 Elbow Exoskeleton Design

4.1 Hardware Setup

4.2 Real-Time Estimation

5 Experiments and Results

5.1 Joint Angle Estimation

Table 1. Error calculated between actual and estimated joint angle

5.2 E-EXO Control

Fig. 2. Joint angle estimation

velocity signal, which is controlled through ESCON’s (motor driver) built-in PI

Fig. 3. E-EXO control structure

Fig. 4. E-EXO position control results using estimated joint angle

Acknowledgment. The reported work is partially supported by EU-AAL Joint Programme

Merve Acer(&) and Adnan Furkan Yıldız

Mechanical Engineering Department, İstanbul Technical University,

Abstract. Wearable tactile sensors are signiﬁcant in biomedical robotic

© Springer Nature Switzerland AG 2019

2 Sensor Description and Test SetuP

The tactile sensor is composed of 9 PZT (PSI-5H4E) elements positioned as 3 3

High input impedance characteristic of piezoelectric material requires an opera-

3 Force Localization Estimation

3.2 Learning Algorithm

3.3 Estimation Results

Table 1. Estimation results

Stefania Russo2 , Nicola Carbonaro1 , and Alessandro Tognetti1(B)

Abstract. We present the development of an innovative stretchable tac-

stretchable and conformable fashion. To detect the force interaction in multiple

2 Materials and Methods

The presence of diﬀerent materials or wires embedded in a tactile sensor is usu-

Fig. 1. Block diagram of our EIT experimental platform.

Evagoras Xydas(&), Banu Abdikadirova, and Kostas Konstantinos