Indoor Navigation With A Swarm of Flying Robots
Indoor Navigation With A Swarm of Flying Robots
Indoor Navigation With A Swarm of Flying Robots
Timothy Stirling and James Roberts and Jean-Christophe Zufferey and Dario Floreano
Abstract Swarms of flying robots are promising in many mapscollectively referred to as Global Information. Previ-
applications due to rapid terrain coverage. However, there ously we presented an aerial swarm search strategy using
are numerous challenges in realising autonomous operation a Robotic Sensor Network paradigm within 3-D simulation
in unknown indoor environments. A new autonomous flight
methodology is presented using relative positioning sensors in [1]. A network of robots embedded in the environment with
reference to nearby static robots. The entirely decentralised ap- local sensing, processing and communication can solve the
proach relies solely on local sensing without requiring absolute complex navigation task without global information. This
positioning, environment maps, powerful computation or long- paper extends our prior simulation work [1] by presenting
range communication. The swarm deploys as a robotic network a new methodology for autonomous flight across the span
facilitating navigation and goal directed flight. Initial validation
tests with quadrotors demonstrated autonomous flight within of the robotic network validated with real flight tests. On-
a confined indoor environment, indicating that they could board relative positioning sensors are utilised with an ap-
traverse a large network of static robots across expansive proach that is computationally simple and robust to varied
environments. illumination. The related work is discussed next and then
the autonomous flight and navigation behaviours are detailed
I. I NTRODUCTION before presenting the results. Finally, limitations and future
Swarms of flying robots are promising for applications work are discussed.
such as search because they can rapidly travel above obsta-
II. R ELATED W ORK
cles, have elevated sensing, and facilitate task parallelisation
and redundancy [1], [2]. However, indoor flying robots have Previous research in indoor flying swarms usually used
severely limited on-board sensing and processing, and GPS external tracking systems in-place of GPS, e.g. [8]. However,
is unreliable due to attenuated signals. Therefore, Swarm on-board sensing is desirable for operation in unknown envi-
Intelligence [3] is promising in simplifying control and ronments. One approach is to use laser scanners to estimate
reducing sensing and processing requirements. This work the robot pose and motion, e.g. [9] and [10] demonstrated
is based on quadrotors [4] which can take-off and land indoor navigation of a quadrotor, but required significant off-
in small spaces, hover over targets and have high ma- board processing. However, such approaches can fail in large
noeuvrability. However, rotorcraft experience drift due to homogeneous environments such as long corridors, or near
turbulence and imbalances [2], which is frequently controlled glass [11]. Additionally, suitable laser scanners only operate
using absolute positioning from external tracking systems in 2-D, but the robots move in 3-D, leading to failure if there
[5], unavailable in unknown environments. On-board sensing are large variations in vertical environment structure [11].
approaches include illumination-dependent cameras or laser Alternatively, using illumination dependent cameras, in-
scanners that can fail in homogeneous environments. These door navigation using a priori image-databases was demon-
approaches are computationally expensive, often requiring strated in [12] and [13]. However, such approaches frequently
off-board processing which delays control feedback due to require off-board processing, and depend upon suitable
transmission times and requires reliable long-range commu- features, so artificial features are often pre-installed [14].
nication. Additionally, navigation usually requires absolute Although recent work towards using on-board processing
positioning or localisation using environment maps [6]. Maps is promising [14], [15], vision-based approaches have many
may be unknown a priori and their online creation requires undesirable properties. They can suffer undesirable control
significant processing. Such approaches also do not scale to feedback caused by errors in pose estimation introducing
large environments [7] or swarms. oscillations and increasing platform motion. This increases
In summary, alternative strategies are required to enable image-blur, further degrading pose estimation and increasing
indoor flying swarms without absolute positioning, long- platform motion, which escalates into an uncontrollable
range communication, centralised processing or environment feedback loop [16]. Similarly, images from flying robots
suffer from vibrations [12]. Similar problems affect optic-
We thank Francois Gervaix and Julien Brahier (HEIG-VD) for flow approaches [17], [18], which require significant illumi-
the tracking system and Maja Varga who assisted in the paper. nation and contrast. Therefore, vision-based methods are not
This work is part of the Swarmanoid Project, a Future Emerging
Technologies (FET IST-022888) project funded by the EC. All work entirely satisfactory.
was conducted at the Laboratory of Intelligent Systems (LIS), Ecole Using large blimps, [19] achieved basic swarm behaviours
Polytechnique Federale de Lausanne (EPFL), Lausanne, Switzerland. such as leader-following and aggregation using on-board
timothy.stirling@gmail.com, james@iuavs.com,
jean-christophe.zufferey@epfl.ch, infrared sensors. However, blimps are susceptible to distur-
dario.floreano@epfl.ch bances and their large size ( 1.0 m) makes them unsuitable
for many applications and environments.
To summarise, previous approaches to indoor flight have
many limitations, requiring pre-installed sensors or land-
marks, appropriate illumination, or off-board processing, etc.
Moreover, prior navigation methods also required global
information such as maps. An alternative navigation method
is to use robotic sensor networks, e.g., a pre-deployed sensor
network guided outdoor flying robots in [20]. Instead of
using pre-deployed networks, the robots themselves can be
used as sensor nodes [21]. Various approaches to deploying
robot sensor networks exist but none are suitable for indoor
flying robots. The most common are based on attraction
and repulsion forces between robots, termed Social Poten-
tial Fields (SPFs) [22]. However, complex tuning of the
force laws is required and determining the parameters for a
desired group behaviour is computationally infeasible [22].
Therefore, in previous work we presented a new approach
suitable for swarms of flying robots [1], which was analysed Fig. 1. Beacons provide simple navigation signals to flying explorers.
within a 3-D dynamics simulator [23]. In this paper, the
autonomous flight behaviours are developed and tested on
flying robots, validating the proposed swarm navigation and
search strategy.
III. M ETHODS
[16] S. Ahrens, D. Levine, G. Andrews, and J. How, Vision-based guid- billa, N. Mathews, E. Ferrante, G. D. Caro, F. Ducatelle, T. Stirling,
ance and control of a hovering vehicle in unknown, GPS-denied A. Gutierrez, L. Gambardella, and M. Dorigo, ARGoS: a modular,
environments, in Proceedings of the International Conference on multi-engine simulator for heterogeneous swarm robotics, in Proceed-
Robotics and Automation (ICRA 09). Piscataway: IEEE Press, 2009, ings of the International Conference on Intelligent Robots and Systems
pp. 26432648. (IROS 11). Los Alamitos: IEEE Press, 2011, pp. 50275034.
[17] J.-C. Zufferey, A. Beyeler, and D. Floreano, Optic Flow to Steer and [24] J. Roberts, J.-C. Zufferey, and D. Floreano, Energy management for
Avoid Collisions in 3D, in Flying Insects and Robots, D. Floreano, J.- indoor hovering robots, in Proceedings of the International Confer-
C. Zufferey, M. V. Srinivasan, and C. Ellington, Eds. Berlin: Springer, ence on Intelligent Robots and Systems (IROS 08). Piscataway: IEEE
2009, pp. 2950. Press, 2008, pp. 12421247.
[18] W. E. Green and P. Oh, Optic flow based collision avoidance on [25] J. Roberts, T. Stirling, J.-C. Zufferey and D. Floreano, 3-D relative
a hybrid MAV, IEEE Robotics and Automation Magazine, vol. 15, positioning sensor for indoor flying robots, Autonomous Robots
no. 1, pp. 96103, 2008. (forthcoming), 2012.
[19] C. Melhuish and J. Welsby, Gradient ascent with a group of minimal- [26] D. Gurdan, J. Stumpf, M. Achtelik, K. M. Doth, G. Hirzinger,
ist real robots: Implementing secondary swarming, in Proceedings of and D. Rus, Energy-efficient autonomous four-rotor flying robot
the International Conference on Systems, Man and Cybernetics, vol. 2. controlled at 1 kHz, in Proceedings of the International Conference
Piscataway: IEEE Press, 2002, pp. 509514. on Robotics and Automation. Piscataway: IEEE Press, 2007, pp.
[20] P. Corke, R. Peterson, and D. Rus, Localization and navigation as- 361366.
sisted by networked cooperating sensors and robots, The International [27] Q. Li, M. D. Rosa, and D. Rus, Distributed algorithms for guiding
Journal of Robotics Research, vol. 24, no. 9, pp. 771786, 2005. navigation across a sensor network, in Proceedings of the 9th annual
[21] A. Howard, M. J. Mataric, and G. S. Sukhatme, Mobile sensor international conference on Mobile computing and networking. New
network deployment using potential fields: A distributed, scalable York: ACM, 2003, pp. 313325.
solution to the area coverage problem, in Proceedings of the 6th [28] M. Kovac, J. M. Germann, C. Harzeler, R. Siegwart, and D. Floreano,
Distributed Autonomous Robotic Systems (DARS 02), vol. 5. Berlin: A perching mechanism for micro aerial vehicles, Journal of Micro-
Springer, 2002, pp. 299308. Nano Mechatronics, vol. 5, no. 34, pp. 7791, 2009.
[29] J. Roberts, T. Stirling, J.-C. Zufferey, and D. Floreano, 2.5D infrared
[22] J. Reif and H. Wang, Social potential fields: A distributed behavioral
range and bearing system for collective robotics, in Proceedings of
control for autonomous robots, Robotics and Autonomous Systems,
the International Conference on Intelligent Robots and Systems (IROS
vol. 27, no. 3, pp. 171194, 1999.
09). Piscataway: IEEE Press, 2009, pp. 36593664.
[23] C. Pinciroli, V. Trianni, R. OGrady, G. Pini, A. Brutschy, M. Bram-