Mechatronics 59 (2019) 104–114

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Motion control of joystick interfaced electric wheelchair for improvement

of safety and riding comfort☆
Jung Hyun Choi, Younghun Chung, Sehoon Oh∗
Department of robotics Engineering, DGIST, 42988, Techno Jungang-daero 333, Hyeonpung-myen, Dalseong-gun, Daegu, Republic of Korea

a r t i c l e i n f o a b s t r a c t

Keywords: Nowadays powered wheelchairs are widely utilized to help people’s locomotion. However, many commercialized
Electric wheelchair powered wheelchairs still provide unsatisfactory ride quality, which is mainly due to not-well-designed control
Joystick operation algorithm. To address this issue, this paper proposes a motion control algorithm that can improve the safety and
Locomotion mode
riding comfort of the user and can be easily implemented on a powered wheelchair.
Yaw moment observer
To design the controller in an intuitive way, locomotion mode of a wheelchair is suggested at first, which consists
Kinematic Kalman Filter
Hall sensor of the longitudinal mode and the rotational mode. Then, the motion reference generator, wheelchair motion
observer and control algorithms are designed in this locomotion mode.
Moreover, this locomotion mode-based motion control does not require encoders, which makes the implementa-
tion of the algorithm affordable for many commercialized wheelchairs.
The detail of the controller and state observer is studied, and experimental results to verify the validity of the
proposed controller and the state observer are explained.

1. Introduction On the other hand, the power-assisted wheelchair has emerged as a

hybrid type of a powered wheelchair and a manual wheelchair, which
Wheelchairs are widely utilized among people who have difficulty has similar appearance to a manual wheelchair but can provide assistive
or weakness using their legs. Manual wheelchairs which are operated motor power responding to the users torque [8,9]. In addition, there are
by the users arms are the most widely utilized type of wheelchair that other types of powered wheelchairs that have different types of struc-
increase mobility for these people, since they consist of simple mecha- ture to perform specific tasks such as climbing stairs[10] or being trans-
nism, and thus are easy to use and affordable for many people. How- formed into an upright shape for daily conversations with other people
ever, recently, the powered wheelchair has been developed to help [11].
wheelchair users by providing electric power for driving. Although the Among various powered wheelchairs, the joystick type wheelchair
powered wheelchair has more complicated structure and should be han- and power-assist wheelchair are successfully commercialized and inten-
dled with care, its locomotion requires less human power than the man- sively researched. For the joystick type wheelchairs, the interface char-
ual wheelchair. acteristics of the joystick have been deeply researched since the joystick
For this reason, nowadays the powered wheelchair has prevailed has an important role to convert humans intention to the wheelchair
remarkably, and various types of powered wheelchairs are developed. controller [12–14].
First of all, the powered wheelchair operated using a joystick is the most Meanwhile, the power-assisted wheelchair utilizes force/torque sen-
successfully commercialized one [1–3], which can be considered a type sors on the push-rim to detect the users intention and provide appropri-
of electric vehicle that is controlled by a joystick. Thanks to the benefit ate motor torques based on the measured human forces [15–17]. Most
of convenience and large power, they have been widely utilized for out- power-assisted wheelchairs use two force/torque sensors to fully detect
door mobility. However, it is hard to manipulate it in a narrow space the users intention, and even using one torque sensor is useful enough
like the office due to the difficulty of steering using the joystick. To this to detect the intention with well-developed detection algorithm [18]. In
end, an electric wheelchair which has omni-wheel [4,5] has been also addition to the user intent detection, the motion states of the wheelchair
developed aiming at holonomic omnidirectional driving [6,7]. need to be observed, and studies have been conducted for this by com-

☆
This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIP) (NRF-2016R1A2B4016163) and
was supported by the Robot industry fusion core technology development project through the Korea Evaluation Institute of Industrial Technology (KEIT) funded by
the Ministry of Trade, Industry and Energy of Korea (MOTIE) (no. 10080547)
∗
Corresponding author.
E-mail address: sehoon@dgist.ac.kr (S. Oh).

https://doi.org/10.1016/j.mechatronics.2019.03.005
Received 7 September 2018; Received in revised form 11 February 2019; Accepted 17 March 2019
Available online 28 March 2019
0957-4158/© 2019 Elsevier Ltd. All rights reserved.
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

bining various motion sensors such as acceleration sensor or gyro sensor The contribution points of this research to design a motion controller
[19,20]. for a joystick interfaced wheelchair are given as follows.
After obtaining the users intention, it is necessary to design the con-
trol of motor torque for proper wheelchair locomotion. In particular, • Definition of locomotion mode for wheelchair motion.
the control of power-assisted wheelchair has been vehemently inves- • Suggestion of locomotion mode-based motion reference generation
tigated: several studies addressed tipping over problem of the power method with joystick interface.
assisted wheelchair [21], and others tackled unintended turning of the • Suggestion of locomotion mode-based wheelchair motion estimation
wheelchair caused by the friction of both wheels traveling on different algorithm without using an encoder.
roads [22–25]. These approaches aimed to overcome dangerous and un- • Suggestion of motion control algorithm based on the intuitive loco-
desired situations and provide more secure wheelchair driving to the motion mode.
user by removing disturbance which perturbs desired wheelchair loco- • Control system that can be easily implemented in off-the-shelf
motion. Other advanced wheelchair motion controllers are designed to wheelchairs without large hardware modification.
attenuate humans load to drive, particularly on slopes. The study in
The definition of the intuitive locomotion mode of wheelchair mo-
[26] developed acceleration control to reduce the required torque for
tion is defined, and the motion reference generation with regard to the
human, and [27] applied gravity compensation control also to reduce
locomotion mode is proposed in Section 2. In Section 3, the state ob-
the load. These controllers can enhance user convenience, and some
server to estimate the wheelchair motion is designed, and control al-
study even utilized the gravity for power regeneration [28].
gorithm to achieve the reference motion is proposed. In Section 4, the
In contrast to the power assisted wheelchair, there is not much
experimental results of the motion estimation and the reference track-
research related to driving control of a joystick interfaced electric
ing are presented to verify the effectiveness the proposed control system.
wheelchair. Conventional research has worked on the analysis the joy-
Conclusions of the study are presented in Section 5.
stick interfaced wheelchair motion or its driving performance. For ex-
ample, study in [29] analyzed the reverse directional stability of the
wheelchair investigating the effect of the caster rotation. Chen and 2. Motion reference generation for a powered wheelchair using
Agrawal [30] introduced a force field of the joystick to assist users joy- joystick based on intuitive locomotion mode
stick motion to enhance user interface. Research in [31] suggested a
collaborative control methodology considering comprehensive human 2.1. System overview
factor to help user motion. Other research [32,33] developed controllers
for the joystick type wheelchair to synchronize two wheels and thus to Fig. 1 shows the powered wheelchair utilized in this system, which
keep its heading angle. However, not any of these studies worked on is equipped with two BLDC motors (including Hall sensors) and a joy-
motion control of joystick wheelchair to improve the quality of driving. stick. The main controller has BLDC motor driver circuit, a digital signal
For many wheelchair users, the driving quality of a wheelchair is processor (TI TMS-320C28346) and two built-in motion sensors (a gy-
an important factor, but there are issues to be addressed for this, e.g., roscope and accelerometer). Two lead-acid batteries are located under
jerky acceleration/deceleration and difficulty in manipulating the head- the seat. The connection between the motors, controller, sensors, bat-
ing direction. Wheelchair users complain about riding comfort when the teries and the joystick is depicted in Fig. 1(b). Two BLDC motors are
acceleration/deceleration is very jerky, and also they may be subject to connected to the real wheels through ring gears, and the front gears
danger, particularly unless the heading angle of the wheelchair is prop- consist of passive casters.
erly controlled. These issues can be addressed by a well-designed motion All the components including motors, joystick, wheels and batter-
controller of a wheelchair. ies (except the controller) are off-the-shelf, which means no additional
To this end, this research proposes a novel wheelchair motion con- modification is added to these components. Only the motor driver of the
troller, which can reduce the jerk peaks and keep its heading angle off-the-shelf wheelchair is replace with a new control board in Fig. 2,
against various road conditions. The proposed controller, at first, de- which has a better DSP(Digital Signal Processor) and motion sensors.
fines the wheelchair locomotion mode that should be controlled to im- This enables us to realize the advanced motion control proposed in this
prove safety and riding comfort. Then the desired motion reference is paper with the minimum hardware modification. Notice that there is
designed, and the wheelchair motion states are estimated based on the no encoder to measure the wheel angle, as it is difficult to attach an
defined motion mode. Finally, the feedback controllers are designed to encoder to any off-the-shelf motorized wheel system.
achieve the desired motion along the locomotion mode. One unique fea- Mechanical specification of the wheelchair is illustrated in Table 1.
ture of the proposed control system is that it does not require any en- Appearance of proposed controller used in this study is depicted in
coder, which is not usually implemented in off-the-shelf wheelchairs. Fig. 2.

Fig. 1. (a) Powered wheelchair and its compo-

nents (b) Signal connections among the compo-
nents.

105
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 2. Appearance of the new control board

to replace the conventional control board.

Table 1 This implies that a human is familiar with the velocity control in the
Specification of the wheelchair used in this study. longitudinal direction and the orientation control in the rotational di-
Motor BLDC motor (24 V 200 W × 2) rection in vehicle operation. Taking this point into consideration, the
Weight 22 kg (including motors and batteries) wheelchair motion description and control will be determined in the
Wheel radius 0.11 m proposed locomotion mode in this study.
Distance between two wheels 0.6 m Fig. 4 illustrates the whole wheelchair operation including a human
user who commands the wheelchair motion based on his/her sensory
system and intention. The coordinates of the signals in Fig. 4 are de-
2.2. Definition of wheelchair locomotion mode fined based on the proposed locomotion mode, which means that the
joystick operation signals, the motion reference to the controller, and
A user provides his/her commands for the wheelchair motion the motion feedback of the wheelchair are all described based on the
through the joystick, which should be converted to the motion refer- proposed locomotion mode such that the control of the wheelchair is
ence for two motors. However, the relations among signals (of users intuitive enough for the user to feel safe and comfortable.
command, joystick angle, motor control reference and wheelchair mo-
tions) are not straightforward enough: the joystick motion has two- 2.3. Joystick interface design and motion reference generation in
dimensional characteristic, which corresponds to a coordinate in the xy locomotion mode
plane, while the wheelchair motion consists of three-dimensional coor-
dinates which are a position in the xy plane and an orientation angle. Fig. 5 illustrates the joystick utilized in this study. The lever of the
The motor torque input is also two-dimensional which does not directly joystick can be titled in two dimensions, and the tilted angle in each
correspond to a coordinate in the xy plane. These separated coordinates x/y direction is depicted as 𝜑/𝜙 respectively as shown in Fig. 5(a). The
joystick joystick
should be transformed and connected with each other for intuitive ma- Hall sensors in the joystick generate the voltages 𝑉longi and 𝑉lat
nipulation of wheelchair motions. propotional to 𝜑, and 𝜙 as shown in (1).
To address this problem, wheelchair locomotion mode is defined in These voltages are the command given by the user and should be
this study to describe and control the wheelchair motion intuitively. converted to the motion reference for the wheelchair as illustrated in
Fig. 3 illustrates the locomotion mode of wheelchair, which consists Fig. 4. In this study, two voltages are converted to the velocity references
joystick
of the longitudinal velocity mode and the rotational velocity mode. in the locomotion mode: 𝑉longi is converted to the longitudinal velocity
The longitudinal mode represents the forward and backward motion joystick
reference, and 𝑉lat is converted to the rotational velocity reference
of wheelchair, and the rotational mode represents the change of the ori- of the wheelchair.
entation of the wheelchair. However, direct conversion of these voltage signals to the references
This locomotion mode definition is similar to the intuitive opera- leads to very sensitive motion of the wheelchair, and thus some signal
tion/driving mode of car/vehicle system, which consists of longitudinal processing is required to realize safe and comfortable operation. A low
control (accelerator and brake) and orientation manipulation (steering). joystick
pass filter is applied to 𝑉longi to prevent very jerky acceleration and
joystick
deceleration. A dead zone is incorporated to 𝑉lat , where small level
angle 𝜙 under a threshold is considered zero. This prevents unnecessary
tremoring of the wheelchair in the rotational mode, which is caused by
sensor noise or small motions in the hand.
Eq. (2) illustrates this signal processing to generate wheelchair mo-
tion reference, 𝑣𝑟𝑒𝑓
𝑙
and 𝜔𝑟𝑒𝑓
𝑧 , where 𝜏 is the time response of the low
pass filter that can tune the sensitivity of the longitudinal motion, and
𝜖 is the dead zone width which also determines the sensitivity of the ro-
tational motion. 𝐾longi
scale and 𝐾 scale are scale factors for motion reference.
lat
Fig. 6 illustrates the signal processing and flow from the joystick to the
wheelchair control, where vl is the longitudinal velocity and 𝜔z is the
rotational velocity.
The proposed reference design allows to set the sensitivity and re-
Fig. 3. Locomotion definition. sponse time of joystick operation individually in each direction. For ex-

106
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 4. Signal flow in the whole wheelchair

control system.

the motors is conducted in the wheel level; the control input to the
wheelchair is the torques to the left and right wheels.
To cope with this discrepancy between the wheelchair motion con-
trol in the locomotion mode and the motor control in the wheel mode, a
transform between two modes should be incorporated [34,35] as shown
in Fig. 9. Jacobian matrix A is introduced for the transformation in
Fig. 9, which is defined as in (3). Jacobian A relates the wheel speeds
(𝜃̇ 𝑅 , 𝜃̇ 𝐿 ) to the wheelchair locomotion velocities vl and 𝜔z , and the in-
Fig. 5. Joystick lever and Hall sensor to generate voltage.
verse of its transpose 𝐴−𝑇 convert the driving force F and the rotating
moment M of the wheelchair to the wheel torques TR and TL .
ample, the wheelchair driving can be tuned less jerky in the longitudinal By introducing this Jacobian, the wheelchair motion control can be
direction, while still keeping the fast response of heading angle opera- designed in the locomotion mode, using the driving force F and the ro-
tion. tating moment M as the control inputs. while the actual control inputs
joystick
are the torques TR and TL .
𝑉longi ∝ sin𝜑
[ ] [ ] [ ]
𝑣𝑙 𝜃̇ 𝑅 ⎡ 1 1 ⎤ 𝜃̇
joystick =𝐴 =⎢
2 2 ⎥ 𝑅
𝑉lat ∝ sin𝜙 (1)
𝜔𝑧 𝜃̇ 𝐿 ⎢− 𝑊
⎣ 2
𝑊 ⎥ 𝜃̇ 𝐿
⎦
2
[ ] [ ]
𝐾longi
scale 𝑇𝑅 −𝑇
𝐹
=𝐴 (3)
𝑣𝑟𝑒𝑓
joystick
𝑙
= 𝑉 𝑇𝐿 𝑀
𝜏𝑠 + 1 longi
{ ( )
scale 𝑉 joystick − 𝜖
𝐾lat sin𝜙 > 𝜖 3.2. Wheelchair motion state observer without using encoder
𝜔𝑟𝑒𝑓
𝑧 = lat (2)
0 sin𝜙 < 𝜖.
In addition to the control input, the output should be obtained in the
3. Control algorithm considering intuitive wheelchair locomotion locomotion mode. In other words, vl and 𝜔z should be obtained for the
mode locomotion mode wheelchair control.
Wheel angle measurement using encoders can be utilized to provide
The whole control system proposed in this study is structured as in velocity observation by differentiating its measurement[20,23,36,37].
Fig. 7, which consists of 1) locomotion mode reference generation, 2) However, the wheelchair utilized in this study does not have an encoder,
feedback controller, 3) mode conversion and 4) locomotion mode state and it is difficult to install an encoder on the wheel of the off-the-shelf.
observer. To address this problem, the Hall sensor, which is already imple-
The motion reference generation using the joystick is discussed in the mented in the motor for the current control, is utilized to obtain accurate
previous section, and the other 2) feedback controller, 3) mode conver- wheelchair velocities.
sion and 4) locomotion mode state observer are studied in this section The number of pole pairs of the Hall sensor implemented in the mo-
tor is 4, which means the sensor can detect 8 pluses per revolution.
3.1. Transform between locomotion mode and wheel mode This resolution is very low and not appropriate for velocity estimation
because of the noise when it is differentiated. The accelerometer embed-
As illustrated in Fig. 8, the whole wheelchair motion is controlled ded on the board can be utilized to measure the wheelchair velocity, but
and observed in the locomotion mode, while the low-level control of integration of the measurement will result in bias or drift.

Fig. 6. Signal processing of joystick voltage for motion ref-

erence generation.

107
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 7. Entire control structure of the system.

Fig. 9(a) shows an experiment result of longitudinal velocity obser-

vation using only an accelerometer or only the Hall sensor. The result
shows that the drift and the noise are very problematic in these cases.
Therefore, in this study, sensor fusion based on Kalman Filter algorithm
is employed to address this issue.
Eq. (4) describes the relationship between the longitudinal velocity
vl and the sensor output aaccelerometer (measured acceleration), 𝜃𝑅
hall and

𝜃𝐿 (right and left wheel angles measured by the Hall sensors).

hall

𝜃𝑅
hall 𝜃𝐿hall 𝑥𝑙
𝑎accelerometer = 𝑣̇ 𝑙 , + = , (4)
2 2 𝑟

Fig. 8. Two different modes of wheelchair and conversion between them.

where r is the radius of the driving wheel as shown in Fig. 9(b), and xl
𝑑𝑥
is the traveled distance of the wheelchair (𝑣𝑙 = 𝑑𝑡𝑙 ).

Fig. 9. (a) Velocity estimation using Hall sensor (dashed

line) and accelerometer (solid line), (b) Relationship be-
tween wheel angle and wheelchair acceleration, (c) Kine-
matic Kalman Filter design.

108
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

[ ]𝑇
By defining the wheelchair motion state as 𝑥 = 𝑣𝑙 𝑥𝑙 ∈ 2 , the 𝐽𝑛 𝑠 + 𝐵𝑛
𝐶𝐹𝑟𝑜𝑡𝐹 = (10)
state space model of the wheelchair motion and the measurement output 𝜏𝑄 𝑠 + 1
is described as follows.
[ ] [ ] 𝑄=
1
(11)
0 0 1 accelerometer 𝜏𝑄 𝑠 + 1
𝑥̇ = 𝑥+ 𝑎 = 𝐴𝑥 + 𝐵𝑎accelerometer (5)
1 0 0
𝐽𝑛 𝑠 + 𝐵𝑛
[ ] 𝜃𝑅
hall 𝜃𝐿hall 𝑄𝑃𝑛−1 = (12)
𝑦= 0 1
𝑥 = 𝐶𝑥 = + (6) 𝜏𝑄 𝑠 + 1
𝑟 2 2
Jn , Bn are the rotational nominal dynamic models of wheelchair that are
Kinematic Kalman Filter[20,38] is designed based on this model, so
to be identified from the wheelchair motion. The rotational dynamics
that the estimated state 𝑥̂ 𝑙 can be utilized as the measurement of the
may be coupled with the longitudinal mode, but it is ignored in this
longitudinal velocity.
model, since the magnitude of the coupling is not significant, and also
The rotational velocity can be measured in a similar way, but the
the coupling effect can be rejected by YMO.
measurement 𝛾 gyroscope of the gyroscope(embed in the main controller
as shown in Fig. 2) can be directly utilized as the rotational wheelchair
4. Experimental verification
velocity. The whole wheelchair motion state is, then obtained as fol-
lows.
The proposed locomotion mode based control system is verified
[ ] through various experiments. Three points are evaluated in terms of
𝑣𝑙 = 1 0 𝑥̂
safety and comfort of wheelchair locomotion: motion reference genera-
𝜔𝑧 = 𝛾 𝑔𝑦𝑟𝑜𝑠𝑐𝑜𝑝𝑒 (7)
tion using joystick interface, acceleration/deceleration modification for
comfortable longitudinal motion and heading angle control for safe and
3.3. Locomotion mode based wheelchair motion control comfortable rotational motion.
Before the driving test, the performance of the motion state observa-
Wheelchair motion control for safety and comfort is designed in the tion using Hall sensor and acceleration is verified at first, and three types
locomotion mode as shown in Figs. 7 and 8, where the motion references of controller 1) conventional controller that is already implemented in
𝑣𝑟𝑒𝑓
𝑙
and 𝜔𝑟𝑒𝑓
𝑧 are given from the joystick. The longitudinal velocity con- the off-the-shelf powertrain using the previous controller board in Fig. 2,
trol and the rotational velocity control are designed separately as shown 2) feedback (FB) controller with estimated motion state feedback, and 3)
in Fig. 8. the proposed controller with feedback control, feedforward control and
Fig. 10 elaborates the proposed locomotion mode based control al- YMO are implemented using the observed motion states and compared
gorithm: the longitudinal velocity control employs a PD (proportional with each other in various conditions.
and differential) control as shown in (8). As there is no integral term in
(8), the tracking performance is not very strong enough and there can
4.1. Wheelchair motion state observer performance verification
be steady-state error.

𝐶𝐹𝑙𝑜𝑛𝑔𝑖
𝐵
= 𝐾𝑝 + 𝐾d 𝑠 (8) The performance of the proposed wheelchair motion state observer
is verified through experiments. A Kinematic Kalman Filter is de-
Meanwhile, the rotational velocity control utilizes Yaw Moment Ob-
signed based on (5) and (6) and implemented in the wheelchair. The
server (YMO) [25] which is a disturbance observer that eliminates any
noise/covariance for the KKF design were tuned based on trial and er-
disturbance in the rotational direction very quickly and firmly. Com-
ror, and finally it was set to (Qv , Qx )=(100,0.1) and 𝑅ℎ𝑎𝑙𝑙 = 0.1.
pared to the longitudinal control, this YMO is very strong feedback con-
Fig. 11 shows the longitudinal velocity observation results, which
trol.
compares 1) the (pseudo) time-derivative of the Hall sensor measure-
This different strategy is taken considering the safety and the ease
ments (Blue solid line), 2) the time integral of the accelerometer outputs
of manipulability for the user. Humans are more sensitive to the errors
(Black dashed line) and 3) the observation result of the proposed KKF
in the heading angle, and thus the disturbance rejection in the rota-
(Red solid line). The result validates that the problems of the conven-
tional mode should be secure enough. In addition to the PD controller
tional velocity measurements, which is the drift phenomenon or large
and YMO, a feedforward controller is utilized in the rotational velocity
noise, can be addressed by the proposed KKF. This result implies that
control to improve the response time. Eqs. (9)–(12) are the controller
the Hall sensor with very low resolution can be successfully utilized for
designs that are implemented in Fig. 10.
velocity measurement with combination of accelerometer.
𝐶𝐹𝑟𝑜𝑡𝐵 = 𝐾𝑝 + 𝐾d 𝑠 (9)

Fig. 10. Proposed locomotion mode based wheelchair motion control. Fig. 11. Velocity estimation comparison.

109
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 12. Motion reference generation from the

joystick input (a) voltage signal from the joy-
stick (b) generated motion reference.

4.2. Locomotion mode-based control performance verification Table 2

Comparison of peak jerk values under various reference generations .
4.2.1. Motion reference generation using joystick Conventional 𝜏1 = 0.159 𝜏1 = 0.227
The performance of the motion reference generation proposed in 𝜏2 = 0.159 𝜏2 = 0.328
Section 2.3 is verified here. Fig. 12 shows the reference generation by
PeakA 12.13 23.39 12.39
the joystick: Fig. 12 (a) is the voltage from the joystick sensors, and PeakD −18.48 −24.26 −10.32
Fig. 12(b) is the reference signals generated from the measured voltage.
The blue line in (a) indicates the voltage proportional to 𝜑 (in Fig. 5),
which leads to the blue line in (b), the reference 𝑣𝑟𝑒𝑓
𝑙
for the longitudinal
Table 2 compares the peak values with three different settings. In
velocity. The red line in (a) indicates the voltage proportional to 𝜙 (in
the case of the conventional motion generation (Fig. 13(a)), the peak
Fig. 5), which results in the red line in (b), the reference 𝜔𝑟𝑒𝑓 𝑧 for the
of decelerating jerk is very large leading to swaying of the users body,
rotational velocity.
which makes the user feel uncomfortable and even unsafe. The case
Three parameters 𝜏,𝐾longi
scale and 𝐾 scale for the reference generation in
lat with 𝜏1 = 𝜏2 = 0.159 (Fig. 13(b)) results in higher jerks both for accel-
(2) are set as 𝜏 = 0.159, 𝐾longi
scale = 2.3, 𝐾 scale = 1 in this experiment. The re-
lat eration and deceleration. The tuned parameter 𝜏1 = 0.227, 𝜏2 = 0.318.
joystick
sult in Fig. 12 (b) shows that the voltage 𝑉longi is successfully smoothed (Fig. 13(c)) can successfully generate similar jerk for acceleration but
and scaled to generate longitudinal velocity reference 𝑣𝑟𝑒𝑓 𝑙
so that the smaller jerk for deceleration.
user feels comfortable enough. Fig. 12(a) shows that there is noise and There is a dead-time (depicted as 𝛿 in Fig. 13(a)) before deceleration
joystick
tremoring in 𝑉lat , which may lead to improper reference generation. in the conventional joystick operation, and the proposed motion control
The result in Fig. 12(b) shows that the dead zone 𝜖, which is set to 0.4 V, can eliminate this dead-time too, which can lead to better manipulability
successfully removed this effect. of the wheelchair.

4.2.3. Disturbance rejection in the rotational mode for safety

4.2.2. Acceleration/deceleration modification in the longitudinal mode for The performance of the rotational wheelchair motion control is veri-
comfortable driving fied here. In particular, the ability of the wheelchair to keep its heading
The tuning factor 𝜏 in the reference generation can affect the com- angle is evaluated through experiments. To this end, the wheelchair was
fort of user, and moreover the time constant can be set differently for set on the level ground (as shown in Fig. 14(a)) and on uneven road (as
acceleration and deceleration respectively, namely, 𝜏 1 for acceleration shown in Fig. 14(b)), which are the road conditions that can be found
and 𝜏 2 for deceleration. To verify the effectiveness of this, it is inves- commonly in daily life.
tigated how the tuning factors 𝜏 1 and 𝜏 2 change the acceleration and In the following experiments, it is examined how easily the user can
deceleration of the wheelchair under the proposed longitudinal velocity go straight in the longitudinal direction without being disturbed; the
control. user pushes the joystick forward, and the straightness of the wheelchair
In the following experiments, the user drove the wheelchair forward motions is evaluated.
in three different situations: (1) with the conventional reference gen- Three types of controllers are adopted: 1)the conventional controller
eration and control (the algorithm that was already implemented in (which is implemented in the off-the-shelf controller), 2)the proposed
the off-the-shelf controller), (2) with the proposed reference generation feedback controller without YMO and 3)the proposed controller with
(𝜏1 = 0.159 and 𝜏2 = 0.159) and the longitudinal velocity control, and (3) YMO. The PD gains are set to 𝐾𝑝 = 0.02, 𝐾d = 0.001, and the nominal
with the proposed reference generation, but with different time constant value of Jn and Bn are set to 𝐽𝑛 = 0.004, 𝐵𝑛 = 0.0035, which were identi-
setting (𝜏1 = 0.227 and 𝜏2 = 0.318) and the longitudinal velocity control. fied from experiments. Lastly 𝜏 is set to 𝜏 = 0.016.
The PD gains are set as 𝐾𝑝 = 0.4, 𝐾d = 0.001 in this experiment. Fig. 15 illustrates the behavior of the wheelchair on level ground,
Fig. 13 compares the longitudinal motions under these different con- which shows that the wheelchair cannot go straight even on level ground
ditions. In all the experiments, the user titled the joystick to the same depending on the control algorithm. Fig. 16 compares the yaw rate mea-
angle as shown in the uppermost graphs in Fig. 13, and the velocities, surements during the experiments of Fig. 15. The solid blue line is the
the accelerations and the jerk [39] of the wheelchair in the longitudinal result with the conventional controller, the red dashed line represents
direction are measured and compared in Fig. 13. Even though the ve- the result with the feedback controller without YMO, and the black solid
locity patterns look similar, the acceleration and the jerk pattern show line is the result with the proposed controller with YMO. Notice that the
differences. Among these, the peak values of the jerks, which are PeakA conventional controller cannot keep the yaw rate 0, which means the
for acceleration and PeakD for deceleration are measured and compared heading angle of the wheelchair is deviated from its initial angle even
to evaluate driving comfort, as the jerk is commonly considered to be on the level ground. On the other hand, the yaw rates are well sup-
related to driving comfort [40]. pressed in the feedback controllers (with and without YMO).

110
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 13. Longitudinal wheelchair motions with different reference generations (a) conventional reference generation (b) proposed reference generation with 𝜏1 =
𝜏2 = 0.159 s (c) proposed reference generation with 𝜏1 = 0.227 s, 𝜏2 = 0.318 s.

Fig. 14. Road conditions to be experimented (a) level

ground environment (b) uneven road environment.

Another experiments were conducted on the uneven road condition are compared in Table 3. The conventional controller shows large yaw
as in Fig. 14(b), and the user also tilted the joystick forward to drive the rate RMSE value which means the heading angle of the wheelchair is not
wheelchair only forward. Fig. 17 illustrates how the wheelchair pro- kept constant and makes a large turn, while the proposed control algo-
ceeded on the uneven road with three different control algorithms. The rithm with YMO can significantly improve the performance of rotational
result shows that only the proposed controller with YMO (the right case) wheelchair motion suppressing rotational perturbation.
could keep its heading angle while others deviated from its original Another index rotational sensitivity (Srot ) is defined to evaluate how
heading angle. long the wheelchair can keep its heading angle without being perturbed
Fig. 18 shows the yaw rate measurements on the uneven road, where in the rotational direction. Eq. (13) is the definition of Srot , which is
the side slope appears around 3 s. The difference in the yaw rates be- the ratio of the lateral deviation D of the wheelchair to the longitudinal
comes more significant in this case: the conventional controller leads to travelled distance U.
large yaw rate deviation, and the feedback controller without YMO also 𝐷
𝑆𝑟𝑜𝑡 = (13)
shows large yaw rate, while the proposed controller with YMO effec- 𝑈
tively suppress yaw rate even on the side slope. This result implies that Fig. 19 illustrates the definition of U and D, which shows that Srot
the user can manipulate the heading angle of the wheelchair without becomes small when the wheelchair keeps its heading angle straight. In
being disturbed when the proposed control algorithm is applied. Table 3, Srot values of three controllers are compared, and the proposed
To evaluate the ability of the control to keep its heading angle quan- controller with YMO shows the best performance in terms of this Srot
titatively, the Root-Mean-Square Error(RMSE) values of the yaw rates too.

111
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 15. Experimental results on level ground

condition with (a) conventional controller (b)
proposed controller without YMO (c) proposed
controller with YMO.

Fig. 16. Experimental results of yaw rate on level ground

driving condition with three different type of controller.

Fig. 17. Experimental results on uneven road

condition with (a) conventional controller (b)
proposed controller without YMO (c) proposed
controller with YMO.

112
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

Fig. 18. Experimental results of yaw rate on uneven road

driving condition with three different type of controller.

Table 3
Experimental results for yaw rate and lateral sensitivity.

Conventional Proposed w/o YMO Proposed with YMO

RMS error (Level ground) 1.68 0.35 0.27

RMS error (Uneven road) 8.56 4.51 0.85
Lateral sensitivity (Level ground) 0.137 0.077 0.010
Lateral sensitivity (Uneven road) 0.188 0.154 0.017

References

[1] Cooper RA, Spaeth DM, Jones DK, Boninger ML, Fitzgerald SG, Guo S. Comparison of
virtual and real electric powered wheelchair driving using a position sensing joystick
and an isometric joystick. Med Eng Phys 2002;24(10):703–8.
[2] Rofer T, Mandel C, Laue T. Controlling an automated wheelchair via joy-
stick/head-joystick supported by smart driving assistance. In: 2009 IEEE interna-
tional conference on rehabilitation robotics; 2009. p. 743–8.
[3] Levine SP, Bell DA, Jaros LA, Simpson RC, Koren Y, Borenstein J. The navchair
assistive wheelchair navigation system. IEEE Trans Rehabil Eng 1999;7(4):443–51.
[4] Wada M, Asada HH. Design and control of a variable footprint mechanism for holo-
nomic omnidirectional vehicles and its application to wheelchairs. IEEE Trans Robot
Autom 1999;15(6):978–89.
[5] Borgolte U, Hoyer H, Bühler C, Heck H, Hoelper R. Architectural concepts of a
semi-autonomous wheelchair. J Intell Robot Syst 1998;22(3–4):233–53.
Fig. 19. Rotational deviation(D) and wheelchair traveling distance(U) to define [6] Kundu AS, Mazumder O, Lenka PK, Bhaumik S. Design and performance evaluation
of 4 wheeled omni wheelchair with reduced slip and vibration. Procedia Comput Sci
rotational sensitivity. 2017;105:289–95.
[7] Wada M. Mechanism and control of a 4wd robotic platform for omnidirectional
wheelchairs. In: 2009 IEEE/RSJ international conference on intelligent robots and
5. Conclusions systems; 2009. p. 4855–62.
[8] Simpson R, Lopresti E, Hayashi S, Guo S, Frisch R, Martin A, et al. The smart power
assistance module for manual wheelchairs. In: 26th International annual conference
In this paper, a novel control system was proposed to enhance on assistive technology for people with disabilities; 2003. p. 19–23.
the driving comfortability and safety of the joystick type powered [9] Kloosterman MG, Snoek GJ, van der Woude LH, Buurke JH, Rietman JS. A systematic
review on the pros and cons of using a pushrim-activated power-assisted wheelchair.
wheelchair. After defining an intuitive wheelchair locomotion mode,
Clin Rehabil 2013;27(4):299–313.
motion references for the wheelchair motion were generated in the pro- [10] Lawn MJ, Ishimatsu T. Modeling of a stair-climbing wheelchair mechanism with
posed locomotion mode. Also the conversion from the force in the pro- high single-step capability. IEEE Trans Neural Syst Rehabil Eng 2003;11(3):323–32.
posed locomotion mode to the motor torque was also presented. To [11] Nickel E, Hansen A, Pearlman J, Goldish G. A drive system to add standing mobility
to a manual standing wheelchair. Assistive Technol 2016;28(4):218–24.
achieve robust motion control for the generated reference by the joy- [12] Kitagawa L, Kobayashi T, Beppu T, Terashima K. Semi-autonomous obstacle avoid-
stick, longitudinal and rotational velocity controllers were developed. ance of omnidirectional wheelchair by joystick impedance control. In: Proceedings
The wheelchair motion in the locomotion mode was estimated by the 2001 IEEE/RSJ international conference on intelligent robots and systems. Expand-
ing the societal role of robotics in the next millennium, 4; 2001. p. 2148–2153 vol. 4.
proposed KKF successfully. The locomotion mode based control con- [13] Dicianno BE, Cooper RA, Coltellaro J. Joystick control for powered mobility: current
cept suggested in this paper was verified through several experiments in state of technology and future directions. Phys Med Rehabil Clin 2010;21(1):79–86.
common road environments. Finally, it was verified that driving perfor- [14] Fattouh A, Sahnoun M, Bourhis G. Force feedback joystick control of a powered
wheelchair: preliminary study. In: 2004 IEEE international conference on systems,
mance (comfortability and safety) was improved by the proposed con- man and cybernetics, 3; 2004. p. 2640–2645 vol. 3.
trol system. [15] Seki H, Tadakuma S. Straight and circular road driving control for power assisted
wheelchair based on fuzzy algorithm. In: IECON 2006 - 32nd annual conference on
IEEE industrial electronics; 2006. p. 3898–903.
Conflict of interest [16] Cooper RA, Corfman TA, Fitzgerald SG, Boninger ML, Spaeth DM, Ammer W,
et al. Performance assessment of a pushrim-activated power-assisted wheelchair con-
None. trol system. IEEE Trans Control Syst Technol 2002;10(1):121–6.

113
J.H. Choi, Y. Chung and S. Oh Mechatronics 59 (2019) 104–114

[17] Kim K, Nam K, Oh S, Fujimoto H, Hori Y. Two-dimensional assist control for pow- [36] Ohmae T, Matsuda T, Kamiyama K, Tachikawa M. A microprocessor-controlled high-
er-assisted wheelchair considering straight and rotational motion decomposition. In: -accuracy wide-range speed regulator for motor drives. IEEE Trans Ind Electron
IECON 2012 - 38th annual conference on IEEE industrial electronics society; 2012. 1982(3):207–11.
p. 4436–41. [37] Tsuji T, Mizuochi M, Nishi H, Ohnishi K. A velocity measurement method for accel-
[18] Kim K, Payne K, Oh S, Hori Y. One-handed propulsion control of power-assisted eration control. In: Industrial electronics Society, 2005. IECON 2005. 31st annual
wheelchair with advanced turning mode. In: 2014 IEEE 13th international workshop conference of IEEE. IEEE; 2005. p. 6-pp.
on Advanced Motion Control (AMC); 2014. p. 633–8. [38] Jeon S, Tomizuka M, Katou T. Kinematic Kalman filter (KKF) for robot end-effector
[19] Oh S, Hori Y. Development of extended wheelchair operation observer to estimate sensing. J Dyn Syst Meas Control 2009;131(2):021010.
precise two-dimensional tilt information. In: ASME 2007 international design engi- [39] Seki H, Sugimoto T, Tadakuma S. Novel straight road driving control of power as-
neering technical conferences and computers and information in engineering con- sisted wheelchair based on disturbance estimation and minimum jerk control. In:
ference. American Society of Mechanical Engineers; 2007. p. 13–20. Fourtieth IAS annual meeting. Conference record of the 2005 industry applications
[20] Oh S, Kong K, Hori Y. Operation state observation and condition recognition for the conference, 2005., 3; 2005. p. 1711–1717 Vol. 3.
control of power-assisted wheelchair. Mechatronics 2014;24(8):1101–11. [40] Huang Q, Wang H. Fundamental study of jerk: evaluation of shift quality and ride
[21] Oh S, Hata N, Hori Y. Integrated motion control of a wheelchair in the longitudinal, comfort. Tech. Rep. SAE Technical Paper; 2004.
lateral, and pitch directions. IEEE Trans Ind Electron 2008;55(4):1855–62.
[22] Seki H, Sugimoto T, Tadakuma S. Novel straight road driving control of power as- Jung Hyun Choi received the B.S. and M.S. degrees in me-
sisted wheelchair based on disturbance estimation and minimum jerk control. In: chanical engineering from Yeungnam University, Korea, in
Fourtieth IAS annual meeting. Conference record of the 2005 industry applications 2010, and 2013, respectively. He is Ph. D. student in the De-
conference, 2005., 3; 2005. p. 1711–1717 Vol. 3. partment of Robotics Engineering of Daegu Gyeongbuk Insti-
[23] Oh S, Hori Y. Disturbance attenuation control for power-assist wheelchair operation tute of Science and Technology(DGIST), Daegu, Korea Since
on slopes. IEEE Trans Control Syst Technol 2014;22(3):828–37. 2016. He worked as a Researcher in the IoT and Robotics Re-
[24] Kim K, Nam K, Oh S, Fujimoto H, Hori Y. Yaw motion control of power-assisted search Division of DGIST from 2013 to 2016. His research in-
wheelchairs under lateral disturbance environment. In: IECON 2011-37th Annual terests include the development of mobile robot control algo-
conference on IEEE industrial electronics society. IEEE; 2011. p. 4256–61. rithm.
[25] Fujimoto H, Saito T, Noguchi T. Motion stabilization control of electric vehicle under
snowy conditions based on yaw-moment observer. In: The 8th IEEE international
workshop on Advanced Motion Control, 2004. AMC ’04.; 2004. p. 35–40.
[26] Katsura S, Ohnishi K. Advanced motion control for wheelchair in unknown environ-
ment. In: 2006 IEEE international conference on systems, man and cybernetics, 6;
2006. p. 4926–31. Younghun Chung received the B.S. in mechanical engineer-
[27] Lee K-m, Lee C-H, Hwang S, Choi J, Bang Y-b. Power-assisted wheelchair with grav- ing from Pusan National university, Korea, in 2018. He is
ity and friction compensation. IEEE Trans Ind Electron 2016;63(4):2203–11. M.S. student in the Department of Robotics Engineering of
[28] Seki H, Ishihara K, Tadakuma S. Novel regenerative braking control of electric pow- Daegu Gyeongbuk Institute of Science and Technology (DG-
er-assisted wheelchair for safety downhill road driving. IEEE Trans Ind Electron IST), Daegu, Korea Since 2018. His research interests include
2009;56(5):1393–400. the development of mobile robot control algorithm.
[29] Ding D, Cooper RA, Guo S, Corfman TA. Analysis of driving backward in an elec-
tric-powered wheelchair. IEEE Trans Control Syst Technol 2004;12(6):934–43.
[30] Chen X, Agrawal SK. Assisting versus repelling force-feedback for learning of
a line following task in a wheelchair. IEEE Trans Neural Syst Rehabil Eng
2013;21(6):959–68.
[31] Carlson T, Demiris Y. Collaborative control for a robotic wheelchair: evaluation
of performance, attention, and workload. IEEE Trans Syst Man Cybern Part B
2012;42(3):876–88. Sehoon Oh received the B.S., M.S., and Ph.D. degrees in
[32] Maruno Y, Zengin AT, Okajima H, Matsunaga N, Nakamura N. Driving experiment electrical engineering from The University of Tokyo, Tokyo,
of front drive type electric wheelchair using yaw-rate control. In: 2012 Proceedings Japan, in 1998, 2000, and 2005, respectively. He was a re-
of SICE Annual Conference (SICE); 2012. p. 1408–13. search associate at The University of Tokyo until 2012, a Visit-
[33] Tsai M, Hsueh P. Synchronized motion control for 2d joystick-based electric ing Researcher at the University of Texas at Austin from 2010
wheelchair driven by two wheel motors. In: 2012 IEEE/ASME international con- to 2011, a Senior Researcher at the Samsung Heavy Indus-
ference on Advanced Intelligent Mechatronics (AIM); 2012. p. 702–7. tries, and a Research Professor at Sogang University. He is
[34] Kim K, Nam K, Oh S, Fujimoto H, Hori Y. Yaw motion control of power-assisted currently an Assistant Professor at the DGIST, Daegu, South
wheelchairs under lateral disturbance environment. In: IECON 2011-37th Annual Korea. His research interests include the development of force
conference on IEEE industrial electronics society. IEEE; 2011. p. 4256–61. control for human-friendly motion control algorithms and as-
[35] Kwak J, Choi W, Oh S. Modal force and torque control with wire-tension control sistive/exercise devices for people. Dr. Oh received the Best
using series elastic actuator for body weight support system. In: IECON 2017 - 43rd Transactions Paper Award from the IEEE Transactions on In-
Annual conference of the IEEE industrial electronics society; 2017. p. 6739–44. dustrial Electronics in 2013.

114