Measurement 120 (2018) 175–181

Contents lists available at ScienceDirect

Measurement
journal homepage: www.elsevier.com/locate/measurement

Experimental comparison of torque balance controllers for power-assisted T

wheelchair driving
⁎
Yoon Heo, Eung-pyo Hong , Yoon-hee Chang, Bora Jeong, Mu-sung Mun
Korea Orthopedics & Rehabilitation Engineering Center, Bupyeong-gu, Incheon, Republic of Korea

A R T I C L E I N F O A B S T R A C T

Keywords: A power-assisted wheelchair amplifies the user’s propulsion power. If the user’s arm strength is unbalanced, this
Power-assisted wheelchair can affect the driving balance. In order to correct this imbalance, a method of producing an assisting torque by
Torque balance control cross-referencing the opposite input torque was developed. One proposed torque balance control scheme in-
Temporal difference volves automatically controlling the cross-reference proportion according to the amplitude ratio of the left and
right input torques. However, this scheme cannot improve the driving performance under all conditions because
instability is inherent to the user’s propelling torque. To resolve this problem, a new torque balance control
method is proposed that considers not only the proportion of input torques but also the temporal difference. This
study examined the usefulness of the proposed torque balance control method based on the temporal difference
through a comparison with the existing method via a driving simulation and experiment.

1. Introduction
A push-rim activated power assisted wheelchair (PAPAW) is a spe-
cial wheelchair that detects the user’s propulsion torque on the push-
rim and assists the wheelchair’s driving torque with a motor device to
lessen the user’s physical fatigue. Its operation is as simple as that of a
manual wheelchair, but its motor assists with the driving torque. Thus,
PAPAW can be helpful in rehabilitating the disabled. The potential risk
factors of shoulder injury have been reported to be reduced, especially
for users of passive wheelchairs [1]. This benefit is also useful for SCI
patients [2].
The PAPAW research topic can be classified into two broad cate-
gories as follow: a method of detecting a user's force applied to a push-
rim and a driving control method for improving the driving perfor-
mance. The power and the signal must be connected to the torque
sensor mounted on the rotating push-rim without twisting the wire so
that the propulsive force of the user could be measured.
Non-contact torque recognition methods have been actively studied
for the purpose of transmitting signals to a rotating wheel [3,4].
Especially, Yamaha Co. developed the first commercialize PAPAW and
applied a circular connector that uses magnetic flux changes [3]. Alber
Co. solved the wire twist problem by integrating the battery, control,
and sensor device into a rotating wheel [5]. In recent years, some
studies have actively been conducted on a torque sensor-less method of
estimating the user propulsive torque by detecting the motion of a
minute wheel caused by a user propelling a wheelchair without using
direct torque sensors on a push-rim [6–8]. These methods were struc-
turally simple because they do not require a special device for power
and signal transmission. However, a problem in disturbance remains.
Hence, much research is still needed.
The second research topic is about driving performance improve-
ment. As we have mentioned earlier, PAPAW is driven by the user’s arm
propulsion. However, people’s left and right arms often have different
strength levels. Therefore, the direction frequently needs to be cor-
rected on the side with lesser power to make the wheelchair go straight.
However, in the case of a PAPAW, the user’s driving propulsion torque
is amplified by the motor power. Hence, unbalanced arm strengths have
a stronger effect on the wheelchair’s driving performance.
Torque balance control is generally used to improve the PAPAW
driving performance. In the torque balance control, the assisting torque
is determined by cross-referencing the opposite input torque to main-
tain a certain ratio [9,10]. Seki particularly suggested a torque balance
control method, where the user’s driving intention is recognized based
on the proportion of the left and right input torques, and the cross-
reference ratio is adjusted to improve both the straight and rotation
driving performances.
However, detecting the driver’s driving intentions by using only the
proportion of torques has limitations because the user’s left and right
propulsion torques are uneven and have a temporal difference. In this
study, we propose a new driving control method to solve this problem
and better reflect the user’s driving intention by additionally con-
sidering the temporal difference between two input torques [11].

⁎
Corresponding author.
E-mail address: ephong@kcomwel.or.kr (E.-p. Hong).

https://doi.org/10.1016/j.measurement.2018.02.024
Received 18 March 2016; Received in revised form 2 May 2017; Accepted 14 February 2018
Available online 15 February 2018
0263-2241/ © 2018 Elsevier Ltd. All rights reserved.
Y. Heo et al. Measurement 120 (2018) 175–181

Fig. 1. PAPAW control system structure.

This study compares the existing torque balance control method and
a new torque balance controller that considers the temporal difference
between the left and right propulsion torques to detect the user’s
driving intention via a driving simulation and an experiment. The re-
sults show that considering the temporal difference improved the
straight and rotation driving performances.

2. Torque balance controller

2.1. Overview

Fig. 1 shows the PAPAW control system. The balance controller

generates the balance torque Tb to decrease the instability of the user
input torque Th . β represents the cross-reference ratio. The torque
controller consists of a low-pass filter and generates the assisting torque
Ta . The assistance ratio is denoted by α. The time constant τ switches Fig. 3. Configuration block of the torque balance controller based on the torque pro-
from τfast to τslow when the user releases the push rim to generate virtual portion and temporal similarity.
inertia [10,12]. The motorized wheel is modeled with the inertia mo-
ment J and viscosity B.
2.2. Torque balance control based on torque and temporal differences
The PAPAW control system detects the user’s propulsion torque on
the left and right push rims and generates assistive torque with motors.
Fig. 3 shows the proposed torque balance controller that references
Since the left and right sides of the system work independently, when
not only the torque proportion, but also the temporal difference of the
the user’s left and right arm strengths are unbalanced, the PAPAW may
left and right side propelling torques. Symbol γ is the proportion of the
be difficult to move in a straight line. Hence, a controlling device is
user propelling torques with a range of −1 to 1 [10]. Symbol τs is the
needed for stable driving to correct the unbalance torque of the user.
temporal similarity which weighted by the temporal difference between
To correct the unbalanced propulsion torque of the user on the left
the left and right propulsive torques, and decreases as the temporal
and right wheels, one method involves referencing the torques of both
difference increases [Fig. 4(a)].
sides at a certain ratio to generate a common torque that is applied to
In the torque balance control block shown in Fig. 3, symbol β de-
the outputs on both sides. Fig. 2 shows the cross-referencing method for
notes the cross-referencing ratio for the left and right input torques that
generating assisting torques on both sides. Here, a and b are the re-
has been defined in the previous study as the balance ratio [10]. τs and
ference ratios of the left and right torques, respectively, to generate the
γ are applied herein in combination for the optimal balance ratio se-
common cross-referenced torque. α and β are the amplification ratios of
lection. Eq. (1) represents the relationship between the user input tor-
the left and right input torques used to generate the assisting torque.
ques (Thl , Thr ) and the balance ratio (β) to generate balanced output
However, this method has certain problems, such as a decreased
torques (Tbl , Tbr ). In addition, the sign of the output torque follows that of
rotation driving performance while improving the straight driving
the input torque, but if the input torque does not occur, it follows the
performance when the proportion of the common torque is too large.
sign of the opposite side [11].
When the proportion of the common torque is too small, the straight
driving performance is reduced. To resolve this problem, Seki proposed Tbl = β|Thl | + (1−β )|Thr |
a torque balance method where the reference ratio is adjusted ac-
Tbr = β|Thr | + (1−β )|Thl | (1)
cording to the proportion of the left and right torques to reflect the
user’s driving intention [10]. However, the torque ratio includes the Fig. 4(a) shows τs according to the temporal difference Δt and Fig. 4(b)
instability due to the user’s imbalance on the left and right sides, so it shows the definition of the balance ratio (β) with the change of γ . The
has limited ability to reflect the user’s driving intention. selection range of β proposed in this study can be varied in the range of
To address this problem, a torque balance control method was re- the A to D section according to τs . By doing so, it becomes possible to
cently proposed that considers the temporal difference in addition to choose more appropriate balance ratio which most closely matches the
the proportion of the left and right input torques [11]. user's driving intent.
For example, the inflection points P1 and P2 move to the left near A
and C when the intention is straight driving. This maximizes β, which
references the input torque of the opposite side despite the large dif-
ference between the left and right torques. As a result, a stable balance
torque can be generated. In contrast, the inflection points P1 and P2
move to the right near B and D when the intention is for a large rota-
tion. Hence, β is minimized.
We performed herein a driving experiment, where the four changing
section of β was set as follows: A γ = −0.8, Bγ = 0, Cγ = 0.5, and
Dγ = 0.8 on the transverse axis. The reference times, Δt1 and Δt2, were
Fig. 2. Cross-referencing scheme for generating balanced assist torques.
set to 200 ms and 400 ms, respectively.

176
Y. Heo et al. Measurement 120 (2018) 175–181

(a) Definition of temporal similarity (b) Range of variation in the balance ratio
Fig. 4. Temporal similarity and balance ratio variation range: (a) definition of the temporal similarity and (b) range of variation in the balance ratio dependent on the temporal similarity
between propelling torques.

3. Driving simulation of the torque balance controllers

As shown in Fig. 5, the driving balance of the PAPAW with regard to

the input torque can be evaluated from the deviation of the wheel-
chair’s initial and final locations Δdist and the posture angle Ø1 of the
final location [10]. In this study, the wheelchair’s straight and rotation
drive characteristics were simulated using input torques with amplitude
and temporal differences in order to compare the performance of the
two types of torque balance controllers.
In the simulation, the control parameter τ was fixed to 0.2 s, and α
was set to 1.0. J and B of the wheelchair model were set to 0.2 and 0.77,
respectively, from the characterization test of the wheel motor [6].
The two-wheel model, which ignores the influence of the caster, was
applied in order to estimate the driving trajectory. Under ideal condi-
tions, the caster can move in any direction without friction, so its effect
can be ignored for the wheelchair kinematic model, as shown in Fig. 6.
The wheel radius of the wheelchair r was 0.3 m, and the distance Fig. 6. Wheelchair kinematic model.
between wheels L was 0.5 m. The angular speeds of the left and right
wheels ωL and ωR can be given for each wheelchair model. The
where VL and VR are the rotation speeds of the left and right wheels
wheelchair position x, y and posture angle θ can be obtained from the
and v is the driving speed of the wheelchair. ω represents the angular
below equations [13]:
speed of the rotation center of the wheelchair.
VL = rωL,VR = rωR (2) In the simulation, three input torques were used to verify the torque
balance controller based on the temporal difference. As shown in
VL + VR
v= Fig. 7(a) and (b), in order to prove the control performance for straight
2 (3)
driving, torque signals were used where τs between the left and right
VL−VR input torques was 1.0 (Δt = 100 ms) or 0.5 (Δt = 300 ms), respectively.
ω=
L (4) To verify the rotation driving performance, a torque signal was applied
to only the left wheel, as shown in Fig. 7(c).
⎛ x ̇ ⎞ ⎛ cosθ 0 ⎞ v
()
⎜⎜ y ⎟⎟̇ = ⎜ sinθ 0 ⎟ ω 3.1. Straight driving simulation
⎝ θ ̇⎠ ⎝ 0 1 ⎠ (5)
Fig. 8(a) and (b) show the simulation results for the straight driving
of a PAPAW when τs = 1.0 and 0.5 (see Fig. 7(a) and (b)). When the
non-balance controller was applied, as shown in Fig. 8(a), the driving
trajectory of the wheelchair largely deviated to the right while it
moving about 0.6 m in the forward direction. In contrast, when Seki’s
controller and the balance controller based on temporal similarity were
used, the positional deviation was notably decreased.
Compared with the existing Seki balance controller, when the
driving controller based on temporal similarity was applied, the de-
viations in the position and posture angle were decreased, so the
wheelchair moved forward in a straight line more stably.
As shown in Fig. 8(b), when τs between the left and right input
torques was decreased to 0.5, the positional deviation generally in-
Fig. 5. Evaluation of the driving performance with the posture error Δdist and posture creased. However, the wheelchair moved in a straight line more stably
angle Ø1.
when the torque balance controller based on the temporal difference

177
Y. Heo et al. Measurement 120 (2018) 175–181

(a) Straight driving (¨t =100 ms) (b) Straight driving (¨t = 300 ms)

(c) Rotation
Fig. 7. Experimental input torques: (a) different input signals with τs = 1.0; (b) τs = 0.5; (c) the input torque is only on the left side with τs = 0.0.

was applied than when the Seki controller was applied.
3.2. Rotation driving simulation
For the rotation driving simulation, the same torque signal as shown
in Fig. 7(c) was applied. Fig. 9 shows the results of the driving simu-
lation. In general, rotation is best facilitated when the left and right
wheels are driven independently from each other. Hence, the biggest
angle of rotation was possible with the non-balance controller.
When Seki’s control method was applied, the change in posture
angle decreased to about 50% of that with the non-balance controller.
When the torque balance controller that considered the temporal dif-
ference was applied, the change in posture angle was about 75% of that
with the non-balance controller. This suggests that, when the temporal
difference is considered, the decrease in rotation performance due to
the limitations of the Seki’s torque balance method can be compen-
sated.

(a) ¨t = 100 ms (b) ¨t = 300 ms

Fig. 8. Simulation results for straight driving with the non-balance and two balance controllers: (a) different input torque with τs = 1.0 and (b) τs = 0.5.

178
Y. Heo et al. Measurement 120 (2018) 175–181

Fig. 9. Simulation results of rotation driving: (a) driving trajectory of the wheelchair when a force was applied to only left wheel; (b) angle change of the wheelchair over time.

4. Driving experiments Table 1

PAPAW specifications.
4.1. Overview
Wheel diameter 24 in
DC motor (RM51B) 22 W/3725 RPM
A consistent torque input was necessary for an objective comparison Gear ratio 1/4 timing belt × gear (19.7:1)
of the torque balance controllers. However, the strengths of the left and Maximum wheel speed 6 km/h
right arms of a human are not the same, nor are they easy maintain Wheel weight 6.5 kg
Sensor FSR sensors
consistently. In this study, a system was developed where the user Battery LiPo 25.9 V, 7.5 A h
transmits the torque signal to the controller wirelessly without
boarding. As shown in Fig. 10, the experimental system consisted of a
prototype PAPAW, control board, and data acquisiton system for ob- 5. Experimental results
taining several types of systematic data. The assisting torque was gen-
erated at the control board when the input torque signal was wirelessly 5.1. Straight driving
transmitted. Table 1 presents the specification of the developed PAPAW
[14]. Fig. 12(a) shows the driving test results of the PAPAW for the torque
input presented in Fig. 7(a). The input torque of the right side was
about 60% that of left side. Thus, when the torque balance was not
4.2. Driving experiment conditions controlled, the wheelchair curved to the right. In contrast, when the
balance control was applied, the straight driving improved. Especially,
A 3D motion analysis system was employed for precise measure- when the balance controller based on the temporal difference was ap-
ment of the driving trajectory of the PAPAW, as shown in Fig. 11(a). plied, the positional deviation was reduced to about 25% that of the
The 3D motion analysis system precisely analyzed the motion of an Seki method.
object moving in three-dimensional space by using a number of cameras As shown in Fig. 7(b), when τs was decreased to 0.5, the positional
and reflective markers that react to infrared rays. In order to measure deviation from the driving trajectory were greater than when τs was 1.
the driving trajectory and posture angle of the PAPAW, three reflective Although the straight driving performance was improved by the Seki
markers were used, as shown in Fig. 11(b). Reflective marker 1 was balance controller, it was further improved by the balance controller
attached to the center of the wheelchair in order to trace its route. considering the temporal difference. These driving experiments mat-
Reflective markers 2 and 3 were attached to the armrest for measure- ched with the simulation results.
ment of the wheelchair posture angle. In the driving experiment, τfast
and τslow of the torque controller were set to 0.2 and 1.0 s, respectively.

Fig. 10. Configuration of the experimental components.

179
Y. Heo et al. Measurement 120 (2018) 175–181

Fig. 11. Construction of the experimental environment: (a) experimental environment and (b) reflective marker setup.

(a) ¨t = 100 ms (b) ¨t = 300 ms

Fig. 12. Results of the straight driving experiment with non-balance and two balance controllers: (a) different input torque with τs = 1.0; (b) τs = 0.5.

Fig. 13. Results of the rotation driving experiment: (a) driving trajectory of the wheelchair, when a force was applied to only left wheel; (b) angle change of the wheelchair over time.

180
Y. Heo et al.
5.2. Rotation driving
Fig. 13 shows the results of the rotation driving experiment with the
PAPAW. When a force was applied to only one side wheel for rotation
driving of the wheelchair, the curve was sharpest with the non-balance
control. This agreed with the simulation result in Fig. 9. When Seki’s
balance control method was applied, the change in posture angle de-
creased to about 40% that of the non-balance control. When the balance
controller based on the temporal difference was applied, the change in
posture angle was 70% that of the non-balance controller. This suggests
that, when the temporal difference is considered, the rotation driving
performance can be improved compared to the case of the Seki method.
6. Conclusion and consideration
The assisting torque of a PAPAW is determined by the user’s pro-
pulsion torque. Thus, when the user’s propulsion torque is unstable, the
wheelchair’s driving performance deteriorates. To improve the wheel-
chair’s driving performance, one control method compensates for the
imbalance in the input torque via cross referencing of the input torques
of opposite sides. This study verified that the straight and rotation
driving performance can be further improved relative to existing
methods when the temporal difference is considered via a driving si-
mulation and experiment.
The driving experiments were performed under three driving con-
ditions, including straight and rotation, which are commonly used in
daily life. The results were verified with a 3D motion analysis system. In
the results, the torque balance controller based on the temporal dif-
ference was confirmed to provide better straight and rotation driving
control than the existing control method proposed by Seki.
This study proposes a method for generating a control torque as
close as possible to a user's intention even if the user's left and right arm
forces are unbalanced. Therefore, overcoming the change of the ex-
ternal environment (e.g., cross slope) that affects the driving situation is
difficult using the proposed method alone. However, the generated
assist torques, those reflect the user’s driving intention, can be set as
references for the feedback control of speed and current. By doing so, it
would be expecting that the PAPAW driving performance could be
improved even in an unbalanced road environment. In the future, we
181
Measurement 120 (2018) 175–181
plan to perform further studies related to the driving performance im-
provement in an unbalanced road environment.
Acknowledgments
This study was supported by the Senior-friendly Product R&D pro-
gram funded by the Ministry of Health & Welfare through the Korea
Health Industry Development Institute (KHIDI) (HI14C1496).
References
[1] M.G. Kloosterman, J.H. Buurke, W. de Vries, L.H. Van der Woude, J.S. Rietman,
Effect of power-assisted hand-rim wheelchair propulsion on shoulder load in ex-
perienced wheelchair users: a pilot study with an instrumented wheelchair, Med.
Eng. Phys. 37 (10) (Oct 2015) 961–968.
[2] B. Guillon, G. Van-Hecke, J. Iddir, N. Pellegrini, N. Beghoul, I. Vaugier, M. Figere,
D. Pradon, F. Lofaso, Evaluation of 3 pushrim-activated power-assisted wheelchairs
in patients with spinal cord injury, Arch. Phys. Med. Rehabil. 96 (5) (May 2015)
894–904.
[3] A. Uchiyama, H. Ogata, Power assisted wheelchair, US Patent 6,354,390B1, 2002.
[4] K.T. Nam, D.J. Jang, Y.C. Kim, Y. Heo, E.P. Hong, A study of a handrim-activated
power-assist wheelchair based on a non-contact torque sensor, Sensors (Basel) 16
(8) (2016).
[5] U. Alber, Drive assistance device for a hand-driven wheel chair, US Patent 6,155,
367A, 2000.
[6] J. Johansson, D. Petersson, Torque Sensor Free Power Assisted Wheelchair, Master
thesis, Univ. of Halmstad, 2007.
[7] W.M. Richter, Motion-based power assist system for wheelchairs, US Patent 0,008,
732A1, 2013.
[8] S.H. Oh, Fundamental Research on Human-friendly Motion Control, Ph. D. Thesis,
Univ. of Tokyo, 2005.
[9] A. Uchiyama, H. Ogata, Electric power-assisted wheelchair, US Patent 5,818,189A,
1998.
[10] H. Seki, T. Sugimoto, S. Tadakuma, Straight and circular road driving control of
power assisted wheelchair based on balanced assisted torque, p. 6.
[11] Y. Heo, E.P. Hong, M.S. Mun, T.H. Choi, Torque balancing for power assisted
wheelchair based on torque and temporal similarity, Int. J. Precision Eng. Manuf.
16 (8) (Jul. 2015) 1729–1734.
[12] H. Seki, N. Tanohata, Fuzzy control for electric power-assisted wheelchair driving
on disturbance roads, Part C: Appl. Rev., IEEE Trans. Syst., Man, Cybern. 42 (6)
(2012) 1624–1632.
[13] G. Campion, G. Bastin, B. D’Andrea-Novel, Structural properties and classification
of kinematic and dynamic models of wheeled mobile robots, IEEE Trans. on robotics
and automation 12 (1) (1996) 47.
[14] Y. Heo, K.T. Nam, E.P. Hong, M.S. Mun, Development of Pushrim Activated Power
Assisted Wheelchair, pp. 164–169.