Development of an Assist Controller with Robot Suit HAL
for Hemiplegic Patients Using Motion Data on the Unaffected Side
Hiroaki Kawamoto1 , Hideki Kandone1 , Takeru Sakurai2 , Ryohei Ariyasu1 ,
Yukiko Ueno1 , Kiyoshi Eguchi3 , and Yoshiyuki Sankai1
Abstract— Among several characteristics seen in gait of
hemiplegic patients after stroke, symmetry is known to be an
indicator of the degree of impairment of walking ability. This
paper proposes a control method for a wearable type lower limb
motion assist robot to realize spontaneous symmetric gait for
these individuals. This control method stores the motion of the
unaffected limb during swing and then provides motion support
on the affected limb during the subsequent swing using the
stored pattern to realize symmetric gait based on spontaneous
limb swing. This method is implemented on the robot suit HAL
(Hybrid Assistive Limbs). Clinical tests were conducted in order
to assess the feasibility of the control method. Our case study
involved participation of one chronic stroke patient who was
not able to flex his right knee. As a result, the walking support
for hemiplegic leg provided by the HAL improved the subject’s
gait symmetry. The feasibility study showed promising basis for
the future clinical study.

I. INTRODUCTION
In individuals with hemiplegia after stroke, walking ability
is impaired. Since walking is one of the major daily physical
activity, it leads to lowered activity of daily living (ADL).
Walking of individuals with hemiplegia is characterized by
slower speed, instability and gait asymmetry [1]. While these
properties are inter-related, gait asymmetry is considered
to be the key indicator to evaluate gait quality [2], being
correlated with challenges in balance control, gait inefficiencies, musculoskeletal degeneration, and hence decreased
daily activities. It is also known that it worsens through time
[3].
Gait asymmetry arises from impaired control in the affected limb [4], especially from difficulty in knee flexion
and ankle dorsi-flexion [5], [6]. These difficulties result in
compensatory motions in gait; circumduction of the affected
limb during swing to prevent the toe from dragging on the
ground [7], excessive plantarflexion of the unaffected ankle
during stance[8], and excessive extension of the knee on the
affected limb during stance to prevent sudden knee loosening
and to attain stability [9].
Methods to improve gait symmetry have been investigated
in the fields of physical therapy and rehabilitation robotics.
1 H.

Kawamoto, H. Kadone, R. Ariyasu, Y. Ueno are with Center for Cybernics Reserch, University of Tsukuba, 1-1-1 Tennnodai,
Tsukuba, 305-8577, Japan {kawamoto, kadone, ariyasu,

ueno, sankai} at ccr.tsukuba.ac.jp
2 T. Sakurai is with Graduate School of System and Information Engineering, University of Tsukuba, 1-1-1 Tennnodai, Tsukuba, 305-8573, Japan,

sakurai at golem.kz.tsukuba.ac.jp
3 K. Eguch is with Faculty of Medicine, University of Tsukuba,
1-1-1 Tennnodai, Tsukuba, 305-8575, Japan, kyeguchi at

md.tsukuba.ac.jp

978-1-4244-7929-0/14/$26.00 ©2014 IEEE

However, there have been no methods which make possible
for the patients to realize symmetrical gait in independent
free walking. Conventional training on treadmill provided by
a physical therapist manually guiding the affected limb does
not assure enough accuracy for the gait to be symmetric,
which is also the case in methods using visual or auditory
biofeedback for guidance. As for robotic devices, active AFO
[10] which assists ankle function and a controller of Lokomat
which extends motion of the affected limb while impeding
motion of the unaffectfed limb [11] were shown to increase
gait symmetry. However, the former does not directly assure
symmetry of gait during training and it relys on the patient’s
hip and knee ability for its improvement. In the latter, the
constrained pattern of the unaffected limb may be different
from the original spontaneous pattern that is appropriate for
his/her physical properties.
In this paper, we propose a control method of Robot Suit
HAL (Hybrid Assistive Limb, Fig.1.Left) [12] to realize
spontaneous symmetric gait in walking of individuals with
hemiplegia. HAL consists of an exoskeletal structure to be
attached on the lower limbs and the waist of the wearer,
actuated joints on the hips and knees, and a sensor system
to detect various information about the wearer. In our control
method, the joint angle sequence of the unaffected limb
during swing is recorded and used to generate the reference
trajectory for the subsequent swing of the affected limb, so
that the affected limb traces the pattern generated by the
unaffected limb in the preceding step. By this way, the wearer
can experience walking in a spontaneous pattern generated
by the unaffected limb, while keeping gait symmetry at a
higher accuracy. The proposed control method was implemented on the robot. Feasibility of the method was tested
in a clinical trial with a person with hemiplegia after stroke
(Fig.1.Right), where he walked in a symmetric gait pattern
while wearing the robot.
II. METHODS
A. Robot Suit HAL
HAL is basically composed of an exoskeleton, several
power units, the main controller, and the sensing system
(Fig.1 Left). The exoskeleton is designed to support the
mechanical functions of the human lower body. It is an
articulated structure, consisting of a frame, active joints at
the hips and knees, passive joints at the ankles. It is attached
to the user’s hips and legs using cuffs and belts. The joints
of the exoskeleton (hip, knee and ankle) have one DOF for
each in the sagital plane. The torques are generated by the

3077

phase 1

phase 2

phase 3

phase 4

swing
support

support
support

support
swing

support
support

Walking
motion

Left leg
Right leg

Fig. 2.

Phase division in walking. The left leg is colored with gray.

HAL & Human θ l θ˙l
Left leg
fl

Central nerve
System

Fig. 1. (Left) Overview of Robot Suit HAL. (Right) A hemiplegic patient
walking with HAL during the experiment.

power units at the hips and knees. Each power unit incorporates an actuator, a motor driver, a microprocessor and
communication interface. The motion support is achieved by
transmitting the torques from the power units to the user’s
legs through the exoskeleton frames.
The control of the HAL system is integrated by its main
controller. It controls the power units for assistance and
safety, gathering sensor data, detecting failures and communicating the sensor data to local computers to understand the
wearer’s and HAL’s condition.
The HAL is equipped with a sensing system based on
several types of sensors to detect the HAL’s state as well as
the wearer’s bioelectric signals. Potentiometers are mounted
on each joint of the HAL and used to measure the joint
angles. The bioelectrical sensors are attached on the skin
surface of the extensor and the flexor muscles of the knee
and the hip joint to detect their activity. Each shoe’s insole
contains two floor reaction force (FRF) sensors to measure
the FRFs generated at the front and the rear of the foot (heel
and ball areas).
B. Assist Controller
1) Phase Division during Walking: Walking motion can
be divided into four phases (Fig. 2). The phase 1 is the swing
phase of the left leg. In this phase the right leg supports
the body, while the left foot leaves the ground surface and
the left leg swings forward. In the phase 2 the left foot
contacts the ground surface at the front of the body so that
the wearer’s weight is supported by both legs. The weight
is then completely transferred to the left foot. The phase 3
is the swing phase of the right leg. The right foot leaves the
ground surface and the right leg swings forward. The body
weight is fully supported by the left leg. The phase 4 is the
phase where the right foot contacts the ground surface at the
front of the body and both legs support the body. The weight
is then completely transferred to the left foot.
2) Control System: Figure 3 shows a control system to
support hemiplegic leg motion on the right side (We suppose
here the right side is the affected side). Through the all
gait phases, the active joints on the unaffected side (left)

θ r ref θ̇r ref

Motion buffer

Tracking
controller

Fig. 3.

τr

fr
HAL & Human
Right leg
θ r θ˙r

Block diagram for HAL control system for hemiplegia patients.

are controlled to compensate viscosity, to allow spontaneous
motion of the wearer. In the phase 1, while the wearer swings
the left leg, HAL detects the angles θ l = [θlh , θlk ]T consisting
of left hip joint angle θlh and left knee joint angle θlk , and
calculates angular velocities θ̇θ l = [θ̇lh , θ̇lk ]T consisting of left
hip joint angular velocity θ̇lh and left knee joint angular
velocity θ̇lk . The θ l and θ̇θ l are recorded in motion buffer
and transformed into reference joint angles θ r re f and the
reference angular velocities θ̇θ r re f , which are used for motion
support in right leg.
As the walking phase transits to the phase 3, the tracking
controller generates assistive torque on the right side τ r =
[τrh , τrk ] consisting of right hip joint torque τrh and left knee
joint torque τrk , which are calculated using PD controller
based on the differences between motion reference (θθ r re f ,
θ̇θ r re f ) and the current right angles θ r = [θrh , θrk ]T consisting
of right hip joint angle θrh and right knee joint angle θrk ,
and right joint angular velocities θ̇θ r = [θ̇rh , θ̇rk ]T consisting
of right hip joint angular velocity θ̇rh and right knee joint
angular velocity θ̇rk .
Detection of transition between the phases [13] is based
on thresholding on the values of FRF sensors attached on
the front and rear parts of the left and right feet.
C. Clinical Evaluation
We implemented a HAL with the proposed control
method, and applied it to a 63 year old male patient with
hemiplegia on the right side resulting from stroke. In his
daily life, he wears an ankle foot orthosis, and walks using a
cane. His walking is characterized by a circumduction gait,
due to the difficulties to flex the right knee joint without

3078

Knee joint range of motion [deg]

Left hip
Right hip

(a)

0.4
-0.1
1.4

Left knee
Right knee

(c)

0.9

(b)

Left

Right

40.0

30.0

20.0

10.0

Without HAL

With HAL

0.4

(b)

-0.1
0

0.5

1

1.5

2

Stance duration [sec]

Angle [rad] Angle [rad]

0.9

(a)
50.0

2.5

Time [sec]
Fig. 4. Hip and knee joint angles of both legs for two gait cycles without
HAL. The swing phase of right (affected) side is colored with gray.

Left

Right

1.0
0.9
0.8
0.7
0.6
0.5
0.4

P2

Phase 3

P4

Phase 1

P2

Phase 3

P4

Left hip

(a)

Angle [rad] Torque [Nm] Angle [rad]

Angle [rad]

Phase 1

2
1.5
1
0.5
0
-0.5
2
1.5
1
0.5
0
-0.5
20
10
0
-10
-20
2
1.5
1
0.5
0
-0.5
2
1.5
1
0.5
0
-0.5
10
5
0
-5
-10

Maximum heel hegiht [mm]

Without HAL

(a')

Target trajectory
Actual trajectory

200
180
160

detection of phase transition and the amount of the assistive
torque based on report by the wearer and observation by the
experimenters. For the latter two weeks, he learned to walk
being aware of the integrated motion between HAL and his
limbs with the fixed parameters. Figure 1(Right) shows the
patient receiving the HAL’s motion support.
We compared the angles of both left and right knee joints
during walking with and without wearing HAL. Moreover,
we also compared the difference between the motion range
of the knee joint angle in swing phase, the support phase
duration, and the maximum heel height from ground in swing
phase. These data were calculated based on data measured
by a motion capture system (VICON MX with 16 T40s
cameras).

(a')

(a)

Target trajectory
Actual trajectory

With HAL

Fig. 6. Differences in the angle range for each knee joint in swing phase
(a), support phase duration (b), maximum heel height in swing phase (c)
with and without wearing HAL

Left knee

Torque [Nm] Angle [rad]

220

(b')

Right knee

(c)

(c)
(b')

(b)

Right knee

2

3

4

Time [sec]

Fig. 5.
HAL.

Right

240

Without HAL

Right hip

1

Left

(c)

Right hip

(b)

0

260

With HAL

III. RESULTS

Hip and knee joint angles of both legs for two gait cycles with

flexing the right hip joint. He gave an informed consent
before participating.
The clinical trial was conducted once a week for four
weeks, and each trial was 60 minutes. For the former two
weeks, we mainly adjusted the parameters related to the

Hip and knee joint angles during walking without wearing
the HAL are shown in Fig. 4. The reference used for the
joints angle is the value measured for the standing posture.
The angles are considered positive during the flexion and
negative during the extension. It is clear that the ranges
of flexion in the right (affected) side of the hip and knee
joints during the swing phase are smaller than that in the
left (unaffected) side (Fig. 4 (a), (b)), and especially there

3079

was the significant difference between both sides in the knee
joint angles. The trajectory of knee joint angle during the
swing phase was not generated smoothly (Fig. 4 (c)).
Figure 5 illustrates the joint angles and asssitive torques of
hip and knee joints in affected side during walking wearing
the HAL. The joint angle trajectories recorded during the
swing phase in the left side (Fig. 5 (a), (a’)) is used as
target trajectories in the joint angle of the right side (Fig. 5
(b), (b’) red). The assistive torque is positive during flexion
and negative during extension. It was observed that the HAL
generates the assistive torque during the swing phase of the
right side based on the target trajectories, and the joint angles
of the right side were provided in almost accordance with
the target trajectories. The knee joint angle during the swing
phase provided smooth trajectory (Fig. 5 (c)).
Figure 6 (a-c) shows the differences, by averaging data
from 10 strides, in the angle range for each knee joint in
swing phase, the support phase duration of each leg, and
maximum heel height of each foot in swing phase with and
without wearing the HAL. Walking without wearing HAL
demonstrated differences between the right and left side. In
particular, the knee joint range of motion and the maximum
heel height of the right side were significantly shorter than
those of the left side. The stance duration of the right side
was longer than that of the left side. In contrast, walking with
wearing HAL reduced the differences between the sides. The
knee joint range of motion and the maximum heel length of
the right side were significantly increased, while those of
the left side were decreased. The difference of the stance
duration was decreased while the stance duration of the left
and right sides was increased.
IV. DISCUSSION
The clinical study was conducted with a hemiplegic patient
who had asymmetric gait pattern in order to evaluate the
feasibility of the proposed control method. The hip and knee
joint angles on the affected side during the swing phase were
quite similar to those on the unaffected side. Furthermore, the
proposed control method reduced the variety of differences
in gait between both sides due to hemiplegia. Therefore we
confirmed that the proposed control method could provide
the motion support to improve walking symmetry for the
patient with hemiplegia. However, the knee joint range of
motion and the maximum heel height on the unaffected side
were decreased when the motion support was provided by
the proposed control method. This may reflect the fact that
the wearer has adjusted the unaffected gait according to the
supported gait on the affected side. The next step would be
to confirm the efficacy of the control method in the clinical
study with several hemiplegic patients.
The control method is expected to contribute to functional
recovery of gait for hemiplegic patients. It is based on
the observation that the control method repeatedly provided
the symmetrical motion support as locomotor treatment. In
conventional rehabilitation, locomotor training tends to aim
to walk as fast and stable as possible by exploiting the
residual function in the unaffected leg, and does not prioritize

the functional recovery to make the hemiplegic gait closer
to symmetry. In contrast, by using the proposed control
method, patients can learn symmetrical gait as far as they can
keep standing posture and advance to the stage of locomotor
training after stroke. In the near future, we will extend the
clinical studies to examine gait functional recovery.
V. CONCLUSIONS
We proposed an assistive control method for hemiplegia
patients to support motion of the affected leg using motion
data of the unaffected leg during waking with HAL. It
realizes symmetric gait based on the spontaneous swing
pattern of the unaffected side. To investigate the efficiency
of the control method, the walking support was applied in
clinical trials with a hemiplegic subject whose gait was asymmetric. As a result, the control method allowed the subject
to perform the motion of unaffected leg on the affected side,
which improved the gait symmetry significantly. In the near
future, we will apply this control method to subjects showing
different types of hemiplegia.
ACKNOWLEDGMENT
This study was supported by the “Funding Program for
World-Leading Innovative R&D on Science and Technology
(FIRST Program)” and “Tateishi Science and Technology
Foundation” .
R EFERENCES
[1] J. C. Wall and G. I. Turnbull, “Gait asymmetries in residual hemiplegia”, Arch Phys Med Rehabil, vol.67 (8), 1986, pp.550-553.
[2] K. K. Patterson, W. H. Gage,D. Brooks, S. E. Black, W. E. McIlroy,
“Evaluation of gait symmetry after stroke: A comparison of current
methods and recommendations for standardization”, Gait Posture,
vol.31, 2010, pp.241-246.
[3] G. I. Turnbull and J. C. Wall, “Long-term changes in hemiplegic gait”,
Gait Posture, vol.3, 1995, pp.550-553.
[4] J. Perry and J. M. Burnfield, “Gait Analysis: Normal and Pathological
Function (2nd ed.)”, SLACK Incorporated, 2010.
[5] B. B. Bhakta, “Management of spasticity in stroke”, British Medical
Bulletin, vol.56 (2), 2000, pp.476-485.
[6] A. Thibaut, C. Chatelle1, E. Ziegler, MA. Bruno, S. Laureys, and
O. Gosseries, “Spasticity after stroke: Physiology, assessment and
treatment”, Brain Injury, Vol. 27, No. 10, 2013, pp. 1093-1105.
[7] DC Kerrigan, EP Frates, S Rogan, “Hip hiking and circumduction:
quantitative definitions”, Am. J. Phys. Med. Rehabil., Vol. 79 (3),
2000, PP. 247-252.
[8] G. Chen ,C. Patten, D.H. Kothari, F. E. Zajac, “ Gait differences
between individuals with post-stroke hemiparesis and non-disabled
controls at matched speeds”, Gait Posture., Vol. 22 (1), 2005, PP.
51-56.
[9] SJ Olney, C Richards, “Hemiparetic gait following stroke. Part I:
Characteristics”, Gait Posture, Vol. 4 (2), 1996, PP. 136-148.
[10] J. A. Blaya and H. Herr, “Adaptive control of a variable-impedance
ankle-foot orthosis to assist drop-foot gait”, IEEE Transactions on
Neural Systems and Rehabilitation Engineering, vol.12(1), 2004,
pp.24-31.
[11] C. Bonnyaud, D. Pradon, J. Boudarham, J. Robertson, N. Vuillerme
and N. Roche, “Effects of gait training using a robotic constraint
(Lokomat) on gait kinematics and kinetics in chronic stroke patients”,
J. Rehabil. Med. Vol.46, 2014, pp.132-138.
[12] K. Suzuki, G. Mito, H, Kawamoto, Y.Hasegawa, Y. Sankai, “IntentionBased Walking Support for Paraplegia Patients with Robot Suit HAL”,
Advanced Robotics, vol. 21 (12), 2007, pp. 1441-1469.
[13] H. Kawamoto, T. Hayashi, T. Sakurai, K. Eguchi, Y. Sankai, “Development of single leg version of HAL for hemiplegia”, Proc. of 31st
Annu. Int. Conf. of the IEEE Engineering in Medicine and Biology
Society, 2009, pp. 5038-5043.

3080