robotics

Article
Transformable Wheelchair–Exoskeleton Hybrid Robot for
Assisting Human Locomotion
Ronnapee Chaichaowarat * , Sarunpat Prakthong and Siri Thitipankul

International School of Engineering, Chulalongkorn University, 254 Phayathai Road, Pathumwan,

Bangkok 10330, Thailand
* Correspondence: ronnapee.c@chula.ac.th

Abstract: This paper presents a novel wheelchair–exoskeleton hybrid robot that can transform
between sitting and walking modes. The lower-limb exoskeleton uses planetary-geared motors to
support the hip and knee joints. Meanwhile, the ankle joints are passive. The left and right wheel
modules can be retracted to the lower legs of the exoskeleton to prepare for walking or stepping
over obstacles. The chair legs are designed to form a stable sitting posture to avoid falling while
traveling on smooth surfaces with low energy consumption. Skateboard hub motors are used as the
front driving wheels along with the rear caster wheels. The turning radius trajectory as the result of
differential driving was observed in several scenarios. For assisting sit-to-stand motion, the desired
joint velocities are commanded by the user while the damping of the motors is set. For stand-to-sit
motion, the equilibrium of each joint is set to correspond to the standing posture, while stiffness
is adjusted on the basis of assistive levels. The joint torques supported by the exoskeleton were
recorded during motion, and leg muscle activities were studied via surface electromyography for
further improvement.

Keywords: exoskeletons; wheelchair; sit-to-stand motion; differential driving

Citation: Chaichaowarat, R.; 1. Introduction

Prakthong, S.; Thitipankul, S.
In most regions around the world, the population aged 65 and older is growing
Transformable Wheelchair–Exoskeleton
faster than the total population. Physical deterioration is inevitable with age. Mobility
Hybrid Robot for Assisting Human
Locomotion. Robotics 2023, 12, 16.
improvements allow the elderly and people with disabilities to access jobs, education, and
https://doi.org/10.3390/
healthcare and keep their societal roles. Wheelchairs are common personal assistive devices
robotics12010016
for individuals with locomotor disabilities. The lever propelling mechanism was applied
to improve the performance of manual wheelchairs [1]. Although traveling by wheels
Academic Editors: Weitian Wang,
on rigid frames is stable, safe, and consumes low energy, steps and stairs remain critical
Michael Bixter and Quanjun Song
obstacles. Elevators and slopes for reducing barriers to locomotion cannot be provided in
Received: 11 December 2022 all public facilities, and their use is sometimes prohibited during emergencies. In addition
Revised: 15 January 2023 to improving wheelchair-friendly infrastructure, reducing the limitations of wheelchairs
Accepted: 16 January 2023 in the presence of obstacles is challenging. Although manual and power-assisted stair-
Published: 18 January 2023 climbing wheelchairs [2–5] have been developed thus far, their structures remain too bulky
to be carried on vehicles by users.
Wheelchair users are usually constrained to the sitting posture, which is sometimes
inconvenient in environments designed for people capable of standing. For elderly per-
Copyright: © 2023 by the authors. sons [6] and patients with Parkinson’s disease or with spinocerebellar degeneration [7], the
Licensee MDPI, Basel, Switzerland.
standing posture provides not only great independence but also has medical advantages
This article is an open access article
in terms of bone metabolism, blood circulation, and inflammation prevention. Upright
distributed under the terms and
wheelchairs [8,9] were developed as alternatives. Standing mobility devices [10] with
conditions of the Creative Commons
passive exoskeletons using gas springs assist users with lower limb motor disabilities in
Attribution (CC BY) license (https://
sitting down or standing up from an ordinary chair and maintaining voluntary postural
creativecommons.org/licenses/by/
4.0/).
transitions. However, wheelchairs that can climb up stairs or obstacles in standing posture

Robotics 2023, 12, 16. https://doi.org/10.3390/robotics12010016 https://www.mdpi.com/journal/robotics

Robotics 2023, 12, 16 2 of 16

and allow free walking have not been mentioned in the literature. The wearable robot suit
HAL [11] can expand the physical capabilities of healthy people. The powered exoskele-
tons Ekso [12], ReWalk [13], and REX [14] were developed for rehabilitating and assisting
the daily locomotion of people with disabled lower limbs. These exoskeletons can assist
patients in walking and keeping their limbs active. However, users must use crutches
to maintain their balance with their upper limbs. In addition, humans and exoskeletons
consume considerable energy during walking [15]. Thus, the distance of movement is
limited by battery capacity. The walking performance and safety of robots also strongly
rely on control systems and the quality of gait phase detection.
The concept of a detachable lower-limb exoskeleton from an adjustable-height wheelchair
was proposed to combine the advantages of traveling by wheel over long distances and by
walking exoskeletons over complex terrains [16]. However, users cannot carry full-sized
wheelchairs for on-demand sitting. The concept of hybrid assistive wheelchair–exoskeleton
robots with a reduced number of actuators was proposed [17]. Nevertheless, its design cannot
guarantee the balance of the user during the configuration transition. A large joint torque was
also required to support the chair leg in equilibrium to maintain the distance between the front
and rear wheels in the sitting configuration. The reconfigurable mechanism was designed for
wheelchair–exoskeleton hybrid robots [18] to secure the user’s balance during sit-to-stand and
stand-to-sit motions without requiring additional support from the upper limbs. Given that
chair legs are not included in the design, the moment due to the weight of the human body
was mainly supported by the linear actuators with high gear ratios. The dynamic load due to
road vibration was transmitted to the gear. The non-backdrive mechanism is not preferred in
consideration of safe physical human–robot interaction [19,20].
The biomechanics of sit-to-stand motion in elderly persons was studied [21] on the ba-
sis of the kinematic data collected using video and the ground reaction force measured with
a force plate, along with muscle activity monitored with surface electromyography (EMG).
The muscle activities of elderly fallers and non-fallers during sit-to-stand motion were
compared [22]. The joint torque and power consumption during motion were estimated on
the basis of the human model [23]. The kinematic model of a lower-limb exoskeleton was
proposed for determining the joint angle and position of the leg during movement [24].
The active impedance control of a lower-limb exoskeleton with the human joint torque
observer was proposed for sit-to-stand movement [25]. The control method of a wearable
robot for the sit-to-stand and stand-to-sit transfers of patients with spinal cord injuries
was presented [26]. The concept of a passive gravity-balanced assistive device using a
counterweight and springs connected to the auxiliary parallelograms considering the hip,
knee, and ankle torques required against joint angles was proposed [27].
We propose a novel wheelchair–exoskeleton hybrid robot that can transform between
sitting and standing configurations as an alternative compact and lightweight personal
mobility vehicle for the elderly and people with disabilities. The lower-limb exoskeleton
uses motors with planetary gears to support the hip and knee joints. Meanwhile, the ankle
joints are passive. The left and right wheel modules can be retracted to the lower legs
of the exoskeleton to prepare for walking. The chair legs are designed to form a stable
sitting posture to avoid falling while traveling on smooth surfaces by using two skateboard
hub motors as the front driving wheels. In this work, the simplified human model was
derived in accordance with wheelchair parameters to simulate the hip and knee moments
required during a sit-to-stand motion to select the actuators driving the exoskeleton joints
without requiring an additional high-force actuator to support the motion. The prototype
of the wheelchair–exoskeleton hybrid robot was built and tested. For assisting sit-to-stand
motion, the desired joint velocities are commanded by a user while the damping of the
motors is set. For stand-to-sit motion, the equilibrium of each joint is set to correspond to
the standing posture, whereas stiffness is adjusted on the basis of the assistive level. During
tests, the exoskeleton joint torques were recorded, and leg muscle activities were studied
via surface EMG. The turning radius trajectory as the result of differential driving in the
wheelchair mode in several scenarios was observed.
Robotics 2023, 12, 16 3 of 16

Robotics 2023, 12, 16 3 of 16

were studied via surface EMG. The turning radius trajectory as the result of differential
driving in the wheelchair mode in several scenarios was observed.
Section 2 introduces the design concept of the wheelchair–exoskeleton hybrid robot
Section the
and derives 2 introduces
human modelthe design concept of
for simulating thethe wheelchair–exoskeleton
joint hybrid
torques during sit-to-stand robot
motion.
and derives the human model for simulating the joint torques during sit-to-stand
Section 3 shows the built prototype and explains the control system. Section 4 describes motion.
Section 3 shows the
the experimental builtand
setup prototype andthe
discusses explains
resultsthe control
of the system. and
sit-to-stand Section 4 describes
stand-to-sit the
exper-
experimental setup and discusses the results of the sit-to-stand and stand-to-sit experiments.
iments. Section 5 presents the results of differential driving tests. Section 6 summarizes
Section
the key 5findings
presents theongoing
and results of differential driving tests. Section 6 summarizes the key
efforts.
findings and ongoing efforts.
2. Concept of the Wheelchair–Exoskeleton Hybrid Robot
2. Concept of the Wheelchair–Exoskeleton Hybrid Robot
2.1. Biomechanics
2.1. Biomechanics of of Sit-to-Stand
Sit-to-Stand Motion
Motion
The wheelchair–exoskeleton
The wheelchair–exoskeletonhybrid hybridrobotrobotcancantransform
transform from fromsitting to standing
sitting con-
to standing
figurations on the sagittal plane, as shown in Figure 1. Trunk flexion
configurations on the sagittal plane, as shown in Figure 1. Trunk flexion (increasing the hip (increasing the hip
angle θ𝜃) )isisnecessary
angle necessaryto tomove
movethe
thebody’s
body’scenter
centerof ofgravity
gravity forward
forward over over the
the front
front wheels
wheels
h
(pivot point) to prepare for standing up. The ground-contacting point
(pivot point) to prepare for standing up. The ground-contacting point is shifted from the is shifted from the
rear caster wheels to the feet during this transition. By creating the knee
rear caster wheels to the feet during this transition. By creating the knee extension moment extension moment
𝑀 k ,, the
M the whole
wholebodybodyisislifted
lifted upward.
upward. AsAsthethe
knee knee is extended
is extended (the knee
(the knee angleangle 𝜃 de-
θk decreases)
creases) adequately, the hip extension moment 𝑀 is required to
adequately, the hip extension moment Mh is required to align the upper body in an uprightalign the upper body in
an upright
posture (theposture
hip angle(thedecreases).
hip angle decreases).
The groundThe ground
reaction reaction
force N is now 𝑁 is now
forcelocated located
anterior to
anterior
the frontto the front
wheels wheels to the
to maintain maintain the equilibrium
equilibrium in the standing
in the standing configuration.
configuration. The pairThe
of
pair of
chair chair
legs legs is rotated
is rotated upward,upward,
and the and the left
left and rightand rightmodules
wheel wheel modules are retracted
are retracted to
to prepare
prepare
for walking. for walking.

Figure
Figure 1.1. Conceptual design of the the wheelchair–exoskeleton
wheelchair–exoskeleton hybrid robot during during the
the sit-to-stand
sit-to-stand
transition on the
transition the sagittal
sagittal plane.
plane. In the sitting configuration, the ground reaction force force position
position is
is
assumedto
assumed tobe
beatatthe
thefront
front wheel.
wheel. The
The trunk,
trunk, head,
head, andand
armsarms
are are simply
simply considered
considered a rigid
a rigid upper-
upper-body
bodyThe
link. link.hip,
Theknee,
hip, knee, and ankle
and ankle joint joint angles
angles are shown.
are shown. TheThe weight
weight of each
of each segment
segment is assumed
is assumed to
to be at its center of gravity. The locations of the link lengths and center of gravity in the
be at its center of gravity. The locations of the link lengths and center of gravity in the standing standing
configuration are shown.
configuration are shown.

2.2. Estimation of Knee and Hip Moments

2.2. Moments from
from the
the Human
Human ModelModel
By considering
By consideringthe thefront
frontwheel
wheelasasthe
thereference
referenceofof the
the horizontal
horizontal position,
position, as as shown
shown in
in Figure
Figure 1, the
1, the offset
offset 𝑑 measured
d measured to theto the ankle
ankle joint isjoint is assumed
assumed to be constant
to be constant at 75 mm at during
75 mm
during sit-to-stand
sit-to-stand motion. motion. The horizontal
The horizontal positionspositions of the center
of the shank’s shank’s ofcenter
gravityofXgravity 𝑋 ,
sc , the knee
the knee
joint Xk , the 𝑋 , the
jointthigh’s thigh’s
center center ofXgravity
of gravity 𝑋 joint
tc , the hip , theXhip
h , and 𝑋 upper
jointthe , and the upper
body’s body’s
center of
center ofXgravity
gravity 𝑋 are respectively
bc are respectively written as equations:
written as equations:
𝑋 X= 𝑑 + 𝑙 − 𝑙 𝑐𝑜𝑠 𝜃 , (1)
sc = d + ( ls − lsc ) cos ( θ a ), (1)

𝑋 X = +d +
=k 𝑑 ls cos𝜃(θ a, ),
𝑙 𝑐𝑜𝑠 (2)
(2)
Xtc = Xk + (lt − ltc )cos(θ a + θk ), (3)
Xh = Xk + lt cos(θ a + θk ), (4)
Xbc = Xh + lbc cos(θ a + θk − θh ). (5)
Robotics 2023, 12, 16 4 of 16

The summation of moments about the pivot point (or the front wheels) required to
prevent falling backward should satisfy the condition as the equation:

ms gXsc + mt gXtc + mb gXbc > 0, (6)

indicating that the total moment is in the clockwise direction. In consideration of the links
over the knee joint, the knee extension moment sufficient to maintain the static equilibrium
is derived as the equation:

Mk = −mt g( Xtc − Xk ) − mb g( Xbc − Xk ). (7)

In consideration of the link over the hip joint, the hip extension moment that is sufficient to
maintain the static equilibrium is derived as the equation:

Mh = mb g( Xbc − Xh ). (8)

For the estimation of the knee and hip extension moments required during sit-to-stand
motion, our simulation is simplified by assuming that the hip angle is constant at 130◦ ,
whereas the knee extension angle varies from 105◦ to 60◦ . Subsequently, the knee angle is
assumed to be constant at 60◦ , whereas the angle of hip extension varies from 130◦ to 50◦ at
a very low speed to avoid considering the dynamics of motion. The ankle angle is assumed
to be constant at 73◦ throughout the transition period. According to the sitting geometry in
Figure 1, the existence of the wheel modules does not allow moving the feet behind to shift
the ground reaction force backward as in normal sit-to-stand motion [28–30]. Bending the
trunk with the hip flexion larger than usual is necessary. The link parameters used in our
simulation are shown in Table 1, in which the locations of the links’ center of gravity are
applied from [31], and the total mass of human M is 73 kg.

Table 1. Human model parameters applied from [31].

Link Mass (kg) Length (mm) CG Position (mm)

ltc = 0.01 × (40.95 × 422.2)
Thigh mt = 0.01 × (14.16 × 2) × M lt = 422.2
measured from hip joint
lsc = 0.01 × (44.59 × 434.0)
Shank ms = 0.01 × (4.33 × 2) × M ls = 434.0
measured from knee joint
lbc = [0.01/(43.46 + 6.94)] ×
603.3 + 242.9
Body-Head mb = 0.01 × (43.46 + 6.94) × M [43.46 × (48.62 × 603.3) +
from vertex to hip joint
6.94 × (603.3 + 49.98 × 242.9)]

The trajectories of the shank, thigh, and upper body during sit-to-stand motion are
simulated, as shown in Figure 2a. The shank is fixed with a constant ankle angle. The
knee is extended with a constant hip angle. Then, the hip is extended with a constant knee
angle. The variation in the body’s CG position against the joint angles is observed. The
body’s CG must be located anterior to the front wheels (X = 0) such that the total moment
computed via Equation (6) is always positive to prevent falling backward. The yellow plot
in Figure 2b shows that the risk of falling backward is high (small magnitude of the total
CW moment) when the upper body’s CG is posterior to the ankle. The ground reaction
force required beneath the feet to counter this total moment depends on the extent that the
legs and body’s CG are shifted forward during sit-to-stand motion.
Figure 2b shows that the maximum knee moment (approximately 120 N·m) is required
during the early phase of knee extension when the thigh and body’s CG are significantly
posterior to the knee joint. A high magnitude of the knee moment is required again during
the latter phase of hip extension when the body is upright, and the body’s CG is posterior
to the knee joint. The maximum hip moment (approximately 130 N·m) is required at the
latter phase of knee extension when the upper body’s CG is extremely anterior to the hip
Robotics 2023, 12, 16 5 of 16

joint. Notably, the maximum knee moment can be reduced if the motion is started from
a low knee angle. For example, approximately 100 N·m is sufficient if the knee begins
extending from 95◦ instead. In addition, the maximum hip moment is reduced because the
hip flexion to move the body’s CG forward has a small angle. If the arms’ weight is also
considered, the knee and hip extension moments will be reduced in accordance with the
limbs’ CG.

Figure 2. Simulated sit-to-stand motion: (a) Simplified kinematics of sit-to-stand motion in our study.
The ankle, knee, and hip joints shown by the red markers connect the shank, thigh, and body links.
The links’ CGs are shown by the green makers. The origin of the plot is the intersection between
the horizontal line crossing the ankle joint and the vertical line crossing the front wheel center;
(b) Simulated knee and hip angles are plotted in blue and orange, respectively. The total duration of
the sit-to-stand motion is approximately 2 min for this quasi-static simulation. The estimated knee
and hip extension moments are plotted in blue and orange, respectively, and the total clockwise
moment computed via Equation (6) is plotted in yellow.

3. Wheelchair–Exoskeleton Hybrid Robot

3.1. Experimental Prototype
The alpha prototype of the wheelchair–exoskeleton hybrid robot was built, as shown
in Figure 3. The lower-limb exoskeleton uses planetary-geared motors to support the hip
and knee joints, while the ankle joints are passive. The spring mechanism required to
support the small rotation of the ankle will be designed in the future. For the hip joints, the
T-Motor AK10-9 with a peak torque of 38 N·m and a weight of 820 g was used. For the knee
joints, the T-Motor AK70-10 with a peak torque of 24.8 N·m and a weight of 521 g was used.
The lower-back assembly connecting the left and right hip joints, thigh, shank, and foot
assemblies were built from CNC-cut carbon fiber parts (8–12 mm thick) reinforced with 3D
printed nylon (PA6-CF) parts. In the standing upright posture to prepare for walking or
stepping over obstacles, the left and right wheel modules are retracted to the lower legs of
the exoskeleton, as shown in Figure 4a, by using Actuonix L16-100-63-6-R linear actuators
with the maximum pulling force of 100 N and the stroke length of 100 mm (~5 s is required
for full-stroke traveling). The foldable chair legs made from carbon fiber tubes (25 mm
diameter) are pinned to the lower-back assembly posterior to the hip joints and driven by
two servo motors assembled on both sides via elastomer cords, as illustrated in Figure 4b.
For safe travel in wheelchair mode on smooth surfaces with low energy consumption, the
left and right wheel modules pinned to the lower legs of the exoskeleton are driven by the
7065 skateboard hub motors located as the front wheels. Meanwhile, the 75 mm casters are
used as the rear wheels to allow turning with the differential driving technique. The total
mass of the exoskeleton prototype is 20.2 kg, and the maximum width is 67.4 cm.
Robotics 2023, 12, 16 6 of 16

Figure 3. Alpha prototype of the wheelchair–exoskeleton hybrid robot in the standing upright
posture (for walking across obstacles) and the wheelchair mode (for safe and low-energy traveling on
smooth surfaces).

Figure 4. (a) Wheel module’s retraction mechanism using the Actuonix linear actuator; (b) Foldable
chair leg driven through the elastomer cord.

3.2. Control System Integration

The schematic in Figure 5 shows the control system implemented in the wheelchair–
exoskeleton hybrid robot prototype. For the operation of the robot in wheelchair mode, the
first microcontroller is programmed to receive commands from the user via a two-channel
analog joystick. The desired maneuvers (i.e., forward, backward, right turn, and left turn)
are achieved by controlling the speeds of the two skateboard hub motors by using the
electronic speed control circuits via pulse-width modulation (PWM). The second micro-
controller also sends PWM signals to control the position of the linear actuators retracting
the left and right wheel modules. For supporting sit-to-stand and stand-to-sit motions
in exoskeleton mode, the microcontroller receives the user’s commands via the second
analog joystick. The CANBUS shield is used for communication with the motors driving
the exoskeleton’s hip and knee joints. The T-Motor’s integrated controller computes the
reference torque τre f via the equation:
. .
τre f = k p (θd − θ ) + k d θ d − θ + τ f f , (9)
Robotics 2023, 12, 16 7 of 16

in accordance with the feedforward torque τ f f , the torsional stiffness k p , the damping
.
coefficient k d , the equilibrium position θd , and the reference joint velocity θ d . Once the
command package is sent to the motor, the feedback data consisting of the current position
.
θ, joint velocity θ, and torque are returned to the microcontroller.

Figure 5. Diagram of hardware integration. The first microcontroller with an analog joystick is used
to control the robot running in wheelchair mode. The skateboard hub motors operated at 24 V are
driven by the electronic speed control circuits receiving commands via pulse-width modulation
(PWM) signals. The second microcontroller is used to control the exoskeleton’s hip and knee motors
supporting sit-to-stand and stand-to-sit motions. The motors operated at 24 V receive the commands
and return their status via CANBUS. The two linear actuators for retracting the left and right wheel
modules are operated at 12 V and controlled via PWM signals.

The positive command received from the second joystick is related to the assistive
levels for supporting the knee and hip extension moments during sit-to-stand motion. The
reference joint velocities are commanded by the user while the damping coefficients of the
motors are set. The negative command from the joystick is used to support the stand-to-sit
motion. The equilibrium position of each joint is set to correspond to the standing posture.
Meanwhile, torsional stiffness is adjusted on the basis of assistive levels.

4. Sit-to-Stand and Stand-to-Sit Experiment

4.1. Experimental Setup and Procedure
An experiment was conducted, as seen in the flowchart shown in Figure 6a, on a healthy
male participant with a height of 172 cm and weight of 65 kg to evaluate the assistive
performance of the exoskeleton prototype in supporting sit-to-stand and stand-to-sit motions.
For the observation of muscle activities, DELSYS’s wireless EMG sensor (Trigno™) electrodes
were attached to both legs, as shown in Figure 6b. The electrodes covered the vastus lateralis
(VL), the bicep femoris (BF), the tibialis anterior (TA), and the gastrocnemius (GC) muscles.
The EMG signals were processed using the sliding root mean square filter with a window
length of 0.125 s and a window overlap of 0.0625 s. The current position and the torque of the
exoskeleton’s knee and hip motors were also recorded at the sampling frequency of 5 Hz.

Figure 6. (a) Experimental setup and procedure; (b) Surface electromyography (EMG) electrodes
Robotics 2023, 12, 16 8 of 16

Trigno™) were attached to both legs of the participant over the vastus lateralis (VL) in front of the
thigh, the bicep femoris (BF) behind the thigh, the tibialis anterior (TA) in front of the shank, and the
gastrocnemius (GC) behind the shank [32]; (c) The participant wearing the exoskeleton prototype
performed sit-to-stand and stand-to-sit motions while recording the EMG of the muscles, along with
the exoskeleton knee and hip motors’ position, velocity, and torque to evaluate assistive performance.

The sit-to-stand and stand-to-sit experiments were conducted three times. The partici-
pant performed the test without wearing the exoskeleton (basic sit-to-stand test) to obtain
reference results. For the observation of the effect of the inertia, damping, and friction
on the human by the robot’s structure, the participant wore the exoskeleton, as shown
in Figure 6c, and performed motions without actuating the motors (passive sit-to-stand
test). The motors were actuated to provide knee extension support during sit-to-stand
and stand-to-sit motions (robot sit-to-stand test) to evaluate the assistive performance of
the exoskeleton.

4.2. Recorded Position and Torque of the Exoskeleton Motors

The results in Figure 7 show the angle and moment recorded from the knee and hip
motors of the exoskeleton’s right leg during the experiment. In the blue plot, command
−1 (at 13.2 s) indicates the user’s intention of receiving the knee and hip extension moments
supporting the sit-to-stand motion. The knee angle (orange plot) decreases from 0 rad
to −1.54 rad (the motor rotates CCW) in less than 1 s. Simultaneously, the hip angle
(yellow plot) increases from 0 rad to 1.54 rad (the motor rotates CW). In consideration of
the moments, the peak torque magnitude of the knee motor (cyan plot) reaches 7.5 N·m,
whereas that of the hip motor (magenta plot) reaches 29 N·m. After standing upright, the
command 0 (at 13.6 s) is sent by the user. The stiffness of the knee and hip joints are set such
that the motors behave as torsion springs with the equilibrium positions corresponding
to the standing posture. The motor stiffness is sufficient to compensate for the weight
of the exoskeleton’s structure. The magnitude of torque when command 0 is sent (from
13.6 to 21.2 s) is related to each motor’s displacement from its equilibrium. The command
signal +1 is sent (from 21.4 s) to evaluate the assistive performance with the virtual spring
concept. While maintaining the equilibriums of the motors at the standing upright posture,
the higher stiffness is now set to the motors, as can be observed on the basis of the higher
torque-to-displacement ratios during the three cycles of squatting. During stand-to-sit
motion (from 32.0 s), the magnitudes of the knee and hip angles converge to the original
angles before standing, whereas those of the torque increase with the motors’ displacement
measured from their equilibriums in the standing posture.

Figure 7. Angle and moment of the exoskeleton’s right knee and hip motors.
Robotics 2023, 12, 16 9 of 16

4.3. Leg Muscle Activity Observed via Surface EMG

The EMG results shown in Figure 8 were recorded during the two cycles of the basic
sit-to-stand test without the exoskeleton. The activity of the VL muscle (quadriceps),
supporting the knee extension moment, was observed during standing, maintaining an
upright posture, and sitting. The peak values of standing and sitting are similar. The activity
of the BF muscle (hamstrings) supporting hip extension is observed during standing and
sitting and is almost zero without movement. The activity of the TA muscle supporting
dorsiflexion (to move the body’s CG forward) occurs prior to that of the other muscles. The
muscle is activated again to prevent falling backward during sitting down. The activity
of the GC muscle supporting plantar flexion is observed in the latter phase of standing to
balance the body in an upright posture and prevent falling forward. The EMG results shown
in Figure 9 were recorded during the passive sit-to-stand test. The activity of the VL muscle,
in this case, is similar to that in the absence of the exoskeleton. However, the second peak
during standing is notable compared with the first peak (more effort is required to support
the robot structure). A reduction in activity is observed while maintaining an upright
posture. The activity of BF is further extended during sitting because the participant needs
to confirm the stable sitting structure. Although the activities of TA and GC have higher
peak values, the collaboration between both legs is more consistent when the exoskeleton
is worn. The EMG results shown in Figure 10 were recorded during the robot sit-to-stand
test. The activity of the VL muscle is reduced in the second peak of standing because the
motors support knee extension.

Figure 8. RMS electromyography (EMG) of the right (blue) and left (green) muscles recorded during
the basic sit-to-stand test (without wearing the exoskeleton).

Figure 9. RMS electromyography (EMG) of the right (blue) and left (green) muscles recorded during
the passive sit-to-stand test (without actuation).
Robotics 2023, 12, 16 10 of 16

Figure 10. RMS electromyography (EMG) of the right (blue) and the left (green) muscles recorded
during the robot sit-to-stand test (with knee extension support).

5. Differential Driving Experiment

Another primary focus of this research is the wheel module and its functionality in
driving the exoskeleton. There are some scenarios that require steering or turning capability
while traveling in the wheelchair mode rather than transforming to the exoskeleton mode
before changing direction. The module is desired to pivot and turn with a minimal turning
radius, as a human turns, without moving forward. The steering mechanism relies on
the implementation of caster wheels in the rear and the two driving hub motors in the
front. The difference between the speed of both wheels leads to turning from the front as
the rear caster wheels move to accommodate the turn [33]. In theory, this situation means
that minimal wheel slips [34] occur and that the wheelchair can turn without a complex
steering mechanism.

5.1. Experimental Setup and Procedure

The two skateboard hub motors are driven by the ODrive motor controller. Through
the command prompt, the wheel speed and current limits are configured for safety. The
calibration of the ODrive is recommended with the wheels over the ground because the
low default current used for calibration is insufficient for creating high output torque.
The percentage of throttle received from the user via the analog joystick is converted
into the wheel speed of the motors. Preset movements, such as pivoting and turning
for the increased consistency of control in certain situations, are also programmed in
the microcontroller.
This experiment evaluated three scenarios, which were repeated for turning left and
right: pivoting, turning with one wheel, and turning with both wheels. All driving cases
were programmed and mapped to command buttons for the convenient repeatability of
consistent motor output. “Pivoting” is defined as one wheel spinning forward as the other
spins in the opposite direction to induce a rotation about the vertical axis at the midpoint
between both driving wheels. “Turning with one wheel” is defined as spinning only one
wheel to induce rotation about the other wheel, which is locked to not spin. “Turning with
both wheels” is defined as allowing both wheels to spin at different speeds to induce the
rotation of the wheelchair due to the difference in wheel speeds. The wheel speed used in
the pivoting and the one-wheel turning cases is 0.5 round/s. The wheel speeds used in the
case wherein both wheels turn are 1.0 and 0.5 round/s.
A top-down imaging setup is required to record the position and orientation of the
wheelchair under various conditions to determine the radius of curvature when the wheel
modules are turning. This setup is established by using a wide-lens phone camera mounted
at a fixed height from the floor and linked to a laptop via TeamViewer. Figure 11 shows that
the markers are attached to the floor in a 5 × 3 array, all within the frame and evenly spaced
out by 1 m for calibration and postprocessing. Another two moving markers are attached
to the wheelchair. Given the distortion of the wide lens, the calibration footage of checkered
Robotics 2023, 12, 16 11 of 16

images must be taken to rectify the images. OpenCV is used to rectify the image by setting
the camera distortion offset value. Then, by using the Tracker program built on the Open
Source Physics Java framework, dimension measurement is done on each reference point to
obtain the 4 × 2 m reference grid. This result is then verified by using a length calibration
video wherein the wheelchair is pushed around within the frame. Considering that the
markers on the wheelchair are a known value, the values obtained from the program can
be acquired with this value to confirm the accuracy of the Tracker program.

Figure 11. Markers are attached to the floor in a 5 × 3 array: (a) Original recording with distortion
from wide lines; (b) Recording processed via OpenCV to yield the 4 × 2 m reference grid.

The pivoting and one-wheel turning cases are conducted at the center reference point
for ease of observation, whereas the case wherein both wheels turn is conducted with an
offset from the center reference to accommodate the larger radius of curvature. The video
recordings are processed via OpenCV to eliminate distortion from wide lines. Figure 12
illustrates that the Tracker is used to track the wheelchair as it turns in each case. Sample
points are taken at both caster wheels once every six frames (or 0.2 s per point, given that
the video is 30 frames per s).

Figure 12. Tracking of the coordinates of both caster wheels during turning.

5.2. Experimental Results

Figure 13 shows that the CG position and orientation of the wheelchair are computed
from the time-varying coordinates of both caster wheels via kinematic relationships as
adapted from a previous study [35]. The left and right wheel modules can enable turning,
given the layout of the design. For all the turning strategies, the arc trajectories of the
wheelchair–human’s CG, left, and right caster wheels are observed to be smoother without
the occupant. The midpoint between the front driving wheels is slightly shifted behind
during pivoting. In consideration of the slope at the time t0 of the left caster’s arc trajectory
(cyan) during turning with one wheel, the delay of caster steering is notable with the
occupant. The position shift is also observed on the locked (front left) wheel. The CG
trajectory lies between the left and the right caster wheels’ trajectories during turning with
both wheels.
Robotics 2023, 12, 16 12 of 16

Figure 13. Examples of the wheelchair turning trajectories with/without the occupant during pivot
left turns, one-wheel left turns, and two-wheel right turns. Blue markers indicate the front driving
hub motors. Black, cyan, and pink markers and curves represent the wheelchair–human’s CG,
left, and right caster wheels, respectively.

The magnitudes of the CG’s velocity and acceleration are estimated and plotted in
Figure 14. As a result of the PID control, the damped vibration response to the step
reference can be observed in the velocity plots. In all scenarios, the settling times appear to
be smaller without the occupant. The lower inertia system (without the passenger mass)
is advantageous when the limited control input (driving torque of the hub motors) is
considered. The effect of the occupant on the response delay is notable during turning with
one wheel, which corresponds to the delay in caster steering. The lower percentages of the
maximum overshoot are observed during turning with both wheels.
The Taubin algebraic method is applied to the CG trajectories to determine the best-fit
circles representing the turning radius of the wheelchair for each case, as shown in Figure 15.
As expected, pivoting provides the smallest radius of curvature because the axis of rotation
(the midpoint between the two front driving wheels) is close to the CG position. The CG
trajectory is not a smooth circle with the occupant, especially when the radius of curvature
is small. Except for that, in the pivoting case, the existence of the passenger slightly reduces
the radius.
Robotics 2023, 12, 16 13 of 16

Figure 14. Estimated CG velocity and acceleration with/without the occupant during pivot turns,
one-wheel turns, and two-wheel turns.

Figure 15. Best fit circles, based on the Taubin algebraic method, showing the turning radius of
wheelchair with/without the occupant during pivot turns, one-wheel turns, and two-wheel turns.
Robotics 2023, 12, 16 14 of 16

6. Conclusions and Future Work

This paper presents a design concept of a wheelchair–exoskeleton hybrid robot that
can transform between sitting and standing configurations as a possible alternative for
assisting the elderly or people with disabilities. The wheelchair mode allows traveling on
smooth surfaces at high speeds with increased safety and reduced energy consumption.
The compact wheel modules using skateboard hub motors can be retracted to prepare for
walking. Given that the back-drivability of the exoskeleton joints is the primary concern
for a safe physical human–robot interaction, the wheelchair parameters are designed on
the basis of the simplified human model such that the actuators’ torque limitation is still
sufficient to support the sit-to-stand motion. The wheelchair–exoskeleton hybrid robot
prototype was built and tested on a healthy male volunteer to investigate the characteristics
of the device for further improvement.
During sit-to-stand and stand-to-sit motions, the wearer must bend their trunk forward
adequately to locate the body’s CG anterior to the front wheels to prevent falling backward.
For users who are able to operate their upper limbs, crutches might be a possible solution
to reduce the requirement of bending the body forward. For improving the design, shifting
the pivot position behind is beneficial in terms of standing stability. The parallel elastic
concepts [36–38] can be applied to safely enhance the torque capacity of the exoskeleton
joints. A motion controller should be developed for walking and stair climbing [39,40].
Analytical and experimental studies on the integral of power consumption over time for the
wheelchair and the exoskeleton modes will be considered. For evaluating the exoskeleton
performance at different assistive levels, the quantitative analysis of EMG collected from
experiments conducted under the controlled condition with appropriate measurement of
body kinematics is required. The velocity profile and desired radius of curvature can be
mapped on the basis of the results of the differential driving experiment.

Author Contributions: Conceptualization, R.C.; methodology, R.C., S.P. and S.T.; software, R.C. and
S.P.; validation, R.C., S.P. and S.T.; formal analysis, R.C. and S.P.; investigation, R.C., S.P. and S.T.;
resources, R.C.; data curation, R.C., S.P. and S.T.; writing—original draft preparation, R.C., S.P. and
S.T.; writing—review and editing, R.C.; visualization, R.C., S.P. and S.T.; supervision, R.C.; project
administration, R.C.; funding acquisition, R.C. All authors have read and agreed to the published
version of the manuscript.
Funding: This research was partially funded by Thailand Science Research and Innovation Fund,
Chulalongkorn University (CU_FRB65_ind (14)_162_21_28), and by the National Research Council
of Thailand.
Data Availability Statement: The data presented in this study are available on request from the
corresponding author. The data are not publicly available due to privacy.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Sarraj, A.R.; Massarelli, R. Design history and advantages of a new lever-propelled wheelchair prototype. Int. J. Adv. Robot Syst.
2011, 8, 12–21. [CrossRef]
2. Sasaki, K.; Eguchi, Y.; Suzuki, K. Step-climbing wheelchair with lever propelled rotary legs. In Proceedings of the 2015 IEEE/RSJ
International Conference on Intelligent Robots and Systems, Hamburg, Germany, 28 September–3 October 2015; pp. 6354–6359.
3. Lawn, M.J.; Ishimatsu, T. Modeling of a stair-climbing wheelchair mechanism with high single step capability. IEEE Trans. Neural
Syst. Rehabil. Eng. 2003, 11, 323–332. [CrossRef] [PubMed]
4. Chatterjee, P.; Lahiri, N.; Bhattacharjee, A.; Chakraborty, A. Automated hybrid stair climber for physically challenged people. In
Proceedings of the International Conference on Electronics, Materials Engineering & Nano-Technology, Kolkata, India, 24–26
September 2021; pp. 1–4.
5. Suryanto, M.F.I.; Badriawan, N.A.; Ningrum, E.S.; Binugroho, E.H.; Satria, N.F. Balance control on the development of electric
wheelchair prototype with standing and stair climbing ability with tracked-wheel mechanism. In Proceedings of the 2018 International
Electronics Symposium on Engineering Technology and Applications, Bali, Indonesia, 29–30 October 2018; pp. 43–47.
6. Munton, J.S.; Ellis, M.I.; Chamberlain, M.A.; Wright, V. An investigation into the problems of easy chairs used by the arthritic and
the elderly. Rheum. Rehabil. 1981, 20, 164–173. [CrossRef] [PubMed]
Robotics 2023, 12, 16 15 of 16

7. Mano, Y.; Sakakibara, T.; Takayanagi, T. Kinesiological analysis of standing-up movement. Excerpta Med. Int Congr. Ser. 1988,
804, 503–512.
8. Levo: Products. Available online: https://www.levo.ch/products (accessed on 11 December 2022).
9. Superior ME. Available online: http://Superiorstanding.mamutweb.com/subdet1.htm (accessed on 11 December 2022).
10. Eguchi, Y.; Kadone, H.; Suzuki, K. Standing mobility device with passive lower limb exoskeleton for upright locomotion. IEEE
ASME Trans. Mechatron. 2018, 23, 1608–1618. [CrossRef]
11. Hayashi, T.; Kawamoto, H.; Sankai, Y. Control method of robot suit HAL working as operator’s muscle using biological and
dynamical information. In Proceedings of the 2005 IEEE/RSJ International Conference Intelligent Robots and Systems, Edmonton,
AB, Canada, 2–6 August 2005; pp. 3063–3068.
12. Mertz, L. The next generation of exoskeletons: Lighter, cheaper devices are in the works. IEEE Pulse 2012, 3, 56–61. [CrossRef]
[PubMed]
13. Zeilig, G.; Weingarden, H.; Zwecker, M.; Dudkiewicz, I.; Bloch, A.; Esquenazi, A. Safety and tolerance of the ReWalk™ exoskeleton
suit for ambulation by people with complete spinal cord injury: A pilot study. J. Spinal Cord Med. 2012, 35, 96–101. [CrossRef]
14. Barbareschi, G.; Richards, R.; Thornton, M.; Carlson, T.; Holloway, C. Statically vs dynamically balanced gait: Analysis of a
robotic exoskeleton compared with a human. In Proceedings of the 2015 IEEE/EMBS Annual International Conference, Milan,
Italy, 25–29 August 2015; pp. 6728–6731.
15. Duddy, D.; Doherty, R.; Connolly, J.; McNally, S.; Loughrey, J.; Faulkner, M. The effects of powered exoskeleton gait training on
cardiovascular function and gait performance: A systematic review. Sensors 2021, 21, 3207. [CrossRef]
16. Borisoff, J.F.; Mattie, J.; Rafer, V. Concept proposal for a detachable exoskeleton-wheelchair to improve mobility and health. In
Proceedings of the 2013 IEEE International Conference on Rehabilitation Robotics, Seattle, WA, USA, 24–26 June 2013.
17. Shankar, T.; Dwivedy, S.K. A hybrid assistive wheelchair-exoskeleton. In Proceedings of the 2015 International Convention on
Rehabilitation Engineering and Assistive Technology, Singapore, 5–7 November 2015.
18. Song, Z.; Tian, C.; Dai, J.S. Mechanism design and analysis of a proposed wheelchair-exoskeleton hybrid robot for assisting
human movement. Mech. Sci. 2019, 10, 11–24. [CrossRef]
19. Chaichaowarat, R.; Nishimura, S.; Krebs, H.I. Macro-mini linear actuator using electrorheological-fluid brake for impedance
modulation in physical human–robot interaction. IEEE Robot Autom. Lett. 2022, 7, 2945–2952. [CrossRef]
20. Chaichaowarat, R.; Nishimura, S.; Krebs, H.I. Design and modeling of a variable-stiffness spring mechanism for impedance
modulation in physical human–robot interaction. In Proceedings of the 2021 IEEE International Conference on Robotics and
Automation, Xi’an, China, 30 May–5 June 2021; pp. 7052–7057.
21. Millington, P.J.; Myklebust, B.M.; Shambes, G.M. Biomechanical analysis of the sit-to-stand motion in elderly persons. Arch. Phys.
Med. Rehabil. 1992, 73, 609–617. [PubMed]
22. Chorin, F.; Cornu, C.; Beaune, B.; Frere, J.; Rahmani, A. Sit to stand in elderly fallers vs non-fallers: New insights from force
platform and electromyography data. Aging Clin. Exp. Res. 2016, 28, 871–879. [CrossRef]
23. Pal, A.R.; Pratihar, D.K. Estimation of joint torque and power consumption during sit-to-stand motion of human-being using a
genetic algorithm. Procodia Comp. Sci. 2016, 96, 1497–1506. [CrossRef]
24. Glowinski, S.; Ptak, M. A kinematic model of a humanoid lower limb exoskeleton with pneumatic actuators. Acta Bioeng. Biomech.
2022, 24, 145–157. [CrossRef]
25. Huo, W.; Mohammed, S.; Amirat, Y.; Kong, K. Active impedance control of a lower limb exoskeleton to assist sit-to-stand
movement. In Proceedings of the 2016 IEEE/RSJ International Conference on Intelligent Robots and Systems, Daejeon, Republic
of Korea, 9–14 October 2016; pp. 3530–3536.
26. Tsukahara, A.; Kawanishi, R.; Hasegawa, Y.; Sankai, Y. Sit-to-stand and stand-to-sit transfer support for complete paraplegic
patients with robot suit HAL. Adv. Robot 2010, 24, 1615–1638. [CrossRef]
27. Fattah, A.; Agrawal, S.K.; Catlin, G.; Hamnett, J. Design of a passive gravity-balanced assistive device for sit-to-stand tasks. J.
Mech. Des. 2006, 128, 1122–1129. [CrossRef]
28. Elibol, E.; Calderon, J.; Llofriu, M.; Moreno, W.; Weitzenfeld, A. Analyzing and reducing energy usage in a humanoid robot
during standing up and sitting down tasks. Int. J. Hum. Robot 2016, 13, 1650014. [CrossRef]
29. Kim, J.; Yang, J.; Yang, S.T.; Oh, Y.; Lee, G. Energy-efficient hip joint offsets in humanoid robot via Taguchi method and bioinspired
analysis. Appl. Sci. 2020, 10, 7287. [CrossRef]
30. Elibol, E.; Calderon, J.; Llofriu, M.; Quintero, C.; Moreno, W.; Weitzenfeld, A. Power usage reduction of humanoid standing
process using Q-learning. In RoboCup 2015: Robot World Cup XIX, Lecture Notes in Computer Science; Almeida, L., Ji, J., Steinbauer,
G., Luke, S., Eds.; Springer: Cham, Switzerland, 2015; 9513; pp. 251–263.
31. De Leva, P. Adjustments to Zatsiorsky–Seluyanov’s segment inertia parameters. J. Biomech. 1996, 29, 1223–1230. [CrossRef]
[PubMed]
32. Chaichaowarat, R.; Granados, D.F.P.; Kinugawa, J.; Kosuge, K. Passive knee exoskeleton using torsion spring for cycling assistance.
In Proceedings of the 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems, Vancouver, BC, Canada, 24–28
September 2017; pp. 3069–3074.
33. Chaichaowarat, R.; Wannasuphoprasit, W. Full-slip kinematics based estimation of vehicle yaw rate from differential wheel
speeds. KSAE Int. J. Automot. Tech. 2016, 17, 81–88. [CrossRef]
Robotics 2023, 12, 16 16 of 16

34. Chaichaowarat, R.; Wannasuphoprasit, W. Kinematics-based analytical solution for wheel slip angle estimation of a RWD vehicle
with drift. Eng. J. 2016, 20, 89–107. [CrossRef]
35. Chaichaowarat, R.; Wannasuphoprasit, W. Wheel slip angle estimation of a planar mobile platform. In Proceedings of the 1st
International Symposium on Instrumentation, Control, Artificial Intelligence, and Robotics, Bangkok, Thailand, 16–18 January
2019; pp. 163–166.
36. Chaichaowarat, R.; Kinugawa, J.; Kosuge, K. Unpowered knee exoskeleton reduces quadriceps activity during cycling. Engineering
2018, 4, 471–478. [CrossRef]
37. Chaichaowarat, R.; Kinugawa, J.; Kosuge, K. Cycling enhance knee exoskeleton using planar spiral spring. In Proceedings of the
2018 IEEE/EMBS Annual International Conference, Honolulu, HI, USA, 18–21 July 2018; pp. 3206–3211.
38. Chaichaowarat, R.; Kinugawa, J.; Seino, A.; Kosuge, K. A spring-embedded planetary-geared parallel elastic actuator. In
Proceedings of the 2020 IEEE/ASME International Conference on Advanced Intelligent Mechatronics, Boston, MA, USA, 6–9 July
2020; pp. 952–959.
39. Javadi, A.; Chaichaowarat, R. Position and stiffness control of an antagonistic variable stiffness actuator with input delay using
super-twisting sliding mode control. Nonlinear Dyn. 2022, 1–23. [CrossRef]
40. Chaichaowarat, R.; Macha, V.; Wannasuphoprasit, W. Passive knee exoskeleton using brake torque to assist stair ascent. In
Proceedings of the 2020 IEEE Region 10 Conference, Osaka, Japan, 16–19 November 2020; pp. 1165–1170.

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.