2017 International Conference on Rehabilitation Robotics (ICORR)

QEII Centre, London, UK, July 17-20, 2017.

Design of an Exoskeleton Ankle Robot for Robot-Assisted Gait

Training of Stroke Patients*
Ling-Fung Yeung1, Corinna Ockenfeld2, Man-Kit Pang3, Hon-Wah Wai3, Oi-Yan Soo4,
Sheung-Wai Li5 and Kai-Yu Tong1

Abstract— Lower Limb Exoskeleton robot that can facilitate

stair walking is a big challenge, most systems could only provide
level ground walking. In this study, a lightweight (0.5kg at
ankle, 0.5kg at waist for control box) and autonomous
exoskeleton Ankle Robot was proposed to provide power
assistance for gait training of chronic stroke patients and it can
facilitate three walking conditions in real-time: (1) level
walking, (2) stair ascending, and (3) stair descending. Chronic
stroke patients (n=3) with drop foot gait deficit and moderate
motor impairment were recruited to evaluate the system under
different walking conditions (Functional Ambulatory Category:
FAC=4.7±0.5 and Fugl-Meyer Assessment for lower-extremity:
FMA-LE=13.7±2.9). The system consisted of a specially
designed carbon fiber AFO, servomotor, gear transmission
system, IMU and force sensors, and control box. The IMU
sensors embedded in the shank measured acceleration and
angular velocity to identify distinct features in leg tilting angle
and leg angular velocity between the three walking conditions.
The results showed the powered ankle dorsiflexion assistance
could reduce dropped foot of the stroke patients in swing phase
and provide better gait pattern. A demo of the ankle robot will
be conducted in the conference.
I. INTRODUCTION
Stroke is a cerebrovascular accident with high incidence
and disability rate. In 2010, the absolute number of global
stroke survivors was 33 million, significantly increased by
84% since 1990 [1]. The immediate effect of stroke is the
damage of motor pathway descending down from the central
nervous system (CNS) to the muscles, causing loss of motor
control and muscle weakness in the contralateral side.
Imbalance of excitatory and inhibitory muscle stimulation
from CNS could lead to hypertonic spasticity and contracture Figure 1. (A) Structure of the Ankle Robot, with arrows indicating the
at the affected body joints, resulting in the velocity-dependent directions of linear acceleration X, Y, Z, and the angular velocity Z’,
resistance to joint movement [2, 3]. Walking ability is an measured by the motion sensor. (B) A stroke patient wearing the Ankle
important predictor of active participation in complex and Robot on his affected side (his right ankle). (C) A stroke patient wearing the
social daily activities, but only 35% of chronic stroke Ankle Robot climbing up a staircase.
survivors were as active as before first stroke [4].
At the ankle joint, muscle weakness in dorsiflexor and
spasticity in plantar-flexor would cause the foot constantly
pointing downward and dragging on the ground during
* Research supported by ITF grant (GHP/001/12) from the Innovation walking, a condition known as drop foot gait. This gait
and Technology Commission, HKSAR. abnormality leads to inadequate propulsion at initial swing,
Ling-Fung Yeung and Kai-Yu Tong are with the Division of Biomedical insufficient foot clearance during late swing, and inefficient
Engineering, Department of Electronic Engineering, The Chinese University shock absorption producing foot slap sound at initial contact
of Hong Kong, Shatin, HK (Kai-Yu Tong is the corresponding author to [5]. Stroke patients tend to compensate by hip hiking or
provide phone: +852 3943 8454; e-mail: kytong@cuhk.edu.hk).
Corinna Ockenfeld is with the Interdisciplinary Division of Biomedical
circumduction of the paretic limb, with exaggerated positive
Engineering, The Hong Kong Polytechnic University, Hong Kong. work associated with raising trunk during swing phase. Since
Man-Kit Pang and Hon-Wah Wai are with the Industrial Centre, The they are not confident walking with the weak support from the
Hong Kong Polytechnic University, Hong Kong. paretic limb, they spend shorter stance duration and advance a
Oi-Yan Soo is with the Department of Medicine & Therapeutics, The shorter step length in paretic side. These inefficient
Chinese University of Hong Kong, Hong Kong.
Sheung-Wai Li is with the Division of Rehabilitation, Department of
asymmetric gait pattern contribute to slower walking speed,
Medicine, The University of Hong Kong, Hong Kong. increasing falling risk, and greater energy expenditure [5].

978-1-5386-2295-7/17/$31.00 ©2017 IEEE 211

 Ankle foot orthoses (AFO) are orthotic devices that are
externally applied on the ankle joint to correct drop foot gait
by restricting the joint position and movement. Conventional
AFOs are fabricated in thermoplastic as an unbroken L-shaped
brace, or articulated AFO with passive resistive elements
across ankle joint, such as springs, oil dampers, and
mechanical locks. These AFO fix the ankle joint in neutral
position and prevent drop foot during swing phase of walking.
However, they would restrict the ankle range of motion (RoM)
in another gait phases and would cause unnatural gait pattern
with poor dynamic balance [6].
More sophisticated AFO designs are integrated robotic
control system to automatically and dynamically facilitate
ankle locomotion. Most robotic AFO acquired gait pattern
from various kinetic and kinematic sensor outputs and
classified gait phase using finite-state machine (FSM). Smart
AFOs developed in Osaka University [7] and in Halmstad
University [8] used computer-controllable resistive elements
such as magneto-rheological (MR) damper to adjust joint
impedance dynamically during walking. Although they
successfully maintained foot clearance in swing and braking
in heel strike for hemiplegic patient, these AFOs only provide
passive torque, while active assistance in ankle plantar-flexion
was found to reduce metabolic cost of walking [9]. Research
team in the University of Michigan [10] developed a robotic
AFO using electromyography (EMG) to detect the user
intention of muscle contraction for proportional myoelectric
control (PMC) of the pneumatic artificial muscle (PAM) in
both dorsiflexion and plantar-flexion directions. However, it is
difficult to be applied for stroke patients because of their weak
muscles on the affected side [11], and co-contraction in the
agonist and antagonist muscles would render the PMC output
unstable and dangerous in the case of walking [3]. Robotic
AFOs developed at MIT [12], Yonsei University [13], and
Arizona State University [14] used series elastic actuator
(SEA) to drive the ankle joint in predefined position
trajectories or torque profiles. Despite high assistive torque
output, they were heavy and bulky ([14] had torque output up
to 60Nm and weighted 1.75kg; [12] weighted 2.6kg), and
required tethered power supply. Wearing these heavy, tethered
devices for long-distance walking (LDW) might increase
energy cost of walking [9] and might impose great hindrances
to the feasibility of bringing these robotic AFO into daily-wear
application or routine clinical practices [11, 15, 16].
Up to date, few robotic AFOs were developed and
evaluated as portable devices for robot-assisted gait training
(RAGT) that targeting at the ankle joint [17]. The robotic AFO
developed at University of Illinois [18] (maximum torque
12Nm, weighted 1.9kg at ankle, 1.2kg at waist) was a
fully-autonomous system with portable power supply (gas
container) and bidirectional pneumatic rotatory actuator. The
powered assistance profiles covered over-ground walking as
well as stairs and inclinations, but the control algorithm had a
one-step delay due to hardware limitation [19]. To achieve
effective RAGT for practical applications, the robotic AFO
must be able to provide correct afferent feedback in response
to the active voluntary walking of the user. The high-intensity
and repetitive nature of the robot would promote the
experience-driven adaptation of the damaged motor pathway
in CNS to the programmed gait pattern via brain plasticity [20,
21].
212
In this study, we proposed a lightweight and autonomous
Ankle Robot (0.5kg at ankle for the robotic AFO, 0.5kg at
waist for the control box and battery) that can provide
powered dorsiflexion/plantar-flexion assistance to
level-ground walking and stair ascending/descending. The
system utilized force sensors to identify gait phase, and used
motion sensors to measure leg tilting angle and leg angular
velocity for classifying walking conditions in real-time for the
current step. It was designed as a gait training device to
facilitate stroke patients in varying terrains according to their
walking intention. Design considerations and the experimental
results of testing the Ankle Robot were discussed.
II. DESIGN CONSIDERATIONS
The follow design criteria were considered when we
developed the Ankle Robot:
• Lightweight: to reduce the metabolic burden of
carrying extra load on the body for LDW.
• Small size: to fit in the limited space around the ankle
without obstructing gait movement.
• High torque output: to generate sufficient torque to
lift up or push off the drop foot.
• Reliable: to classify gait phase accurately and reliably
in real-time walking.
• Therapeutic effect: to provide correct afferent
feedback to the user during gait training.
A. Exoskeleton AFO
The Ankle Robot was an articulated AFO fabricated as a
monolithic carbon-fiber reinforced polymer (CFRP) brace,
which was lightweight but strong (Fig. 1). The upper leg brace
covered the anterior tibia and fastened around the shank by
ratcheting buckled straps (M2 inc., USA) and the lower foot
sole was inserted in the shoe of user. Soft paddings were used
to provide comfort and improve fitting. The two CFRP parts
were mounted on the gearbox with hinged ankle joint allowing
one degree-of-freedom (1 DOF) movement in the sagittal
plane about the lateral malleolus, with RoM 20˚ dorsiflexion
and 30˚ plantar-flexion limited by mechanical stoppers.
B. Actuation System
Torque output 13.53Nm was required to move spastic
ankle to 20˚ dorsiflexion from dropped foot position [22]. We
have chosen an off-the-shelf brushless DC motor Dynamixel
MX-106R servomotor (ROBOTIS, South Korea) with built-in
PID control, maximum torque output 10Nm, maximum
angular speed 55 rpm with gear reduction ratio 1:225,
operating at 12V, 5.2A. The motor was installed in front of the
anterior tibia where there was no interference with lower-limb
movement (Fig. 1). Motor torque was transmitted to the ankle
joint at lateral malleolus through gear transmission with 1:1.67
gear ratio, which further amplified the Ankle Robot maximum
torque output to 16.7Nm. The motor and gears were encased
in a stainless steel gearbox with plastic shielding protecting
the user from movable parts. A 12V 1800mAh Lithium
Polymer (LiPo) battery supplied power to both the actuator
and the microprocessor unit (MPU), and had capacity enough
for 5-hour operation of Ankle Robot without recharging based
on power drawn from the electronics.
 The Ankle Robot (AFO plus the gearbox) weighted 0.5kg
at ankle, and the control box (LiPo battery plus MPU)
weighted 0.5kg at waist. The LiPo battery was carried at the
waist inside the control box away from the distal limb, where
loading at distal limb would increase metabolic rate in walking
[23].
C. Sensory System
All sensors were attached on the Ankle Robot device itself,
including two force sensitive resistors (FSR) and an inertial
measurement unit (IMU) (Fig. 1). Two FSR-402 (Interlink
Electronics, USA) embedded in the foot sole under forefoot
and heel respectively to detect foot loading pattern as an
ON/OFF switch. Resistance across the FSR dropped if it was
loaded, which could be measured using voltage divider and be
identified by passing a certain threshold. Hysteresis was added
to the threshold crossing detection algorithm to avoid unstable
polarity switching in foot loading. Figure 2. Block diagram of the finite state machine (FSM) of the Ankle Robot,
which identified two gait phases (swing phase and stance phase) and three
TM walking conditions (level walking, stair ascend, and stair descend).
An MPU6050 6-axis MotionTracking device with
accelerometer and gyroscope (InvenSense, USA) was
trailing affected leg had to tilt forward, with the tilting angle
mounted on the leg brace and oriented in a direction to
proportional to the step height. It agreed with previous
measure the leg linear accelerations with backward direction
research of gait analysis in stair walking pattern suggested
as +X, upward direction as +Y, and the leg angular velocity on
there was a certain leg tilting angle related to a specific
the sagittal plane (forward rotation about +Z) (Fig. 1). Signals
walking condition [28]. (2) The low leg angular velocity at
were sampled at 30Hz, low-pass filtered at 4Hz cut-off
pre-swing phase discriminated stair ascending from level
frequency, in which 30Hz sampling rate can preserve 99% of
walking, as the leg was brought upward in stair ascending,
the signal power in gait [24]. Previous studies demonstrated
instead of rolling forward about the ankle joint during terminal
the feasibility of using accelerometer and gyroscope to
stance of level walking.
classify level-ground walking and stair ascending/descending
of dropped foot patients [25, 26]. Leg tilting angle was Motor output profile for different walking conditions were
computed from accelerometer and gyroscope signals. It was designed based on previous studies of normal swing phase gait
found that the leg tilting angle and the leg angular velocity in pattern in level walk [29], stair ascending [30, 31], and stair
the sagittal plane could be used to classify walking conditions, descending [31]. The level of assistance could be adjusted by
whereas motion in frontal and transverse planes were calibration. Both level walking and stair ascending had ankle
negligible and irrelevant to the classification of walking or dorsiflexion assistance for foot clearance, while higher
stair ascending/descending. powered dorsiflexion assistance was provided for stair
climbing to lift the foot up to the next step; stair descending
D. Control Algorithm
had ankle plantar-flexion assistance at the late swing phase to
The user of the Ankle Robot was recommended to initiate prepare for weight acceptance at initial contact when the leg
their walking or stair ascending/descending with their descended to the next step. Short bursts of ankle
unaffected leg leading the affected leg. The control algorithm plantar-flexion assistance were introduced to both the push-off
would identify the pre-swing phase using foot loading pattern for propulsion and the weight acceptance phase for shock
and would then classify the walking condition based on the absorption.
signals obtained from the IMU sensor. The classification was
triggered when both heel FSR and forefoot FSR were Computations were performed using Arduino Pro Mini
unloaded, i.e. the foot lifted off from the ground, indicating the with ATmega328-5V-16MHz microprocessor (Atmel, USA).
swing phase began (Fig. 2). Motor output profiles were This MPU board was placed inside the control box together
executed during swing phase for ankle powered assistance. with the LiPo battery. The control box at the waist level was
The end of the gait cycle was the weight acceptance and was connected to the sensors and actuator on the AFO. The Ankle
identified when either one of the heel or forefoot FSR was Robot could operate fully independently, with option to
loaded. Short bursts of plantar-flexion assistance were communicate in wireless to computer or smart phone using
generated to help push-off propulsion and weight acceptance. Bluetooth. The wireless communication allowed therapists to
conveniently record sensor signals and perform system
Block diagram of the FSM for the Ankle Robot control configuration.
algorithm is showed in Fig. 2. FSM was utilized extensively
for intent recognition of amputee gait in powered lower-limb III. METHODS
prosthesis, such as C-Leg [27]. The classification rule of This pilot study evaluated the performance of the Ankle
walking conditions was two folds: (1) The large leg tilting Robot to classify three walking conditions (level walk, stair
angle at pre-swing phase indicated stair descending, as the ascend/descend) using the control algorithm, and to provide
leading unaffected leg descended to the lower step, the powered assistance to the ankle joint for foot clearance during
swing phase in three chronic stroke patients. Their
demographic characteristics are showed in Table I. They had

213
moderate level of motor impairment in the affected ankle and
walked with drop foot gait. They used walking cane as 90 Stance Swing Stance Swing (a)

Leg Forward Tilting Angle

Phase Phase Phase Phase
walking assistance, but were capable of standing and walking 70
independently without manual contact assistance for an Level Walk

(Degree)
50
extended period of time. Clinical assessment scores were Stair Ascend
provided in Table II. They have provided informed consent to 30 Stair Descend
participate in this pilot study. 10
A. Testing Control Algorithm -10
In this experiment, the stroke patients put on the Ankle 250
Robot and walked with the robot without powered assistance. Stance Swing Stance Swing
(b)

in Sagittal Plane (Degree/s)

They took as much time as they needed for adaptation to the 150 Phase Phase Phase Phase

Leg Angular Velocity

added weight on the ankle. Then stroke patients performed
10-minute level walking and 10-minute stair ascend/descend
with the unpowered Ankle Robot. During walking, the Ankle
Robot computed and recorded the leg tilting angle and the leg
angular velocity signals.
B. Testing Powered Assistance
In this experiment, the ankle joint angles of the three stroke
patients were recorded under three conditions: (1) walking
without wearing the Ankle Robot (NoRobot), (2) walking with
the Ankle Robot without powered assistance (NoPower), and
(3) walking with the Ankle Robot with powered assistance
(Powered). A six-camera motion capture system (Vicon
Nexus, Oxford Metrics) was used for the recording of
kinematic gait pattern and the ankle joint angle was computed
using BodyBuilder of the Vicon system. The gait patterns
were normalized to 100% gait cycle and averaged among the
stroke patients for three walking steps.
IV. RESULTS
Typical form of the sensor signals of the three walking
conditions were showed in Fig. 3. There were some distinct
differences between the three walking conditions at the
50 Level Walk
Stair Ascend
-50 Stair Descend
-150
-250
Figure 3. Sensor signals of a single subject showed the distinct differences
between three walking conditions: level walk (dark solid line), stair ascend
(red solid line), and stair descend (red dashed line). (a) Leg tilting angle
distinguished stair descend condition; and (b) leg angular velocity in sagittal
plane distinguished level walk condition. The time frames for classification of
walking condition were indicated by arrows.
Results showed that the powered ankle dorsiflexion
assistance from the Ankle Robot could reduce the dropped
foot of the stroke patients in the late swing phase of the
affected side (Fig. 4). Wearing the unpowered Ankle Robot
also reduced dropped foot but with a lesser extent, which
showed effects similar to an articulated passive AFO. In this
gait pattern, the stroke patients did not require to raise their
affected side exaggeratedly for foot clearance in swing phase,
which might lead to a more efficient gait pattern.
25
Ankl Angle in Affected Side

Pre-swing Phase
pre-swing phase: (1) The stair descending condition had larger 20
leg tilting angle than that in level walking and stair ascending, 15
(Degree)

(2) The level walking condition had larger leg angular velocity 10
than stair ascending and descending. This 5
results agreed with our control algorithm. It showed that we 0
can use the control algorithm to classify the walking condition -5 0 10 20 30 40 50 60 70 80 90
in real-time using the IMU sensor signals. -10 Gait Cycle %
TABLE I. DEMOGRAPHIC CHARACTERISTICS OF STROKE PATIENTS.
-15
(a) -20
Age Affected Stroke Onset
ID (year) Gender Side Stroke Type (year) 25
Ankle Angle in Unaffected Side

A 72 M Left Ischemic 2 20
Pre-swing Phase
B 58 F Right Hemorrhage 3.5 15
10
C 58 M Right Hemorrhage 13.5
(Degree)

5
0
TABLE II. CLINICAL SCORES OF STROKE PATIENTS. -5 0 10 20 30 40 50 60 70 80 90
Timed 6MWT -10 Gait Cycle %
ID FAC FMA-LE MAS BBS 10mWT (m/s) (m) -15
A 4 10 2 44 0.37 – 0.49 116.4 -20 Walk without Ankle Robot
B 5 17 2 53 0.30 – 0.37 99.8 Walk with Unpowered Robot
(b)
C 5 14 1 50 0.50 – 0.60 178.2 Walk with Powered Robot

NOTE- FAC=Function Ambulation Categories, FMA-LE=Fugl-Meyer Assessment for Figure 4. Averaged ankle joint angle of stroke patients (n=3) with ±1 SD in
Lower-Extremity, MAS=Modified Ashworth Scale, BBS=Berg Balance Scale, 10mWT=Ten-minute three conditions: walking without Ankle Robot, walking with unpowered
Walk Test (from self-selected speed to fast speed), 6MWT=Six-minute Walk Test.
robot, and walking with powered robot.

214
 However, no evidence of the Ankle Robot enhancing the
gait propulsion. On the other hand, the unaffected side showed
a reduction in ankle RoM. The addition of an external device
might cause the stroke patients walked in a more cautious and
stiffer gait pattern. Future study would focus on long-term
adaptation to the powered assistance from Ankle Robot as a
gait training program.
V. CONCLUSION
In this paper, an exoskeleton powered AFO was developed
as a gait training device for long-distance walking of stroke
patient. We designed a novel control algorithm using the leg
tilting angle and leg angular velocity for real-time
classification of walking conditions: level walking, stair
ascending/descend. The control algorithm was evaluated on
three chronic stroke patients. The powered assistance could
reduce the dropped foot gait of the stroke patients. However,
more experiments would be needed to prove its therapeutic
effects.
ACKNOWLEDGMENT
The study was financially supported by an ITF grant
(GHP/001/12) from the Innovation and Technology
Commission of the Hong Kong Special Administrative
Region.
REFERENCES
[1] V. L. Feigin, et al., "Global and regional burden of stroke during
1990-2010: findings from the Global Burden of Disease Study 2010",
Lancet, 383, no. 9913, pp. 245-54, 2014.
[2] C. Beyaert, R. Vasa and G. E. Frykberg, "Gait post-stroke:
Pathophysiology and rehabilitation strategies", Neurophysiol Clin, 45,
no. 4-5, pp. 335-55, 2015.
[3] S. Tasseel-Ponche, A. P. Yelnik and I. V. Bonan, "Motor strategies of
postural control after hemispheric stroke", Neurophysiol Clin, 45, no.
4-5, pp. 327-33, 2015.
[4] A. Singam, C. Ytterberg, K. Tham and L. von Koch, "Participation in
Complex and Social Everyday Activities Six Years after Stroke:
Predictors for Return to Pre-Stroke Level", PLoS One, 10, no. 12, pp.
e0144344, 2015.
[5] G. Chen, C. Patten, D. H. Kothari and F. E. Zajac, "Gait differences
between individuals with post-stroke hemiparesis and non-disabled
controls at matched speeds", Gait Posture, 22, no. 1, pp. 51-6, 2005.
[6] M. Alam, I. A. Choudhury and A. Bin Mamat, "Mechanism and design
analysis of articulated ankle foot orthoses for drop-foot",
ScientificWorldJournal, 2014, no., pp. 867869, 2014.
[7] H. Naito, Y. Akazawa, K. Tagaya, T. Matsumoto and M. Tanaka, "An
ankle-foot orthosis with a variable-resistance ankle joint using a
magnetorheological-fluid rotary damper", Journal of Biomechanical
Science and Engineering, 4, no. 2, pp. 182-191, 2009.
[8] W. Svensson and U. Holmberg, "Ankle-foot-orthosis control in
inclinations and stairs", 2008 IEEE Conference on Robotics,
Automation and Mechatronics, no., pp. 301-306, 2008.
[9] S. Galle, W. Derave, F. Bossuyt, P. Calders, P. Malcolm and D. De
Clercq, "Exoskeleton plantarflexion assistance for elderly", Gait
Posture, 52, no., pp. 183-188, 2016.
[10] G. S. Sawicki, K. E. Gordon and D. P. Ferris, "Powered lower limb
orthoses: applications in motor adaptation and rehabilitation",
Rehabilitation Robotics, 2005. ICORR 2005. 9th International
Conference on, no., pp. 206-211, 2005.
[11] S. Duerinck, E. Swinnen, P. Beyl, F. Hagman, I. Jonkers, P. Vaes and P.
Van Roy, "The added value of an actuated ankle-foot orthosis to restore
normal gait function in patients with spinal cord injury: a systematic
review", J Rehabil Med, 44, no. 4, pp. 299-309, 2012.
215
[12] 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, 12, no. 1, pp. 24-31,
2004.
[13] S. Hwang, J. Kim, J. Yi, K. Tae, K. Ryu and Y. Kim, "Development of
an active ankle foot orthosis for the prevention of foot drop and toe
drag", 2006 International Conference on Biomedical and
Pharmaceutical Engineering, no., pp. 418-423, 2006.
[14] J. Ward, T. Sugar, J. Standeven and J. R. Engsberg, "Stroke survivor
gait adaptation and performance after training on a powered ankle foot
orthosis", Robotics and Automation (ICRA), 2010 IEEE International
Conference on, no., pp. 211-216, 2010.
[15] K. A. Shorter, J. Xia, E. T. Hsiao-Wecksler, W. K. Durfee and G. F.
Kogler, "Technologies for powered ankle-foot orthotic systems:
Possibilities and challenges", Mechatronics, IEEE/ASME Transactions
on, 18, no. 1, pp. 337-347, 2013.
[16] A. M. Dollar and H. Herr, "Lower extremity exoskeletons and active
orthoses: challenges and state-of-the-art", Robotics, IEEE Transactions
on, 24, no. 1, pp. 144-158, 2008.
[17] M. Zhang, T. C. Davies and S. Xie, "Effectiveness of robot-assisted
therapy on ankle rehabilitation--a systematic review", J Neuroeng
Rehabil, 10, no., pp. 30, 2013.
[18] K. A. Shorter, G. F. Kogler, E. Loth, W. K. Durfee and E. T.
Hsiao-Wecksler, "A portable powered ankle-foot orthosis for
rehabilitation", J Rehabil Res Dev, 48, no. 4, pp. 459-72, 2011.
[19] Y. David Li and E. T. Hsiao-Wecksler, "Gait mode recognition and
control for a portable-powered ankle-foot orthosis", IEEE Int Conf
Rehabil Robot, 2013, no., pp. 6650373, 2013.
[20] M. Landers, "Treatment-induced neuroplasticity following focal injury
to the motor cortex", Int J Rehabil Res, 27, no. 1, pp. 1-5, 2004.
[21] J. C. Moreno, F. Barroso, D. Farina, L. Gizzi, C. Santos, M. Molinari
and J. L. Pons, "Effects of robotic guidance on the coordination of
locomotion", Journal of neuroengineering and rehabilitation, 10, no. 1,
pp. 1, 2013.
[22] Z. Zhou, Y. Sun, N. Wang, F. Gao, K. Wei and Q. Wang,
"Robot-Assisted Rehabilitation of Ankle Plantar Flexors Spasticity: A
3-Month Study with Proprioceptive Neuromuscular Facilitation",
Front Neurorobot, 10, no., pp. 16, 2016.
[23] R. C. Browning, J. R. Modica, R. Kram and A. Goswami, "The effects
of adding mass to the legs on the energetics and biomechanics of
walking", Med Sci Sports Exerc, 39, no. 3, pp. 515-25, 2007.
[24] E. K. Antonsson and R. W. Mann, "The frequency content of gait",
Journal of biomechanics, 18, no. 1, pp. 39-47, 1985.
[25] H. Y. Lau and K. Y. Tong, "The reliability of using accelerometer and
gyroscope for gait event identification on persons with dropped foot",
Gait Posture, 27, no. 2, pp. 248-57, 2008.
[26] H. Y. Lau, K. Y. Tong and H. Zhu, "Support vector machine for
classification of walking conditions using miniature kinematic
sensors", Medical & biological engineering & computing, 46, no. 6, pp.
563-573, 2008.
[27] H. A. Varol, F. Sup and M. Goldfarb, "Multiclass real-time intent
recognition of a powered lower limb prosthesis", IEEE Trans Biomed
Eng, 57, no. 3, pp. 542-51, 2010.
[28] R. Riener, M. Rabuffetti and C. Frigo, "Stair ascent and descent at
different inclinations", Gait & posture, 15, no. 1, pp. 32-44, 2002.
[29] P. M. Mills and R. S. Barrett, "Swing phase mechanics of healthy
young and elderly men", Human movement science, 20, no. 4, pp.
427-446, 2001.
[30] S. Nadeau, B. J. McFadyen and F. Malouin, "Frontal and sagittal plane
analyses of the stair climbing task in healthy adults aged over 40 years:
what are the challenges compared to level walking?", Clin Biomech
(Bristol, Avon), 18, no. 10, pp. 950-9, 2003.
[31] A. Protopapadaki, W. I. Drechsler, M. C. Cramp, F. J. Coutts and O. M.
Scott, "Hip, knee, ankle kinematics and kinetics during stair ascent and
descent in healthy young individuals", Clin Biomech (Bristol, Avon),
22, no. 2, pp. 203-10, 2007.