Content uploaded by Juhui Qiu
Author content
All content in this area was uploaded by Juhui Qiu on Jan 11, 2021
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=ierd20
Expert Review of Medical Devices
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ierd20
The realization of robotic neurorehabilitation in
clinical: use of computational intelligence and
future prospects analysis
Jiali Yang , Zhiqi Zhao , Chenzhen Du , Wei Wang , Qin Peng , Juhui Qiu &
Guixue Wang
To cite this article: Jiali Yang , Zhiqi Zhao , Chenzhen Du , Wei Wang , Qin Peng , Juhui
Qiu & Guixue Wang (2020): The realization of robotic neurorehabilitation in clinical: use of
computational intelligence and future prospects analysis, Expert Review of Medical Devices, DOI:
10.1080/17434440.2020.1852930
To link to this article: https://doi.org/10.1080/17434440.2020.1852930
Accepted author version posted online: 30
Nov 2020.
Submit your article to this journal
Article views: 9
View related articles
View Crossmark data
Publisher: Taylor & Francis & Informa UK Limited, trading as Taylor & Francis Group
Journal: Expert Review of Medical Devices
DOI: 10.1080/17434440.2020.1852930
The realization of robotic neurorehabilitation in clinical:
use of computational intelligence and future prospects
analysis
Jiali Yang1, Zhiqi Zhao1, Chenzhen Du1, Wei Wang2, Qin Peng3, Juhui Qiu1*, Guixue Wang1*
1 Key Laboratory for Biorheological Science and Technology of Ministry of Education, State Key Laboratory
of Mechanical Transmission, State and Local Joint Engineering Laboratory for Vascular Implants,
Bioengineering College of Chongqing University, Chongqing 400030, China
2 Chongqing Key Laboratory of Translational Research for Cancer Metastasis and Individualized Treatment,
Chongqing University Cancer Hospital, Chongqing, China
3 Institute of Systems and Physical Biology, Shenzhen Bay Laboratory, Shenzhen, 518132, China
* Correspondence: jhqiu@cqu.edu.cn (J.Q.), wanggx@cqu.edu.cn (W.G.); Tel.:+86-23-65112675
Author Contributions
Zhao, Peng and Yang had full access to all the data in the study and take responsibility for the
integrity of the data and the accuracy of the data analysis. Concept and design: Yang, Qiu, Du. Drafting
of the manuscript: All authors. Critical revision of the manuscript for important intellectual content:
Qiu, Wang, Yang.Funding
This work was funded by the National Natural Science Foundation of China (31971242), Key
grants from Chongqing Science and Technology Bureau to G.W (cstc2019jcyj-zdxm0033) and the
National Key Technology R & D Program of China (SQ2018YFC010099) as well as Fundamental
Research Funds for the Central Universities (2020CDCGJ011) and Chongqing Municipal Education
Commission, China (KYYJ202001).
Conflicts of Interest
The authors declare no conflict of interest.
ACCEPTED MANUSCRIPT
Review
State of the art and advances of robotic
neurorehabilitation with computational intelligence
Jiali Yang1, Zhiqi Zhao1, Chenzhen Du1, Wei Wang2, Qin Peng3, Juhui Qiu1,*, Guixue Wang1,*
1 Key Laboratory for Biorheological Science and Technology of Ministry of Education, State Key Laboratory
of Mechanical Transmission, State and Local Joint Engineering Laboratory for Vascular Implants,
Bioengineering College of Chongqing University, Chongqing 400030, China
2 Chongqing Key Laboratory of Translational Research for Cancer Metastasis and Individualized Treatment,
Chongqing University Cancer Hospital, Chongqing, China
3 Institute of Systems and Physical Biology, Shenzhen Bay Laboratory, Shenzhen, 518132, China
* Correspondence: jhqiu@cqu.edu.cn (J.Q.), wanggx@cqu.edu.cn (W.G.); Tel.:+86-23-65112675
ACCEPTED MANUSCRIPT
Abstract
Introduction: Although there is a need for rehabilitation treatment with the increase in the aging
population, the shortage of skilled physicians frustrates this necessity. Robotic technology has been
advocated as one of the most viable methods with the potential to replace humans in providing physical
rehabilitation of patients with neurological impairment. However, because the pioneering robot devices
suffer several reservations such as safety and comfort concerns in clinical practice, there is an urgent
need to provide upgraded replacements. The rapid development of intelligent computing has attracted
the attention of researchers concerning the utilization of computational intelligence algorithms for
robots in rehabilitation.
Areas Covered: This article reviews the state of the art and advances of robotic neurorehabilitation
with computational intelligence. We classified advances into two categories: mechanical structures and
control methods. Prospective outlooks of rehabilitation robots also have been discussed.
Expert opinion: The aggravation of global aging has promoted the application of robotic technology in
neurorehabilitation. However, this approach is not mature enough to guarantee the safety of patients.
Our critical review summarizes multiple computation algorithms which have been proved to be
valuable for better robotic use in clinical settings and guide the possible future advances in this
industry.
Keywords: neurorehabilitation; robot-assisted; computational intelligence; mechanical structures;
control methods;
Article Highlights
Robot-aided interventions have been playing an increasing role in the treatment of stroke patients
over recent years subject to the increasing aging population.
Original robotic equipment has some potential risks that are likely to cause second injuries. Advance
technologies of computational intelligence algorithms are applied for optimizing the application of
robots in rehabilitation.
Robotic neurorehabilitation rapidly progresses in certifying clinical operation with the help of
computational intelligence algorithms, not only in mechanical structures, but also in control methods.
For a brighter future of rehabilitation robots, plenty of constructive suggestions are considered for
improving computation intelligence algorithms and a variety of advanced technologies should be
greatly considered.
ACCEPTED MANUSCRIPT
1. Introduction
Stroke is one of the most prevalent disabling health-care diseases among the elderly. Globally, it is
the second-leading cause of disability and mortality in this group of individuals [1,2]. Statistics show
that the risk of stroke increases with age, with close to 75% of all strokes occurring among people older
than 65 years [3-5]. There is a worldwide upward trend of the aged population, with individuals over 60
years estimated to rise to 2 billion by 2050. Accordingly, the increasing number of elderly people poses
immense challenges to the existing healthcare systems, particularly rehabilitation medicine [6]. The
main cause of stroke-associated disability is varying degrees of damages to different parts of the central
nervous system (CNS) [7]. According to the neuronal plasticity theory, after external damages, the CNS
can be reorganized to restore its desirable functioning [8-10]. Patients with motor impairment require
regular training for reshaping the injured central nerve and enable the CNS to restore control of the
motor muscles [11,12]. Professional and standardized rehabilitation exercises are indispensable in the
rehabilitation nursing of such patients. However, there is a chronic, global shortage of qualified
rehabilitation physio-therapists [13]. The current manual or automated therapy no longer meets the
dynamic needs of patients, thus researchers are exploring the power of robotic technologies for
reducing the work pressure of physical therapists [14-16].
Robot-mediated therapy has been proved to have a huge potential in neurorehabilitation [17,18].
Compared with traditional physical therapy, robot-mediated therapy shows its strength in both the
duration and precision of rehabilitation training. Unlike physical therapists who often feel tired and
unable to maintain the same operating force during the long-term training process, rehabilitation robots
could provide accurate and unified rehabilitation training during the long-term training [19]. In addition,
from the perspective of rehabilitation assessment, robotic neurorehabilitation collects relevant data of
patients. Using dynamics and kinematics methods, corresponding biomechanical models are established,
so that neurologists and therapists could better understand the patient’s recovery status and evaluate the
patient’s recovery [20]. Because of these advantages, robot-mediated therapy is thought to be the
solution to increasing rehabilitation needs. However, most traditional robotic devices are functionally
inflexible and do not guarantee safety and reliability. In fact, some procedures are uncomfortable for
patients and some even cause secondary injuries [21,22]. Therefore the current dilemma revolves
around the provision of evidence-based, safe and effective rehabilitation robotic technologies with the
practical clinical application [23]. Multiple useful computational intelligence algorithms have been
suggested to address this concern [24]. Indeed, computational intelligence algorithms are one of the
most advanced methods that are frequently applied in the design, control and optimization of robots.
Their designs such as genetic algorithms (GA) often simulates the natural biological behavior and
physical structure of an individual, inspired by the natural selection and natural genetic mechanism of
biological evolution [25]. In the searching process, the used intelligent computing algorithm constantly
updates its corresponding parameters by finding a feasible solution within acceptable time and space.
This has provided an invaluable improvement of robotic neurorehabilitation [26,27].
In this study, we highlight recent advances in robotic rehabilitation embracing modern inventions
such as computational intelligence algorithms. The articles were classified into two categories;
optimization of design and robotic control, with the latter encompassing trajectory control and
recognition of the intended motion. We then analyzed recent advances in computational intelligence
algorithms regarding rehabilitation robots. The rest of this paper is organized as follows: Part one
summarizes the mechanical structure and operational rationale of rehabilitation robots, which utilizes
computational intelligence algorithms. Part two discusses trajectory control methods and models for
ACCEPTED MANUSCRIPT
reco
g
reha
b
algo
r
Clin
i
five
fro
m
AN
D
Eng
l
crite
2. A
p
mo
v
[29]
.
chal
l
gnizing inte
n
b
ilitation pro
c
r
ithms. We s
e
icalKey data
b
years where
m
databases,
b
D
“computat
i
l
ish-language
e
ria.
p
plication of
c
A rehabilita
t
v
ements [28].
.
Though a lo
t
l
enges yet to
n
ded motions
c
ess and pros
p
e
arched relev
a
b
ase. Advanc
e
only articles
b
ut all includ
e
i
onal intelli
g
articles are c
o
Figure
c
omputationa
l
t
ion robot is
The structur
a
t
of rehabilitat
be addressed
,
using simila
r
p
ect. Part fou
a
nt articles in
e
d literature
w
from Januar
y
e
d “robotic”
O
g
ence” AND
o
unted in this
1: Flow diagra
m
l
intelligence
a medical
d
a
l design of s
u
i
on devices h
a
,
such as hig
h
r
intelligence
ur
p
rovides an
n
Medline, Pu
b
w
as ensure b
y
y
2015 were
O
R “robots”
O
“optimizati
o
study. In the
m
of selection
p
algorithms in
d
evice that is
u
ch equipme
n
a
ve been appl
i
h
cost and co
m
algorithms.
P
expert opini
o
b
Med, PubM
e
y
searching st
u
conside
r
ed.
S
O
R“medical e
q
o
n” as key
w
end, 30 relat
e
p
rocess for fin
a
structural de
s
highly accu
n
t greatly infl
u
i
ed in clinical
m
plex structu
r
P
art three di
s
o
n of comput
a
e
d Central,
W
u
dies publish
e
S
earch strate
g
q
uipment” A
N
w
ords or se
a
e
d articles ma
t
a
l review
s
ign
r
ate and ada
p
u
ence their p
r
settings, the
r
r
e [30]. The c
o
s
cusses the c
u
a
tional intelli
g
W
eb of Scienc
ed from the
r
g
ies varied sl
i
N
D “rehabilit
a
a
rch theme.
t
ched our sel
e
p
tive to rep
e
r
actical appli
c
re are still s
e
omplexity of
u
rrent
g
ence
e
and
r
ecent
i
ghtly
a
tion”
Only
e
ction
e
titive
c
ation
e
veral
these
ACCEPTED MANUSCRIPT
devices, which requires direct supervision by clinicians during use, coupled with the high cost of the
devices, greatly limits their impact on productivity, thus limiting their wide range of applications in the
clinical environment [31]. Optimization of robotic structure design reduces the complexity of the
designed equipment to meet the performance requirements, as well as the cost of the equipment to obtain
more comprehensive equipment, effectively improving its clinical use value [32]. Therefore, developing
a robotic design remains one of the research concerns that has existed for decades.
2.1. Common parallel robots
Structurally, parallel robots are attractive in rehabilitation because of their optimal stiffness, speed
and flexibility, suitable for high torque and precision [33,34]. In 2015, Wang et al developed a
redundant actuate ankle, parallel rehabilitation robot. Structurally, it features a new modified
multi-objective differential evolution algorithm (NMODE) designed to address parameter optimization
challenges and adopts other computational intelligence algorithms such as POEMA and MODEA
[35,36]. Consequently, its performance is satisfactory [37]. Several studies have also reported the use
of other new components. Zi et al conceptualized a novel waist rehabilitation robot with pneumatic
artificial muscles (PAMs) in the device drive. PAMs are widely used in the rehabilitation industry
because they are cheap, offer excellent force to weight ratio and variable installation possibilities
[38,39]. The robot is divided into two parts; one section twists the waist whereas the other section
completes the lower limb traction. Satisfactory cable length is obtained successfully by applying the
particle swarm optimization algorithm (PSO), a common computational intelligence algorithm with
good robustness and adaptability [40]. Well designed robots with computational intelligence algorithms
can perform successful rehabilitation, thus offer the early prospect for their clinical application.
Multiple experiments have been performed to establish the best practical fruition of different
computational intelligence algorithms. Jamwal et al reviewed the performance of a wearable parallel
robot for ankle joint treatment, which uses a non-dominated sorting genetic algorithm (NSGA II) to
search for Pareto optimal solutions. It achieved six key performance objectives and suitable design was
acquired [41]. Building on their work, they further explored a biased fuzzy sorting genetic algorithm
(BFSGA) for a parallel robot which encourages satisfying results in the extreme zones [42,43]. On his
part, Lamine et al quantitatively explored the effectiveness of gait-training with a related machine. It
controls posture during the gait training by moving the orthosis. The quantitative analysis of the normal
gait is the input of the inverse dynamics model. The optimization algorithm applied decreased the
maximum tension in cables by 13%, achieving free collision of the cable / end actuator [44].
Additionally, Gabardi et al developed a thumb-exoskeleton that utilizes just one linear actuator for five
degrees of freedom. GA was used to find suitable link lengths, having been successfully validated by
two different experiments to be effective in such optimization [45]. Enferadi et al creatively invented a
new spherical parallel robot consisting of a fixed platform that could completely rotate about an axis. It
was applied in a relatively wide clinical settings because of the corresponding intelligence algorithms
that greatly decreases computational time in robot [46]. In 2019, Laribi et al on their part employed
computational intelligence algorithms for optimizing cable tensions for cable-driven parallel robots,
which further promoted the application of such designs [47]. Taken together, although computational
intelligence algorithms are adopted for different designs, every algorithm has a unique application
conditions for optimal performance. However, the diversity of algorithm types offers several selection
options for further research.
2.2. Linkage mechanisms design
Among the parallel robots, a lack of linkage mechanisms presents a significant setback [48,49].
ACCEPTED MANUSCRIPT
The equipment is positioned in two-dimensional planar space, imposing high rigidity and light weight
[50,51]. In 2015, Gao et al presented a symmetrical five-bar linkage composed of two revolute and
three prismatic joints, thus taking in to account the stiffness constraint. They addressed the stiffness
challenge by kinematic analysis and employed it as an objective function in GA optimization. Finally,
suitable mechanisms are incorporated in the robot-mediated training, which not only ensures high
rigidity but also makes the workspace that covers the human upper limb range of motion [52]. For the
dimension synthesis challenge, Shao et al chose a traditional computational algorithm: GA. Structurally,
it exemplified the cam-linkage robot design which consists of a two degree-of-freedom seven-bar crank
slider. Interestingly, this design achieved greater performance compared to the original robot [53].
Elsewhere, McDaid et al developed a novel robot based on a two degree of freedom, five-bar linkage,
named the Pro-Gait. It permits the operator to move relative to the device and handles joint dislocations
while still allowing the accurate realization of various gait modes. They also applied related
computational intelligence algorithm in improving the performance of the joint workspace,
singularities distance and geometric size of the robot which achieves the required motion and transmit
force to the user [54].
Besides these studies, there are other numerous research findings on linkage optimization. For
instance, Singh et al came up with a four-bar linkage robot for gait disorder, utilizing the global
coordinate system, a non-conventional mechanism, that moves with the hip trajectory. In this first trial,
traditional PSO was preferred for minimal tracking error and less complex assembly procedures [55].
Subsequently, they developed a hybrid teaching learning particle swarm optimization (HTLPSO)
algorithm, considered as an excellent method that offers a further improvement to designed linkage
[56,57]. The optimization process was fused with the swing and stance phase in an attempt to reduce
the error between the target and actual trajectory. Compared to the well-established GA, HTLPSO
decreases 4.17% and 24.14% in the function evaluations of swing and stance phase when converges to
an optimum value. Consequently, the structure of parallel robots underlines the usefulness of
mechanical design to its application in rehabilitation medicine.
2.3. Related component
Some researchers have focused on boosting the performance of some components of the robots
such as the compliant planar spring, crucial for gravity-balance. It is used in adjusting the rigidity of the
device and to provide optimal equilibrium [58,59]. The controversial static performances of the spring
system significantly influences the overall design of the device. Chau et al adopted a novel
multi-objective genetic algorithm (MOGA) for estimating errors of compliant planar spring. Using this
algorithm, many functionality tests have been conducted effectively and guaranteed the performance of
this static spring planar [60]. The improvement of such components has also boosted the overall
performance of the robots.
An overview of the rehabilitation equipment and computational intelligence algorithms are shown
in Table 1. The table provides an insight into the rational overall structural design of rehabilitation
medical equipment concerning performance. Parameters for the choice of a rehabilitation robot present
a significant concern for the structural design of the equipment. The application of computational
intelligence algorithms efficiently optimizes the mechanical design of the robots, further improving the
stability of the equipment as well as ensures physical safety of patients in clinical application.
ACCEPTED MANUSCRIPT
Table 1: Overview of the structural design of rehabilitation robotic device with computational intelligence
algorithms
Ref The
algorithms Strength Optimized mechanisms Application
[37] NMODE Faster convergence speed and more
non-dominated solutions
Improved output transfer,
torque and constraint
processing
Ankle
[40] PSO Robustness and adaptable Suitable cable length
design Waist
[41] NSGA II
Faster convergence and lower
computational complexity
Suitable design, offers
Parto solutions Ankle
[42] BFSGA
Performs well in exploring the extreme
zones of the Pareto front
Provides multiple Pareto
optimal solutions for robot
design
Ankle
[44] Not mentioned Not mentioned Low maximum tension
(decrease 13%) Gait
[45] GA Versatile and robustness Have good adaptability
with optimized link lengths
Thumb
[46] GA Versatile and robustness Have a larger and
singularity-free workspace Upper Limb
[47] GA Versatile and robustness
Smallest robot size and
minimum cable tension
distributions
Upper Limb
[52] GA Versatile and robustness
The device achieves
highest rigidity and
singularity-free workspace
Upper-limb
[53] GA Versatile and robustness Small and more compact Gait
[54] GA Versatile and robustness
Achieve desired
workspace, singularities
and physical size
Gait
[55] PSO Good robustness and adaptability Reduce the tracking error
during operation Lower limb
[56]
HTLPSO
Better convergence and solution
Allows for smooth
movement and accurately
track all designated points
Lower limb
[60] MOGA Simple to operate and highly efficient
Eliminate the modeling
errors and improve the
overall static performances
and robustness
Upper limb
NMODE = New modified multi-objective differential evolution Algorithm, PSO = particle swarm optimization
algorithm, NSGA II = Non-dominated Sorting Genetic Algorithm, BFSGA = biased fuzzy sorting genetic algorithm,
GA = Genetic algorithm, HTLPSO = hybrid teaching learning particle swarm optimization algorithm, MOGA =
Multi-objective genetic algorithm.
ACCEPTED MANUSCRIPT
3. Application in the control field
Because robotic neurorehabilitation medical devices operate in close contact with patients, the
safety of patients must be guaranteed [61]. The stress applied and movement of the limb or joint must
be regulated accurately to avoid secondary injury [21,22]. Meanwhile, the realization of accurate and
unified rehabilitation training is inseparable from the optimization and improvement of the control
technology [62]. Consequently, robotic controls should be highly accurate and efficient to ensure the
wide application of robotic neurorehabilitation in clinical settings [63-65]. However, because the human
muscles are nonlinear, it is difficult to precisely set the parameters of controllers, which are influenced
by external disturbances [66]. Multiple computational intelligence algorithms are employed to improve
the control accuracies through appropriate training of the controller. This provides optimal models that
can better recognize motion input signals.
3.1. Training on trajectory control of rehabilitation robots
Recovery training requires steady, safe and gradual movements, because any sudden movement
may cause unbearable damage to the patients [67]. Consequently, the error of the controller trajectory
should be at a bare minimum. Convention controllers always produce large overshoots in control
trajectory tracking, thus do not meet the threshold for acceptable performance. As shown in the
following sections, some studies have been done in this field focusing on the combination of
computational intelligence algorithms and controllers.
3.1.1. Optimization of convention controllers
Computational intelligence algorithms have solved disturbance related rejections in conventional
controllers. PSO algorithm developed by Taha et al to optimize the performance controllers in gait
rehabilitation is such an example. The proposed proportional–derivative (PD) controller with modified
parameters shows strong robustness in response to a disturbance around motion, which reflects its
potential application in the rehabilitation of trajectory control [68]. Joyo et al experimented with the
fusion of both PSO and artificial bee colony algorithm (ABC) in finding the optimal parameters of the
proportional–integral–derivative (PID) controller. Compared with PSO-PID with regard to performance
of maximum overshoot, rise and settling time and use of maximum sensitivity function under
disturbance, ABC-PID has a superior practical application, thus offers promising solutions for research
and development of rehabilitation equipment [69].
In 2017, Belkadi et al utilized a robust adaptive control method with a PID controller based on a
modified particle swarm optimization (MPSO) algorithm. The adaptive controller reduced the error in
the trajectory without establishing the model of nonlinear systems in advance and drew ideal
conclusions both in position and velocity [70]. Furthermore, Ayas et al explored the applicability of
fractional order proportional–integral–derivative (FOPID) controller, an extension of the conventional
PID controller that has become one of the most practical fractional order control way in linear systems
[71,72]. It employs robotics control, drive systems, power electronics among others. Based on our
analysis, both the cuckoo search algorithm (CSA) and PSO algorithms are recommended for providing
satisfactory controller parameters in terms of predefined performance index [73-75]. An optimized
controller could eliminate the influence of external interference on its operation, thus achieving high
precision in controlling the desired trajectory [76]. Both of these tests utilize computational intelligence
algorithms for optimization, and the corresponding excellent performance underlines the great potential
of their use in robot control.
3.1.2. Optimization of the intelligent controller
ACCEPTED MANUSCRIPT
Apart from the traditional control methods, computational intelligence algorithms play a crucial
role in identifying some parameters of advanced and intelligent controllers. For instance, Li et al
designed sliding controllers based on the Radial Basis Function (RBF) of the neural network. This
model provides matching condition for system dynamics and uncertain disturbances. To determine
suitable related parameters for RBF neural network, GA was introduced, achieving accurate control for
knee joint movement [77].
In addition, Azar et al. reviewed the impedance control method with regard to minimal error and
good stability, where GA was adopted to enhance the safety of patients. Impedance control is an
intelligent method which always guides the position and force by adjusting the mechanical impedance
to the external forces generated by environmental disturbance. Its application greatly impacts the
security of the trajectory [78]. Based on the previous work by FOPID controller, Ayas et al developed
and incorporated a fuzzy logic controller (FLC) with a cuckoo search algorithm (CSA). Compared with
traditional PID controllers, this hybrid model reduced steady-state tracking errors by 50% [79].
Meanwhile, Silawatchananai et al used PSO-based fixed structure H∞ control which produced better
tracking performance than the conventional PID controller [80]. Compared with traditional robot
controllers, these advanced controllers display better robustness and stability in the recovery of crucial
factors, development of rehabilitation controls and trajectory tracking.
The details of the methods for trajectory control referred above are concluded in Table 2. As these
details show, more and more intelligent control schemes are applied in the control of rehabilitation
robots. With the development of control technology, more rehabilitation equipment for patients are
poised to be developed endlessly.
Table 2: Overview of how rehabilitation robots control trajectory tracking with computational intelligence
algorithms.
Refer The
algorithms Strength of algorithms Optimized controller Application
[68] PSO Fast convergence and good
stability
More robust under
disturbances and uncertainties Gait
[69] ABC
Good global search ability and
more robust
Less overshoot and no
steady-state error Upper limb
[70] MPSO
Quick adaptation to changes and
good stability
Obtain Satisfactory Track
control effect both in position
and velocity
Lower limb
[76] CSA
Have Strong ability to find the
desired solution
Few steady-state tracking
errors under external
interference
Ankle
[77] GA Highly versatile and robust
Offers accurate adjustment to
the joint angle more accurately
and effectively under external
interference
Knee
[78] GA Highly versatile and robust More secure and synchronized
correction lower limbs
[79] CSA
Simple to operate because of few
parameters and has strong global
search ability
Effective in trajectory tracking
subject to effects of external
disturbance
Ankle
ACCEPTED MANUSCRIPT
[80] PSO Fast convergence and good
stability
Achieves better tracking
performance Arm
PSO = particle swarm optimization algorithm, ABC = artificial bee colony algorithm, MPSO = Modified particle
swarm optimization algorithm, CSA = Cuckoo search algorithm, GA = Genetic algorithm.
3.2. Motion recognition of rehabilitation robots
Motion recognition is a technology that utilizes the physical structure and various sensory systems
to recognize patterns and intentions of training patients [81,82].It is achieved by decoding the
bioelectricity signal excited when a neuron carrying information on human behavior transmits the
signal between tissues / organs [83]. Some scientists believe that this technology presents the best
invention key to achieving efficient human-machine coordination in the most comfortable way possible
[84-86]. The bigger aim is to combine motion recognition and rehabilitation by building on the success
of related work of this technology.
3.2.1. Judgment of motion angle
The judgment of the angle of motion is one of the most crucial parts of motion recognition. Pasha
Zanoosi et al focused on improving the stability of predicting joint trajectories. Using experimental
data on human joint trajectories under tilting disturbance, it was shown that joint trajectories and
muscle activation could be successfully predicted. In subsequent experiments, they proposed a
multi-objective optimization model based on non-dominated sorting genetic algorithm (NSGA) which
emphasized stability and muscle stresses as key functions. Together, the two findings revealed the
stability of human work based on predictable joint trajectories [87]. In other words, NSGA is useful in
improving the technology of judgment for the angel of motion. Similar research performed by Feng et
al. adopting an improved cuckoo search algorithm (CSA), which is a combination of traditional CSA
and steepest descent method, aimed at offering better prediction of wrist joint angles. The wrist joint
angle model developed based on support vector regression (SVR) and steepest descent cuckoo search
algorithm (SDCS) has been shown to have high accuracy and is timely [88]. It scored well on its ability
to predict angles, thus offer valuable reference for surface electromyography (sEMG) based control for
rehabilitation mechatronic systems. With the help of computational intelligence algorithms, the
prediction of angles of motion is certainly becoming more stable and accurate concerning updating the
recovery efficiency of rehabilitation training.
3.2.2. Prediction of motion torque
The judgment of the torque during the training process is desirable to ensure the patient’s safety. In
2018, Buongiorno et al applied GA optimized EMG-driven neuro-musculo-skeletal models (NMS) to
predict the human body torque. These models were capable of predicting the shoulder and elbow
moments with a low error and computational cost, thus prospective in rehabilitation cases [89].
Similarly, Peng et al put forward a neuromusculoskeletal model based on the hill model. They used GA
to find various model parameters from data on sEMG and joint torques obtained with the assistance of
an upper-limb rehabilitation robot. Their findings demonstrates the usefulness of this new method in
modeling robot rehabilitation, or neuromuscular assessment [90]. Simultaneously, Noughaby et al also
utilized the hill model and GA. However, they mainly focused on reducing the interaction of forces
between the operator and the equipment. This effectively achieved better torque estimation than the hill
model. GA used to identify suitable parameters for the hill model registered satisfactory results [91].
Together, these studies significantly make adequate progress in predicting human torque, a parameter
that promotes the establishment of safe clinical systems for medical rehabilitation devices.
3.2.3. Recognition of motion pattern
ACCEPTED MANUSCRIPT
Significant progress has been made in the study of motion pattern recognition. Zhang et al
identified patient motion pattern using a signal acquisition system of sEMG. To achieve a better
recognition effect, Levenberg–Marquardt (LM) and PSO algorithms are considered. Eight different
composite actions are achieved with an average recognition rate of 93.1% and an average response
time of less than 0.25 s [92]. In the same year, Cai et al combined sEMG signals with a support vector
machine (SVM) model made from GA for upper limb identification, which showed good accuracy [93].
A model motion of the healthy side may help in the treatment of the affected side more effectively
hence can be a suitable strategy for recovery training of stroke patients.
The aforementioned studies on motion recognition are summarized in Table 3. The results shown
in the Table indicate that motion recognition plays a crucial part in robot myoelectric control, a
potential control method for human-machine interface (HMI) control [94,95]. Its main target is to
ensure the intention of the patients and achieve patient-cooperative training. Future, rehabilitation
robots should be intelligent and collaborative which is inseparable from the application of intention
recognition.
Table3. Overview of motion recognition for rehabilitation robots with computational intelligence algorithms.
Refs Algorithms
name
Strength of
algorithms Application effect Application
[87] NSGA
Could avoid
premature
convergence
Predict joint motion trajectories
more accurately exposed under
assumed perturbation
Lower limbs
[88] SDCS
Good convergence
and robustness
Reduces the training time and
improves the accuracy of
predicting the angle of wrist
joint
Lower limbs
[89] GA
Strong versatility
and good robustness
predict the moment of shoulder
and elbow with a low error rate Shoulder and elbow
[90] GA
Strong versatility
and good robustness
Search the optimal parameters of
a neuromusculoskeletal model Upper limbs
[91] GA
Strong versatility
and good robustness
Reduced force interaction
between people and machines by
about 25%
Knee
[92] PSO
Fast convergence
and good stability
Improve the real-time
performance and the recognition
rate
Upper limb
[93] GA
Strong versatility
and good robustness
Has higher classification
accuracy
Upper Limb
NSGA = non-dominated sorting genetic algorithm, SDCS = steepest descent cuckoo search algorithm, GA =
genetic algorithm, PSO = particle swarm optimization algorithm.
4. Prospective outlooks
Although the usage of computational intelligence algorithms improves the performance of robot
irrespective of the design and control, several challenges remain to be solved. Thus, research should be
conducted to improve the application of computational intelligence algorithms in rehabilitation. And a
variety of high and advanced technologies are considered to integrate into rehabilitation filed.
ACCEPTED MANUSCRIPT
shor
t
intel
appl
i
com
p
exp
e
to i
m
Fig
u
PS
O
algo
r
mak
i
that
reha
b
dev
e
on
t
anal
y
acq
u
The data re
v
t
comings. Fir
l
ligence algor
i
ication. Sec
o
p
utational in
t
e
rience, whic
h
m
prove the o
p
u
re 2 indicate
O
, accounting
r
ithms have
ing it difficu
l
are intelligen
t
Figure2.
A
p
p
mechanical st
r
In addition,
b
ilitation ro
b
e
lopment of
b
t
ime. Big da
t
y
zing and p
r
u
ired from s
e
v
iewed here
stly, there is
n
i
thms. There
fo
o
ndly, there
t
elligence alg
o
h
limi
t
s the p
e
p
eration effec
that the mos
t
for over 50
low accurac
y
l
t to find opt
i
t
should be es
t
lication of dif
f
r
uctural design
;
science an
d
b
ots, such as
b
ig data techn
o
t
a technolog
y
r
ocessing a l
a
e
nsors fitted
show that e
a
n
o unified an
d
fo
re, there is
a
are no con
v
o
rithms. Thu
s
rformance of
t
s caused by
t
commonly
u
% of all al
g
y
and diversi
t
mal solution
s
t
ablished for
b
fe
rent type of
c
;
(b) in robot tr
a
d
technologie
artificial in
t
o
logy has en
a
y
is a cuttin
g
a
rge number
in rehabilita
t
a
ch computat
i
d
complete th
e
a
need to dev
e
v
entional gu
i
s
, most resea
r
such algorith
the initial sel
u
sed computa
t
g
orithms. Th
e
t
y, require g
r
s
for specific
b
etter applica
t
c
omputational
i
a
ining trajector
y
s in multipl
e
t
elligence (
A
a
bled doctors
g
-edge techn
o
of historical
t
ion robots.
T
i
onal intellig
e
e
oretical syst
e
e
lop standard
i
i
delines for
r
chers select
m
s. Future re
ection of par
a
t
ional intellig
e
e
se conventi
o
r
adients and
a
problems.
W
t
ion of rehabi
l
ntelligence alg
y
control; (c) i
n
e
frontier fi
e
I),
b
ig data
to obtain dat
a
o
logy that c
r
data [96-98
]
T
he a
p
plicat
i
e
nce algorith
m
e
m for the cur
r
i
zed algorith
m
selecting p
a
parameters b
a
s
earch shoul
d
a
meters. Thir
d
e
nce algorith
m
o
nal computa
a
lways fall i
n
We
propose th
a
l
itation.
o
rithms in this
n
motion intenti
o
e
lds are
p
ro
m
and virtual
r
a
about the r
e
r
eates mathe
m
]
. Such data
i
on of big d
m
has its sp
r
rent computa
t
m
s to improv
e
a
rameters us
e
ased on thei
r
d
develop stra
t
dly, data sho
w
m
s are the G
A
a
tional intelli
g
n
to local opt
i
a
t more algor
i
s
review. (a) I
n
on recognition.
m
ising to e
m
reality (VR).
e
covery of p
a
m
atical mode
are often u
s
d
ata technolo
g
e
cific
t
ional
their
e
d in
r
own
t
egies
w
n in
A
and
g
ence
i
mum
i
th
m
s
n
m
ploy
The
tients
l
s by
s
ually
g
y in
ACCEPTED MANUSCRIPT
reha
b
tech
n
that
Fig
u
reha
b
effe
c
sign
i
p
erc
envi
r
reco
v
opp
o
dev
e
trea
t
just
s
engi
n
b
e d
e
5. C
o
algo
r
algo
r
desi
g
dev
e
liter
a
6. E
x
ove
r
sect
i
b
ilitation ins
t
n
ologies are
t
the active p
a
u
re 3 shows t
h
b
ilitation rob
o
c
tive combin
a
i
ficantly imp
r
AI has
b
ec
o
eives inform
a
r
onment. In
t
v
ery speed [
1
o
rtunities in
e
lopment and
t
men
t
[110].
T
stay in the la
b
Research o
n
n
eers to coo
p
e
veloped to p
r
o
nclusion
This article
r
ithms in t
h
r
ithms have
b
g
n and robot
e
lopment situ
a
ture in the re
x
pert opinion
Robot-aided
r
recent yea
r
s
i
ons, robotic
n
t
ruments ca
n
t
he most effe
c
a
rticipation o
f
h
e gene
r
al app
o
ts with VR
a
tion is achie
r
ove which sp
e
Figure3. T
h
o
me an impo
r
a
tion throug
h
t
he rehabilita
t
1
08,109]. The
traditional h
e
medical inst
i
T
hus, rehabili
t
b
oratory for s
c
n
rehabilitati
o
p
erate with th
e
r
omote the fu
t
reviews the
h
e medical
f
b
een discusse
d
control strat
e
ation and re
h
cent five yea
r
interventions
due to the i
n
n
eurorehabili
t
n
facilitate t
h
c
tive in the a
c
f
patients in t
r
l
ication proc
e
technology
d
ved, the pati
e
e
ed up the re
c
h
e process of a
p
r
tant field wi
t
h
machines a
n
t
ion industry,
combination
e
alth fields,
h
i
tutions to c
o
t
a
t
ion robots
a
c
ientific resea
r
o
n robots is
e
rapists and
c
t
ure develop
m
application
f
ield. Severa
l
d
, which sho
w
e
gies. We als
o
h
abilitation d
e
r
, hence may i
n
have been
p
l
n
creasing agi
n
ation has a h
u
h
e assessmen
t
c
tive training
t
reatment is
v
e
ss of VR in a
d
o not mobili
z
e
nt’s mental
p
c
overy of peo
p
p
plying VR wit
h
t
h the wides
p
n
d uses it a
s
AI can be u
s
of rehabilita
t
h
elping pati
e
o
mbine medi
c
a
re more like
l
r
ch.
an increasi
n
c
linical neuro
p
m
ent of roboti
c
of rehabilita
t
l
rehabilitati
o
w
great appli
c
o
suggests
p
r
o
emands. Thi
s
nspire furthe
r
l
aying an incr
e
n
g
p
opulation
u
ge therapeut
i
t
of the rec
o
of patients [9
v
ital in neuro
l
robot-assiste
d
z
e the subjec
p
articipation
p
le with disa
b
h
rehabilitatio
n
p
read applica
t
s
knowledge
s
ed in rehabi
l
t
ion robots a
n
e
nts to find
b
c
al resources
l
y to be appli
e
n
gly developi
n
p
hysiologists
.
c
neurorehabi
t
ion robots
w
o
n robots
w
c
ation potenti
a
o
spects fo
r
f
u
s
review is b
a
r
studies on re
h
e
asing role i
n
[111,112]. A
s
i
c potential t
o
o
very proces
s
9
-101]. Rese
a
l
ogical rehabi
d
device. Ho
w
t
ive initiativ
e
and neurolin
g
ilities.
n
robots.
t
ion [105-10
7
for its better
l
itation traini
n
d this techno
l
b
etter soluti
o
in the proce
s
ed
in clinical
s
n
g cross-fiel
d
Advanced t
e
l
itation.
w
ith computa
t
w
ith computa
t
a
l both in me
c
ture directio
n
a
sed on the
a
h
abilitation r
o
the treatmen
t
s
we summari
o
improve bot
h
s
of patients
a
rchers have
f
i
litation [102
-
w
ever, most c
u
e
of patients.
g
uistics effec
t
7
]. It contin
u
r
adaptation
t
n
g and moni
t
l
ogy will hel
p
o
ns via equi
p
s
s of rehabili
t
s
ettings rathe
r
d
, which re
q
e
chnologies s
h
a
tional intelli
g
t
ional intelli
g
chanical stru
c
n
s according
t
a
nalysis of r
e
o
bots.
n
t of stroke p
a
i
zed in the pr
e
t
h training du
r
. VR
f
ound
-
104].
u
rrent
If an
t
will
ously
t
o the
t
oring
p
fin
d
p
ment
t
ation
r
than
q
uires
h
ould
g
ence
g
ence
c
tures
t
o the
e
lated
a
tients
vious
r
ation
ACCEPTED MANUSCRIPT
and precision as well as rehabilitation assessment [19, 20]. Despite the lack of high-quality evidence,
robotic neurorehabilitation is more advanced than traditional therapy. Robotic interventions have
undeniably been confirmed to be more efficient and requires less labor, translating into higher recovery
efficiency and shorter hospital stay [14-16]. However, the majority of clinicians and patients question
the reliability and safety of neurorehabilitation robots and fear secondary injuries during their training
process [113]. Moreover, the current rehabilitation products in the market generally have disadvantages,
such as complex structure and high price. Therefore, for robot-mediated therapy to be the mainstream
rehabilitation treatment in hospitals, the risk of these problems must be minimized. The occurrence of
computational intelligence algorithms exactly satisfies these needs. Their use effectively reduces the
complexity of the equipment and the production cost, and the optimization of the control methods
ensures the durability and accuracy in the clinical application of the rehabilitation equipment, making the
large-scale clinical use of rehabilitation robots possible [31,62]. Generally, computational intelligence
algorithms successfully improve the effectiveness and economic benefits of robotic neurorehabilitation
and they could realistically be implemented in clinical settings. However, the combination of
computational intelligence algorithms in the field of rehabilitation robots is not perfect, as many
problems and limitations still exist. For example, the blindness of algorithm types and parameter
selection has largely prevented the development of this technology. In the future, research in this field is
still indispensable. Given the existing problems and limitations, a complete theoretical system of
computational intelligence algorithm should be established, and the guidance of parameter selection is
required. There is no clear end in this field, and the pursuit of algorithm performance improvement is
endless. Researchers should endeavor to create new computational intelligence algorithms for achieving
better clinical application.
Furthermore, the study of computational algorithms is not the only promising area in rehabilitation.
Based on the application of computational intelligence algorithms, the involvement of multiple advanced
technologies such as big data, VR and AI also further develop neurorehabilitation robots [100,105]. The
application of big data technology is a good opportunity to promote the development of rehabilitation
assessment which is a promising area in the robotic neurorehabilitation field [96-98]. Established
mathematical models with training data acquired from sensors help neurologists and therapists evaluate
the patient’s recovery and adjust the training plan accordingly [114-116]. VR technology is a potential
method which has been confirmed to be beneficial for better achievement of active training [102-104].
Most rehabilitation pieces of equipment are mainly designed for passive training so far. If VR technology
can be maturely used, a large number of robotic neurorehabilitation that could realize the function of both
active and passive training will emerge, and the value of rehabilitation equipment in clinical applications
will be further enhanced. Moreover,AI is a study direction for achieving intelligent robots [105-107].
Through real-time monitoring of patient motion data combined with corresponding rehabilitation
assessment indicators, rehabilitation robots using AI technology will hopefully realize the timely
adjustment of training plans, making rehabilitation training more efficient and clinically valuable
[108,109]. By excluding these technical factors, some questions could be answered for better clinical
application. The exploration of how to use rehabilitation robots correctly and efficiently also is urgent
work. There is no agreement in the literature about the best duration and frequency of treatment. Only a
few articles focus on how to use rehabilitation devices properly to treat patients in different periods
according to their severity and find out possible side effects. Most clinical data are required for
answering these questions. Through statistics and analysis of clinical data, corresponding conclusions
are promising to draw, which are conducive to the production of robotic neurorehabilitation guidelines.
ACCEPTED MANUSCRIPT
Because therapists are not required to understand rehabilitation devices as engineers, robotic guidelines
are essential for the application of robotic devices in clinical settings. Additionally, factors affecting the
effect of robotic neurorehabilitation have to be searched. Most rehabilitation pieces of equipment
always achieve treatment by the repetitive movement of a simple action at present [63]. Training
trajectories have excessively stereotyped movements, which is less helpful for the actual recovery. The
generation of more efficient and effective rehabilitation training trajectories is a promising avenue for
further research. Moreover, robot-aided therapy and conventional therapy are always regarded as two
opposing sides for comparison in multiple studies. However, few researchers focus on whether a
combination of the two in treatment could achieve better results. Generally, to achieve a true robotic
application in clinical settings, various questions should be answered.
Though various difficulties still exist in the process of large-scale robotic applications in clinical
settings, we strongly believe robotic neurorehabilitation is one of the emerging and sunrise industries in
the future. Although the field of rehabilitation and robotic neurorehabilitation is still in its infancy, with
the increase in aging population in need of rehabilitation treatment,this field is expected to evolve into
a hot healthcare topic attracting significant attention in a few years to come [6]. In the next five years, this
field will still be in a period of rapid development. The high and advanced technologies such as big data,
VR and AI should greatly improve and increase the quality and function of rehabilitation. In addition,
there should be more conferences on rehabilitation to promote communication and learning among
researchers and scholars. The standard operating procedure of robots in clinical settings will be regulated
by robotic guidelines rather than operate according to the personal experience of physical therapists. The
hospitals will establish a medical insurance system for robotic treatment and make a personalized
rehabilitation training plan according to the injury situation of each patient [117]. As a result, the process
of robotic treatment will gradually form a feasible and effective standard process procedure, making it
possible for rehabilitation equipment to be widely used in clinical settings.
Author Contributions
Zhao, Peng and Yang had full access to all the data in the study and take responsibility for the
integrity of the data and the accuracy of the data analysis. Concept and design: Yang, Qiu, Du. Drafting
of the manuscript: All authors. Critical revision of the manuscript for important intellectual content: Qiu,
Wang, Yang.
Funding
This work was funded by the National Natural Science Foundation of China (31971242), Key
grants from Chongqing Science and Technology Bureau to G.W (cstc2019jcyj-zdxm0033) and the
National Key Technology R & D Program of China (SQ2018YFC010099) as well as Fundamental
Research Funds for the Central Universities (2020CDCGJ011) and Chongqing Municipal Education
Commission, China (KYYJ202001).
Declaration of Interest
The authors have no relevant affiliations or financial involvement with any organization or entity
with a financial interest in or financial conflict with the subject matter or materials discussed in the
manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert
testimony, grants or patents received or pending, or royalties.
Reviewer Disclosures
Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.
ACCEPTED MANUSCRIPT
References
Papers of special note have been highlighted as either of interest (*) or of considerable interest
(**) to readers
1. Mozaffarian D, Benjamin EJ, Go AS, et al. Heart Disease and Stroke Statistics-2015 Update A
Report From the American Heart Association [Article]. Circulation. 2015 Jan;131(4):E29-E322.
2. Nordin NAM, Abd Aziz NA, Aziz AFA, et al. Exploring views on long term rehabilitation for
people with stroke in a developing country: findings from focus group discussions. Bmc
Health Services Research. 2014 Mar 10;14.
3. He Q, Wu C, Luo H, et al. Trends in in-hospital mortality among patients with stroke in China.
PLoS One. 2014;9(3):e92763.
4. Meng G, Huang Y, Yu Q, et al. Adopting Text Mining on Rehabilitation Therapy Repositioning
for Stroke. Front Neuroinform. 2019;13:17.
5. Pothiban L, Srirat C. Association between stroke knowledge, stroke awareness, and
preventive behaviors among older people: A cross-sectional study. Nurs Health Sci. 2019
Sep;21(3):399-405.
6. Smithard DG. Stroke in Frail Older People. Geriatrics (Basel). 2017 Jul 15;2(3).
7. Markus TM, Tsai SY, Bollnow MR, et al. Recovery and brain reorganization after stroke in adult
and aged rats. Ann Neurol. 2005 Dec;58(6):950-3.
8. Carmichael ST, Kathirvelu B, Schweppe CA, et al. Molecular, cellular and functional events in
axonal sprouting after stroke. Exp Neurol. 2017 Jan;287(Pt 3):384-394.
9. Leifer D. Neuronal plasticity and recovery from stroke [Article]. Neuroscientist. 1998
Mar;4(2):68-70.
*Review the neuroplasticity after stroke in rehabilitation
10. Carmichael ST. Plasticity of cortical projections after stroke. Neuroscientist. 2003
Feb;9(1):64-75.
11. Kwakkel G, Kollen BJ, Krebs HI. Effects of robot-assisted therapy on upper limb recovery after
stroke: A systematic review. Neurorehabilitation And Neural Repair. 2008
Mar-Apr;22(2):111-121.
12. Prange GB, Jannink MJA, Groothuis-Oudshoorn CGM, et al. Systematic review of the effect of
robot-aided therapy on recovery of the hemiparetic arm after stroke [Review]. J Rehabil Res
Dev. 2006 Mar-Apr;43(2):171-183.
13. Houtrow AJ, Pruitt DW. Meeting the Growing Need for Pediatric Rehabilitation Medicine
Physicians. Arch Phys Med Rehabil. 2016 Apr;97(4):501-506.
14. Zeng X, Zhu G, Yue L, et al. A Feasibility Study of SSVEP-Based Passive Training on an Ankle
Rehabilitation Robot. J Healthc Eng. 2017;2017:6819056.
15. Rosati G, Gallina P, Masiero S. Design, implementation and clinical tests of a wire-based robot
for neurorehabilitation. Ieee Transactions on Neural Systems And Rehabilitation Engineering.
2007 Dec;15(4):560-569.
16. Rosati G. The place of robotics in post-stroke rehabilitation. Expert Review Of Medical
Devices. 2010 Nov;7(6):753-758.
17. Volpe BT, Krebs HI, Hogan N. Is robot-aided sensorimotor training in stroke rehabilitation a
realistic option? [Article]. Curr Opin Neurol. 2001 Dec;14(6):745-752.
**overview of therapeutic potential in robotic neurorehabilitation
18. Krebs HI, Hogan N, Aisen ML, et al. Robot-aided neurorehabilitation [Clinical Trial; Controlled
ACCEPTED MANUSCRIPT
Clinical Trial; ; Research Support, Non-U.S. Gov't; Research Support, U.S. Gov't, Non-P.H.S.].
IEEE transactions on rehabilitation engineering : a publication of the IEEE Engineering in
Medicine and Biology Society. 1998 1998;6(1):75-87.
19. Yoo DH, Kim SY. Effects of upper limb robot-assisted therapy in the rehabilitation of stroke
patients. Journal Of Physical Therapy Science. 2015 Mar;27(3):677-679.
20. Abdullah HA, Tarry C, Datta R, et al. Dynamic biomechanical model for assessing and
monitoring robot-assisted upper-limb therapy [Article; Proceedings Paper]. J Rehabil Res Dev.
2007;44(1):43-61.
21. Chen B, Ma H, Qin LY, et al. Recent developments and challenges of lower extremity
exoskeletons. J Orthop Translat. 2016 Apr;5:26-37.
22. Merrick MA. Secondary injury after musculoskeletal trauma: A review and update [Review]. J
Athl Train. 2002 Apr-Jun;37(2):209-217.
23. Yang GZ, Riener R, Dario P. To integrate and to empower: Robots for rehabilitation and
assistance [Editorial Material]. Sci Robot. 2017 May;2(6):2.
24. Blum C, Roli A. Metaheuristics in combinatorial optimization: Overview and conceptual
comparison. Acm Computing Surveys. 2003 Sep;35(3):268-308.
25. Holland JH. GENETIC ALGORITHMS [Article]. SciAm. 1992 Jul;267(1):66-72.
26. Yang XS, Chien SF, Ting TO. Computational intelligence and metaheuristic algorithms with
applications. ScientificWorldJournal. 2014;2014:425853.
*A detailed introduction to computational intelligence algoritms
27. Lopez-Rubio E, Elizondo DA, Grootveld M, et al. Computational intelligence techniques in
medicine. Comput Math Methods Med. 2015;2015:196976.
28. Poli P, Morone G, Rosati G, et al. Robotic technologies and rehabilitation: new tools for stroke
patients' therapy. Biomed Res Int. 2013;2013:153872.
29. Bamforth SE, Brookes NJ. Customer-focused design methodologies and their suitability for
the rehabilitation industry. Westminister: Professional Engineering Publishing Ltd; 2001.
English. (Pham DT, Dimov SS, Ohagan V, editors. Advances In Manufacturing Technology - Xv).
30. Ferreira F, Chaves MEA, Oliveira VC, et al. Effectiveness of robot therapy on body function and
structure in people with limited upper limb function: A systematic review and meta-analysis.
PLoS One. 2018;13(7):e0200330.
*Review the limitations of robotic neurorehabilitation in clinical settings
31. Shi D, Zhang WX, Zhang W, et al. A Review on Lower Limb Rehabilitation Exoskeleton Robots
[Review]. Chinese Journal Of Mechanical Engineering. 2019 Aug;32(1):11.
32. Weber LM, Stein J. The use of robots in stroke rehabilitation: A narrative review.
NeuroRehabilitation. 2018;43(1):99-110.
33. Jamwal PK, Hussain S, Xie SQ. Review on design and control aspects of ankle rehabilitation
robots. Disabil Rehabil Assist Technol. 2015 Mar;10(2):93-101.
34. Hussain S. State-of-the-art robotic gait rehabilitation orthoses: Design and control aspects.
Neurorehabilitation. 2014 2014;35(4):701-709.
35. Cabrera JA, Nadal F, Muñoz JP, et al. Multiobjective constrained optimal synthesis of planar
mechanisms using a new evolutionary algorithm. Mechanism and Machine Theory.
2007;42(7):791-806.
36. Ali M, Siarry P, Pant M. An efficient Differential Evolution based algorithm for solving
multi-objective optimization problems. European Journal of Operational Research. 2011.
ACCEPTED MANUSCRIPT
37. Wang C, Fang Y, Guo S. Multi-objective optimization of a parallel ankle rehabilitation robot
using modified differential evolution algorithm. Chinese Journal of Mechanical Engineering.
2015;28(4):702-715.
38. Yong Kwun L, Shimoyama I. A skeletal framework artificial hand actuated by pneumatic
artificial muscles. Advanced Robotics. 2012;13(3):349-350.
39. Zhao X, Zi B. Design and analysis of a pneumatic muscle driven parallel mechanism for
imitating human pelvis. Proceedings of the Institution of Mechanical Engineers, Part C:
Journal of Mechanical Engineering Science. 2013;228(4):723-741.
40. Zi B, Yin G, Zhang D. Design and Optimization of a Hybrid-Driven Waist Rehabilitation Robot.
Sensors (Basel). 2016 Dec 14;16(12).
41. Jamwal PK, Hussain S, Xie SQ. Three-Stage Design Analysis and Multicriteria Optimization of a
Parallel Ankle Rehabilitation Robot Using Genetic Algorithm. IEEE Transactions on
Automation Science and Engineering. 2015;12(4):1433-1446.
42. Jamwal PK, Hussain S. Design optimization of a cable actuated parallel ankle rehabilitation
robot: A fuzzy based multi-objective evolutionary approach. Journal of Intelligent & Fuzzy
Systems. 2016;31(3):1897-1908.
43. Jamwal PK, Hussain S. Multicriteria Design Optimization of a Parallel Ankle Rehabilitation
Robot: Fuzzy Dominated Sorting Evolutionary Algorithm Approach. IEEE Transactions on
Systems, Man, and Cybernetics: Systems. 2016;46(5):589-597.
44. Lamine H, Laribi MA, Bennour S, et al. Design study of a cable-based gait training machine.
Journal of Bionic Engineering. 2017;14(2):232-244.
45. Gabardi M, Solazzi M, Leonardis D, et al. Design and Evaluation of a Novel 5 DoF
Underactuated Thumb-Exoskeleton. IEEE Robotics and Automation Letters.
2018;3(3):2322-2329.
46. Enferadi J, Shahi A. On the position analysis of a new spherical parallel robot with orientation
applications. Robotics and Computer-Integrated Manufacturing. 2016;37:151-161.
47. Laribi MA, Carbone G, Zeghloul S. On the Optimal Design of Cable Driven Parallel Robot with
a Prescribed Workspace for Upper Limb Rehabilitation Tasks. Journal of Bionic Engineering.
2019;16(3):503-513.
48. Zhong CH, Yang XJ. Driving torque reduction in linkage mechanisms using joint compliance for
robot head [Article]. Chinese Journal Of Mechanical Engineering. 2015 Sep;28(5):888-895.
49. Lu Y, Leinonen T. Unified type synthesis of planar and spatial mechanisms by systematic
linkage technique. Beijing: China Machine Press; 2004. English. (Huang T, editor. Eleventh
World Congress In Mechanism And Machine Science, Vols 1-5, Proceedings).
50. Liu XJ, Wang QM, Wang JS. Kinematics, dynamics and dimensional synthesis of a novel 2-DoF
translational manipulator. Journal Of Intelligent & Robotic Systems. 2005 Jan;41(4):205-224.
51. Dai JS, Delun W, Cui L. Orientation and Workspace Analysis of the Multifingered Metamorphic
Hand—Metahand. IEEE Transactions on Robotics. 2009;25(4):942-947.
52. Gao J, Li M, Allison G, et al. Optimal dimensional synthesis of a symmetrical five-bar planar
upper-extremity neuromotor device. Chinese Journal of Mechanical Engineering.
2015;28(4):684-690.
53. Shao Y, Xiang Z, Liu H, et al. Conceptual design and dimensional synthesis of cam-linkage
mechanisms for gait rehabilitation. Mechanism and Machine Theory. 2016;104:31-42.
54. McDaid AJ. Design, Analysis, and Multicriteria Optimization of an Overground Pediatric
ACCEPTED MANUSCRIPT
Robotic Gait Trainer. IEEE/ASME Transactions on Mechatronics. 2017;22(4):1674-1684.
55. Singh R, Chaudhary H, Singh AK. A novel gait-based synthesis procedure for the design of
4-bar exoskeleton with natural trajectories. J Orthop Translat. 2018 Jan;12:6-15.
56. Singh R, Chaudhary H, Singh AK. A Novel Gait-Inspired Four-Bar Lower Limb Exoskeleton To
Guide the Walking Movement. Journal of Mechanics in Medicine and Biology. 2019;19(04).
57. Singh R, Chaudhary H, Singh AK. A new hybrid teaching-learning particle swarm optimization
algorithm for synthesis of linkages to generate path [Article]. Sadhana-Acad Proc Eng Sci.
2017 Nov;42(11):1851-1870.
58. Su HJ, McCarthy JM. A polynomial homotopy formulation of the inverse static analysis of
planar compliant mechanisms [Article]. J Mech Des. 2006 Jul;128(4):776-786.
59. Jin M, Zhang X, Zhu B, et al. Spring-joint method for topology optimization of planar passive
compliant mechanisms. Chinese Journal of Mechanical Engineering. 2013;26(6):1063-1072.
60. Chau NL, Le HG, Dao T-P, et al. Design and Optimization for a New Compliant Planar Spring of
Upper Limb Assistive Device Using Hybrid Approach of RSM–FEM and MOGA. Arabian Journal
for Science and Engineering. 2019;44(9):7441-7456.
61. Cristian A, Green J. Patient Safety and Quality Improvement in Rehabilitation Medicine
[Article]. Phys Med Rehabil Clin N Am. 2012 May;23(2):221-+.
62. Miao Q , Zhang M, Cao J, et al. Reviewing high-level control techniques on robot-assisted
upper-limb rehabilitation. Advanced Robotics. 2018;32(24):1253-1268.
63. Saglia JA, Tsagarakis NG, Dai JS, et al. Control Strategies for Patient-Assisted Training Using the
Ankle Rehabilitation Robot (ARBOT). IEEE/ASME Transactions on Mechatronics.
2013;18(6):1799-1808.
64. Langhorne P, Bernhardt J, Kwakkel G. Stroke Care 2 Stroke rehabilitation [Article]. Lancet.
2011 May;377(9778):1693-1702.
65. Krebs HI, Palazzolo JJ, Dipietro L, et al. Rehabilitation robotics: Performance-based
progressive robot-assisted therapy [Article]. Auton Robot. 2003 Jul;15(1):7-20.
66. Gaino R, Covacic MR, Teixeira MCM, et al. Electrical stimulation tracking control for paraplegic
patients using T–S fuzzy models. Fuzzy Sets and Systems. 2017;314:1-23.
67. Koceska N, Koceski S, Zobel PB, et al. Control Architecture for a Lower Limbs Rehabilitation
Robot System. New York: Ieee; 2009. English. (2008 Ieee International Conference on
Robotics And Biomimetics, Vols 1-4).
68. Taha Z, Abdul Majeed APP, Zainal Abidin AF, et al. A hybrid active force control of a lower limb
exoskeleton for gait rehabilitation. Biomed Tech (Berl). 2018 Jul 26;63(4):491-500.
69. Joyo MK, Raza Y, Ahmed SF, et al. Optimized Proportional-Integral-Derivative Controller for
Upper Limb Rehabilitation Robot. Electronics. 2019;8(8).
70. Belkadi A, Oulhadj H, Touati Y, et al. On the robust PID adaptive controller for exoskeletons: A
particle swarm optimization based approach. Applied Soft Computing. 2017;60:87-100.
71. Li W, Huang D, Zhang M, et al. Bifurcation control of a generalized VDP system driven by
color-noise excitation via FOPID controller. Chaos, Solitons & Fractals. 2019;121:30-38.
72. Lakshmanaprabu SK, Najumnissa Jamal D, Sabura Banu U. Multiloop FOPID Controller Design
for TITO Process Using Evolutionary Algorithm. International Journal of Energy Optimization
and Engineering. 2019;8(3):117-130.
73. Yang XS, Deb S. Cuckoo Search via Levey Flights. In: Abraham A, Herrera F, Carvalho A, et al.,
editors. 2009 World Congress on Nature & Biologically Inspired Computing. World Congress
ACCEPTED MANUSCRIPT
on Nature and Biologically Inspired Computing. New York: Ieee; 2009. p. 210-+.
74. Sharma R, Rana KPS, Kumar V. Performance analysis of fractional order fuzzy PID controllers
applied to a robotic manipulator. Expert Systems with Applications. 2014;41(9):4274-4289.
75. Calderón AJ, Vinagre BM, Feliu V. Fractional order control strategies for power electronic buck
converters. Signal Processing. 2006;86(10):2803-2819.
76. Ayas MS, Altas IH, Sahin E. Fractional order based trajectory tracking control of an ankle
rehabilitation robot. Transactions of the Institute of Measurement and Control.
2016;40(2):550-564.
77. Li M, Meng W, Hu J, et al. Adaptive Sliding Mode Control of Functional Electrical Stimulation
(FES) for Tracking Knee Joint Movement. 2017 10th International Symposium on
Computational Intelligence and Design (ISCID)2017. p. 346-349.
78. Azar WA, Akbarimajd A, Parvari E. Intelligent Control Method of a 6-DOF parallel robot Used
for Rehabilitation Treatment in lower limbs. Automatika. 2017;57(2):466-476.
79. Ayas MS, Altas IH. Fuzzy logic based adaptive admittance control of a redundantly actuated
ankle rehabilitation robot. Control Engineering Practice. 2017;59:44-54.
80. Silawatchananai C, Parnichkun M. Haptics control of an arm exoskeleton for virtual reality
using PSO-based fixed structure H∞ control. International Journal of Advanced Robotic
Systems. 2019;16(3).
81. Ott C, Lee D, Nakamura Y, et al. Motion Capture based Human Motion Recognition and
Imitation by Direct Marker Control. 2008 8th Ieee-Ras International Conference on
Humanoid Robots. IEEE-RAS International Conference on Humanoid Robots. New York: Ieee;
2008. p. 406-412.
82. Fukuda O, Tsuji T, Kaneko M, et al. A human-assisting manipulator teleoperated by EMG
signals and arm motions. IEEE Transactions on Robotics and Automation.
2003;19(2):210-222.
83. Cedras C, Shah M. MOTION-BASED RECOGNITION - A SURVEY [Article]. Image Vis Comput.
1995 Mar;13(2):129-155.
84. Zhu B, Wan Z, Xu Y. The Recognition of Motion Intention of Knee Joint Based on Piezoelectric
Signals. Journal of Physics: Conference Series. 2019;1176.
85. Zhang L, Liu G, Han B, et al. sEMG Based Human Motion Intention Recognition. Journal of
Robotics. 2019;2019:1-12.
86. Nas K, Gur A, Cevik R, et al. The relationship between physical impairment and disability
during stroke rehabilitation: effect of cognitive status [Article]. Int J Rehabil Res. 2004
Sep;27(3):181-184.
87. Pasha Zanoosi AA, Naderi D, Sadeghi-Mehr M, et al. Investigation of the effects of human
body stability on joint angles’ prediction. Multibody System Dynamics. 2015;35(2):111-129.
88. Feng X, Ai Q, Meng W. KUKA Real-Time Control through Angle Estimation of Wrist from sEMG
with Support Vector Regression. Proceedings of the 2018 2nd International Conference on
Computer Science and Artificial Intelligence - CSAI '182018. p. 389-393.
89. Buongiorno D, Barsotti M, Barone F, et al. A Linear Approach to Optimize an EMG-Driven
Neuromusculoskeletal Model for Movement Intention Detection in Myo-Control: A Case
Study on Shoulder and Elbow Joints. Front Neurorobot. 2018;12:74.
90. Peng L, Hou Z-G, Luo L, et al. An sEMG-Driven Neuromusculoskeletal Model of Upper Limb for
Rehabilitation Robot Control. 2016. (2016 Ieee International Conference on Robotics And
ACCEPTED MANUSCRIPT
Biomimetics).
91. Noughaby AG, Vossoughi GR. The Control of an Exoskeleton and The Reduction of Interaction
Force Using Human Intent Detection by EMG Signals and Torque Estimation. 2018 6th Rsi
International Conference on Robotics And Mechatronics. RSI International Conference on
Robotics and Mechatronics ICRoM2018. p. 536-541.
92. Zhang X, Dai J, Li X, et al. sEMG-based shoulder-elbow composite motion pattern recognition
and control methods for upper limb rehabilitation robot. Assembly Automation.
2019;39(3):394-400.
93. Cai S, Chen Y, Huang S, et al. SVM-Based Classification of sEMG Signals for Upper-Limb
Self-Rehabilitation Training. Front Neurorobot. 2019;13:31.
94. Oskoei MA, Hu HS. Myoelectric control systems-A survey [Review]. Biomedical Signal
Processing And Control. 2007 Oct;2(4):275-294.
95. Millan JD, Rupp R, Muller-Putz GR, et al. Combining Brain-Computer Interfaces and Assistive
Technologies: State-of-the-Art and Challenges. Front Neurosci. 2010;4.
96. Ma Y, Wang Y, Yang J, et al. Big Health Application System based on Health Internet of Things
and Big Data. IEEE Access. 2017;5:7885-7897.
97. Alhussain T. Medical Big Data Analysis Using Big Data Tools and Methods [Article]. J Med
Imaging Health Inform. 2018 May;8(4):793-795.
98. Stroppa F, Loconsole C, Frisoli A. Convex polygon fitting in robot-based neurorehabilitation.
Applied Soft Computing. 2018;68:609-625.
99. Brutsch K, Koenig A, Zimmerli L, et al. Virtual reality for enhancement of robot-assisted gait
training in children with central gait disorders. J Rehabil Med. 2011 May;43(6):493-9.
100. Holden MK. Virtual environments for motor rehabilitation: Review [Review]. CyberPsychol
Behav. 2005 Jun;8(3):187-211.
101. Tutak JS. VIRTUAL REALITY AND EXERCISES FOR PARETIC UPPER LIMB OF STROKE SURVIVORS
[Article]. Teh Vjesn. 2017 Sep;24:451-458.
102. Pennycott A, Wyss D, Vallery H, et al. Towards more effective robotic gait training for stroke
rehabilitation: a review [Review]. J NeuroEng Rehabil. 2012 Sep;9:13.
103. Koenig A, Omlin X, Bergmann J, et al. Controlling patient participation during robot-assisted
gait training [Article]. J NeuroEng Rehabil. 2011 Mar;8:12.
104. Nielsen JB, Willerslev-Olsen M, Christiansen L, et al. Science-Based Neurorehabilitation:
Recommendations for Neurorehabilitation From Basic Science [Review]. J Mot Behav.
2015;47(1):7-17.
105. Brunette ES, Flemmer RC, Flemmer CL. A Review of Artificial Intelligence. New York: Ieee;
2009. English. (Gupta GS, Mukhopadhyay SC, editors. Proceedings Of the Fourth International
Conference on Autonomous Robots And Agents).
106. Novak D, Riener R. Control Strategies and Artificial Intelligence in Rehabilitation Robotics
[Article]. AI Mag. 2015 Win;36(4):23-33.
107. Parker G. Are robots smart? [Editorial Material]. Industrial Robot. 2001;28(4):279-280.
108. Lisboa PJG. A review of evidence of health benefit from artificial neural networks in medical
intervention [Review]. Neural Netw. 2002 Jan;15(1):11-39.
109. Su Y, Su L. Intelligent Medical Rehabilitation Training Instrument Based on Movement
Coordination. IEEE Access. 2020;8:73373-73387.
**Interesting review of some potential technologies in rehabilitation
ACCEPTED MANUSCRIPT
110. Ling W, Yu G, Li Z. Lower Limb Exercise Rehabilitation Assessment Based on Artificial
Intelligence and Medical Big Data. IEEE Access. 2019;7:126787-126798.
111. Oña ED, Garcia-Haro JM, Jardón A, et al. Robotics in Health Care: Perspectives of Robot-Aided
Interventions in Clinical Practice for Rehabilitation of Upper Limbs. Applied Sciences.
2019;9(13).
112. Rehmat N, Zuo J, Meng W, et al. Upper limb rehabilitation using robotic exoskeleton systems:
a systematic review. International Journal of Intelligent Robotics and Applications.
2018;2(3):283-295.
113. Yue Z, Zhang X, Wang J. Hand Rehabilitation Robotics on Poststroke Motor Recovery. Behav
Neurol. 2017;2017:3908135.
114. Lorish TR, Sandin KJ, Roth EJ, et al. STROKE REHABILITATION .3. REHABILITATION EVALUATION
AND MANAGEMENT [Article]. Arch Phys Med Rehabil. 1994 May;75(5):S47-S51.
115. Lynch EA, Luker JA, Cadilhac DA, et al. Rehabilitation Assessments for Patients With Stroke in
Australian Hospitals Do Not Always Reflect the Patients' Rehabilitation Requirements [Article].
Arch Phys Med Rehabil. 2015 May;96(5):782-789.
116. Zeng X, Zhu G, Zhang M, et al. Reviewing Clinical Effectiveness of Active Training Strategies of
Platform-Based Ankle Rehabilitation Robots. J Healthc Eng. 2018;2018:2858294.
117. Proietti T, Crocher V, Roby-Brami A, et al. Upper-Limb Robotic Exoskeletons for
Neurorehabilitation: A Review on Control Strategies. IEEE Rev Biomed Eng. 2016;9:4-14.
ACCEPTED MANUSCRIPT