Biol Cybern (2015) 109:141–162
DOI 10.1007/s00422-014-0635-1
REVIEW
Development of a control system for artificially rehabilitated
limbs: a review
M. S. H. Bhuiyan · I. A. Choudhury · M. Dahari
Received: 14 June 2013 / Accepted: 28 October 2014 / Published online: 10 December 2014
© Springer-Verlag Berlin Heidelberg 2014
Abstract Development of an advanced control system for
prostheses (artificial limbs) is necessary to provide function-
ality, effectiveness, and preferably the feeling of a sound liv-
ing limb. The development of the control system has intro-
duced varieties of control strategies depending on the appli-
cation. This paper reviews some control systems used for
prosthetics, orthotics, and exoskeletons. The advantages and
limitations of different control systems for particular appli-
cations have been discussed and presented in a comparative
manner to help in deciding the appropriate method for perti-
nent application.
Keywords Prosthesis · Control system · Non-linear
control strategy · Control system element
1 Introduction
Amputation or a limb defect creates the need for getting help
from different supporting devices. Amputees have been aided
by a walking stick/cane, walking frame, crutch, wheelchair,
motor wheelchair, stroller, motor stroller, and more recently
by artificial prosthetics (Gao et al. 2010). The amputee with
a transfemoral and knee level amputation generally uses a
wheelchair, whereas an amputee with lower-level amputa-
tions (transtibial and foot amputation) primarily uses a pros-
M. S. H. Bhuiyan (B) · I. A. Choudhury · M. Dahari
Manufacturing System Integration, Department of Mechanical
Engineering, Faculty of Engineering, University of Malaya,
50603 Kuala Lumpur, Malaysia
e-mail: sayem_um@yahoo.com
I. A. Choudhury
e-mail: imtiaz@um.edu.my
M. Dahari
e-mail: mahidzal@um.edu.my
thesis (Karmarkar et al. 2009). A prosthesis is a device that
replaces a missing extremity, such as an arm or leg. The extent
of an amputation or loss and location of the missing extremity
largely determine the type of prosthesis to be used. Recently,
the design of prosthetics has been significantly improved by
incorporating some extra features as well as introducing new
controlling methods.
In newer and more improved designs of prosthetics, more
control is given to the users by employing different con-
trol systems, muscle of carbon fiber, mechanical linkages,
motors, computer microprocessors, and innovative combi-
nations of these technologies (Wen-Wei Hsu et al. 1999).
More importantly, essential factors such as weight-force
ratio, strength, durability, adaptability, wearability, degree of
freedom, resistance to environment, and functional capabili-
ties are being addressed seriously to develop a more efficient
prosthesis. The operating power consumed by the prosthe-
sis is another essential factor to be considered, which varies
depending on the prosthesis type. According to Johnson et
al. (1997), a transfemoral amputee must use approximately
80 % more energy to walk than a normal person (with two
whole legs). Currently, new plastics and other materials, such
as carbon fiber, have allowed prosthetics to be stronger and
lighter, limiting the amount of extra energy necessary to oper-
ate the limb (Sawers and Hahn 2011; Esposito et al. 1997).
Because of the exceptional strength-to-weight characteristics
and quality of superior biocompatibility, a majority of today’s
upper-limb and lower-limb prostheses have now been made
from composites with underlying polymer matrix (Scholz et
al. 2011). More advanced control systems and more com-
patible choice of material will ensure more efficient func-
tionality of prostheses. An automated control system could
provide more realistic movement to the prosthetic’s compo-
nents and to itself. For instance, a myoelectric control system,
which controls the limbs by converting muscle movements
123
142
to electrical signals may allow the amputees to control the
prosthesis more directly (Jawhar et al. 2011).
The latest research has focused on the development of an
artificial prosthetic limb, which would be capable of imi-
tating the behavior of a biological limb. The desired move-
ment of the prosthesis is obtained with the help of different
types of joints, sensors, controllers, and actuators (Martins et
al. 2012). Application of artificial intelligence has made the
control system more spontaneous with the ability of decision
making.
This article presents a brief discussion on the importance
of prosthesis construction followed by a thorough study on
control system development, and the different components of
the system. A comprehensive review has been carried out on
available control systems for prosthesis operation, and thus
their suitability for particular application has been weighed
and illustrated meticulously to help the future user in choos-
ing an appropriate control system for a pertinent problem. A
systematic approach has been followed to highlight advan-
tages and limitations of various control systems for specific
applications.
2 Importance of prosthesis reconstruction
The necessity of prosthesis reconstruction arises because of
the increasing rate of limb loss due to accidents, war, dis-
eases, or congenital defects. In recent years, there has been
an increased rate of diseases causing paraplegia due to spinal
cord injury or from different types of stroke incidences. Func-
tional electrical stimulation (FES) has been applied as a tool
to improve trunk stability in paraplegic subjects during activ-
ities of daily living (ADL) (Kukke and Triolo 2004; Newham
and Donaldson Nde 2007; Vanoncini et al. 2012). However,
the major drawbacks of FES are that it causes early damage of
cells, sagging of muscles, and in the long run it loses the abil-
ity to stimulate the muscle (Matjačić et al. 2003; Vanoncini et
al. 2012). In this case, use of orthotics is preferred over FES
by the patients with paraplegia. Orthoses are used to sup-
port the patient with limb problems using different types of
passive or active attachments without stimulating the muscle
externally.
Damage to limbs is another commonly seen defect among
the disabled due to diseases, congenital defects, accidents, or
war, which most often results in amputation (Sagawa et al.
2011). Industrial, vehicular, and war-related accidents are the
leading causes of amputations in developing countries, such
as those in large portions of Africa (Burger et al. 2004). In
more developed countries like those in North America and
Europe, disease is the primary cause of amputations (Rosen-
feld et al. 2000). Cancer, infection, and circulatory disease
are the leading diseases that result amputation (Albertini et
al. 2000). Trauma is another prevailing cause of amputation
123
Biol Cybern (2015) 109:141–162
(Gallagher and MacLachlan 2004). In addition to trauma,
dysvascular disease is a predominant cause of lower limb
amputation (60 % case) (Resnik et al. 2012; Bryant and Pan-
dian 2001). However, upper limb amputation due to disease
is found to be relatively rare (3 % case) (Resnik et al. 2012). A
study carried out by Esquenazi and Meier Iii (1996) reported
that the estimated ratio of arm to leg amputations was 1:3.
Similar results have been obtained from a recent study con-
ducted by Watve et al. (2011). They have also added that the
most common level of amputation is below the knee. Accord-
ing to Sagawa et al. (2011), approximately 1.7 million people
experienced limb loss in the United States in 2007, and more
than 185,000 new amputations are performed each year in
that country. Ziegler-Graham et al. (2008) stated in a sur-
vey report that the number of amputees in the USA in 2005
was 1.6 million, which is expected to more than double to
3.6 million by 2050. These people will need to have limbs
replaced using the correct prosthesis arrangement to ensure
a high-quality, active life.
A study (Couture et al. 2011) on patients with limb dis-
eases, defective limbs, or seriously injured limbs showed that
amputation helps patients to have greater functional indepen-
dence and greater body image satisfaction than those patients
without amputations with limb defects. On the other hand,
amputation alters the biomechanics of the amputee’s body
movement.
According to Gailey et al. (2008), the intact limb of lower-
limb amputees is often stressed or favored more during their
everyday activities. This primarily causes osteoarthritis of the
knee and/or hip joints of the intact limb, which in the long run
causes osteoporosis that limits sufficient loading of the lower
limb through the long bones. A poor prosthetic fit and align-
ment, postural changes, leg-length discrepancy, amputation
level, and general deconditioning commonly cause back pain
to the lower-limb amputee. The right selection of prosthe-
sis with appropriate physics would be helpful in this case,
which is usually defined by the amputee’s body construc-
tion and his/her activity. The correct choice of a control sys-
tem complements the other part of prosthesis development,
which is usually performed by some external device such as
electrical, mechanical, and artificial intelligence. All of the
difficulties associated with amputation necessitate that the
amputee obtain a tailor-made prosthesis with an appropriate
control system, which should be capable of balancing their
body dynamics and providing desired movement to its user.
3 Development of a control system for prostheses
The prosthesis consists of some mechanical elements replac-
ing the missing extremity and some electrical components
that support desired movement in the prosthesis arrangement.
The electrical and mechanical components independently or
Biol Cybern (2015) 109:141–162
combined with artificial intelligence form a system that gives
control of the prosthesis to its user and it is regarded as the
control system for the prosthesis.
3.1 Background
Prior to developing a control system for a prosthesis, one
has to study thoroughly the body dynamics and function-
ality of the subject’s impaired limb. Some intervention and
assessment tools are available to identify the major factors
that contribute to the risk of falling (Maki et al. 2011). The
relationships among static ground friction, hip torque, and
the trajectory of the prosthetic foot has been investigated
with a two-dimensional model proposed by van Keeken et
al. (2012). The static and dynamic balancing of perturbation
of an amputee has been assessed by measuring the weight
distribution force below the prosthetic foot (Nederhand et al.
2012). In another investigation, the center of pressure (COP)
and center of mass (COM) were indicated as the major para-
meters to be controlled to balance an amputee’s body dynam-
ics (Mergner 2010). The gait phase mechanism (stance and
swing) under different walking conditions was identified to
be useful in determining the intended limb motion (Jiménez-
Fabián and Verlinden 2012). On the other hand, Nagano et
al. (2005) studied muscle force, work, and power output of
major lower limb muscles during jumping. Hemami and Dar-
iush (2012) modeled and simulated lower-limb movement as
a system of connected rigid bodies. They studied the behav-
ior of the central nervous system (CNS) when producing the
desired trajectories of motion and the force of contact prior
to the control of the prosthesis. The projection of force in
the sensory cortex was implemented with a standard integral
unit controller in the spinal cord and a well-defined neural
population unit. It was extended to the case for direction of
motion with two other neural units of visual input and spindle
efference.
An effective control system is required to recreate a sim-
ilar force-length-velocity mechanism, and thus to imitate
the biological limb control faculty to produce the required
movement to the prosthesis. The effectiveness of the pros-
thesis depends on its ability to simulate the functionality of
a real, healthy biological limb. Therefore, an efficient pros-
thesis should have all the same faculties as a natural limb.
A natural limb has the degree of freedom to produce differ-
ent directional movement, ability to receive commands from
the possessor and perform accordingly, e.g. for lower limb:
standing, walking, running, jumping, and for upper limb:
holding, lifting, and carrying, (Previdi 2002; Previdi et al.
2004). All movements of a biological limb are reportedly
nonlinear due to gravity and muscle contraction dynamics
(Previdi 2002; Previdi et al. 2004). The moment generated at
the joint during a non-isometric muscle contraction depends
very strongly on muscle length and velocity. These variables
143
are functions of a set of joint angles and their derivatives,
which are nonlinear. In addition, gravity and elastic passive
moment cause a nonlinear term to be present in the equation
of limb movement. In a prosthesis control system, similar
nonlinear movements are reproduced by representing the var-
ious nonlinear movements into different nonlinear equations
(Schauer and Hunt 2000). Development of a control system
for a nonlinear system like a prosthesis is much more chal-
lenging than controlling a linear system (Tan 2009). Different
types of closed-loop, nonlinear control strategies have been
used to control different movements of prosthetics (Pujana-
Arrese et al. 2010; Yap et al. 2007).
Depending on the aim, the prosthesis controllers are
equipped with different task controllers that can be formed
in several ways based on application. The control action of
a prosthesis is usually accomplished by controlling the posi-
tion and force/torque parameters. Based on the physical para-
meters, the prosthesis control system can be classified by
position, torque/force, and force interaction controllers. The
position control scheme is mostly implemented as a low-level
controller, which is commonly used to ensure that the limb
joints turn in a desired angle (Timmermans et al. 2009). Force
interaction controllers are generally used either in the form
of impedance or admittance controllers. The impedance or
admittance controllers are reported to work only if they are
followed by either the force or the position controller, respec-
tively. The combination of the impedance or admittance con-
troller is considered a high-level controller, while the force
or position controller is known to be the low-level controller.
The torque/force controller is mostly used for a physical
parameter-based control system (Nef et al. 2009). The high
level controller, on the other hand, is the impedance con-
troller that controls the interaction force between the human
and the prosthesis by receiving position as input and pro-
ducing force as output. The output force of the impedance
model becomes the reference force to the force/torque con-
troller. Besides the impedance control, the admittance control
is used to control the interaction force between the amputee
and prosthesis. The admittance model receives forces and
produces positions, rather than receiving positions and pro-
ducing forces. A position controller based on the position
references of the admittance model output is used to control
the angle of the prosthesis joints (Carignan et al. 2009). The
impedance and admittance controller produce desired move-
ment in the prosthesis by controlling the interaction force
between the amputee and prosthesis (Anam and Al-Jumaily
2012). The control system for the prosthesis, on the other
hand, has been gradually transformed from mechanical to
electrical and now to a bioelectrical control system (Le et al.
2010; Zhang et al. 2012; del-Ama et al. 2012; Resnik et al.
2012).
Some essential components such as sensors, controller/
microprocessors, actuators, and interfacing units were shown
123
144
• Mechanical
• Electrical
• Hydraulic
• Pneumatic
• Hybrid
Actuator
• Mechanical
• Electrical Control System of
Sensor
• Biosensor Prosthesis
• Bioelectrical
Interfacing
• Non-invasive
o Fuzzy Interfacing
system (FIS)
o Adaptive Neuro-Fuzzy
interfacing (ANFIS)
• Invasive
o Fully exploiting
neuroplastic interfacing
o Natural Interfacing
Fig. 1 Components of a typical control system used for prosthetics
by Klute et al. (2009) to be required for developing a control
system for prosthetics. A more effective combination of the
different elements mentioned would provide better control of
the prosthesis. Figure 1 depicts the essential components of
a typical control system applied in controlling a prosthesis.
In the control system of a prosthesis, mainly four types
of sensors, five types of controllers, five types of actuators,
and two types of interfacing units are used. The components
listed in Fig. 1 are discussed in the following subsections.
3.2 Sensors
Sensors in a prosthesis control system are used to sense
force, torque, position, angle, proximity, power, speed, veloc-
ity, strength, etc., enabling the process to develop an effi-
cient prosthesis. Through stages of improvement, the sensors
presently have reached at a considerable advanced state that
could competently capture the nerve signal. The gait fea-
ture signals are now obtainable using an inertial miniature
accelerometer sensor (Alaqtash et al. 2011). The sensors’
data captured from the limb movement are then input into
a rule-based relational matrix of a fuzzy logic controller for
modulation based on the fuzzy data set (Magdalena 1997).
The sensor can be mechanical, electrical, biological, or
many more types. Zhang and Yin (2012) have developed
and deployed a mechanical sensor based on a shape memory
alloy (SMA) to sense and thus control movement of artifi-
cial skeletal muscle. Dutta et al. (2011) used Ag/AgCl (silver
123
Biol Cybern (2015) 109:141–162
• Proportional (P)
• Proportional Derivative
(PD)
• Proportional Integral
Controller
Derivative (PID)
• Linear-Quadratic
Regulator (LQR)
• Sliding mode controller
(SMC)
metal and silver chloride) electrodes with 2 cm interelectrode
distance following the Surface ElectroMyoGraphy for the
Non-Invasive Assessment of Muscles (SENIAM) method to
collect surface EMG signals. Akagi et al. (2012) employed
a displacement sensor consisting of two photo reflector elec-
trodes to assess muscle movement of the prosthesis. A tri-
axial accelerometer with a dynamic time warping (DTW)
method was applied on the inner forearm of some volun-
teers to recognize limb movement (Muscillo et al. 2011).
Tormene et al. (2009) used 29 strain sensors distributed on
the upper limb of a patient under rehabilitation to observe
their limb movement. Jae-Myung and Yong-Myung (2006)
have employed flex sensors attached to biceps brachii mus-
cles and triceps brachii muscles of alternate arms to identify
the arm’s movement. Inertial sensor technology, accelerome-
ters, and gyroscopes were employed to assess a Paralympics
amputee runner with prosthesis before changing to a new
prosthetic (Lee et al. 2012).
From a thorough study carried out by Schultz and Kuiken
(2011), it has been found that the main challenge in con-
trolling the prosthesis is to read the intention of the user to
perform a movement or action, which creates a need to cap-
ture the signal from the nervous system of the remaining limb
or from the amputee’s brain. This is commonly divided into
two types: invasive and non-invasive. Some sensor examples
are implantable electrodes, pattern recognition of myoelec-
tric signals, targeted re-innervation. Sensors that intercept
nerve signals include extraneural electrodes and intraneural
Biol Cybern (2015) 109:141–162
electrodes. Some information can be obtained directly from
the brain, such as cortical spike activity (action potential
and field potential), local field potential, intracortical stimu-
lation for sensory feedback, electrocorticogram recordings,
or electroencephalogram recordings. Even if there is suffi-
cient motor command information available for multifunc-
tion prosthesis control, adequate control would still require
some degree of sensory feedback. Both tactile and proprio-
ceptive feedback play important roles in volitional movement
(Le et al. 2010). Tactile feedback allows for modulation of
grip forces and hand postures for grasping and object manip-
ulation (Crevecoeur et al. 2011). Proprioceptive feedback is
essential for accurate motor control and joint coordination
(Vaugoyeau et al. 2011).
A large number of articles on sensing techniques avail-
able for clinical application have been revisited by Tiwana
et al. (2012). From their study, the use of tactile sensors in
the rehabilitation of amputees became very obvious, particu-
larly to investigate the severity of the stump–socket inter-
action. To acquire information about the force or torque
applied to a prosthesis during object manipulation or dur-
ing walking is done using visual, auditory, electrical, tactile,
and vibrotactile stimulation. The latest addition to sensing
techniques is a specially designed chemical sensor compris-
ing a biological recognition element and a physicochemi-
cal transducer known as a biosensor. The functionality of
biosensors and actuators are typically based on the concept
of the bio-microelectromechanical systems (MEMSs) (Pon-
mozhi et al. 2012). A barrier to obtaining implantable biosen-
sors is biocompatibility. To eliminate this difficulty, subcuta-
neously implantable needle-type electrodes were developed
(Wang 2001). Because of unique physical and chemical prop-
erties of carbon nanotubes (CNTs), such as their high aspect
ratio, ultra-lightweight, high mechanical strength, high elec-
trical conductivity, and high thermal conductivity, the CNT
has been applied in the biomedical field (Sinha and Yeow
2005). A large variety of potentiometric biosensors were
developed using biocatalytics and bioaffinity-based biosens-
ing schemes. However, only a few of them are considered to
be applicable to biomedical analysis (Wise 2007). The func-
tionality, quality, and longevity of total knee arthroplasty and
other kinds of implants can be enhanced by the improved
design of sensors (Ponmozhi et al. 2012).
3.3 Controllers
The controller is another essential element of a control sys-
tem, which uses the captured signal as input and produces
corresponding output to help facilitate the movement of the
prosthesis to the desired position. The use of some funda-
mental control units such as proportional (P), proportional
derivative (PD), proportional integral (PI), proportional inte-
gral derivative (PID), and linear quadratic regulator (LQR)
145
have been observed extensively in different types of control
systems (Weir 2008). Chang et al. (2011) have proposed a
novel T–S fuzzy model-based controller applied to a one-
dimensional manipulator actuated by a pneumatic artificial
muscle (PAM). The T–S fuzzy model-based controller can
decompose a nonlinear system into a set of linear subsys-
tems. Thus, the simple linear control techniques can suc-
cessfully reproduce complicated nonlinear systems. Timmer-
mans et al. (2009) have used proportional (P), and propor-
tional derivative (PD) controllers to control the arm position
of an exoskeleton robot, ARMIN III. A PID controller has
been applied by Aliff et al. (2012) to control expanding and
contracting of a prosthesis. Kaluza and Cioacă (2012) mod-
eled an artificial neural network-based controller inspired by
biological central pattern generators for a toy robot. Their
proposed neural network-based controller has achieved a
great level of success in controlling the body dynamics of
the toy, and thus effective control of its movement.
3.4 Actuators
The actuator is the next important unit of the prosthesis con-
trol system to execute the instruction from the controller,
and thus to trigger the end effector. The conventional DC
motors and micromotors, some exotic shape-memory alloys,
ultrasonic motors, and pneumatic artificial muscles have been
used as actuators during the past decade. Wide ranges of actu-
ator such as mechanical, electrical, hydraulic, pneumatic, and
hybrid have been used to actuate different systems depending
on the nature of the end effector and its application (Varol
et al. 2009). The large variety of applications has resulted
in a diversity of actuation methods. Micro-ultrasonic, ultra-
sonic, and piezoelectric motors are identified to be the most
promising when compared with hydraulic actuators, shape-
memory alloys, and contractile polymer gels with regards
to size, weight, applicability, precision, hysteresis and non-
repeatability, energy consumption, operating frequency, effi-
ciency, power density, anthropomorphism, and cost (Cura
et al. 2003). Recent advancements in shape-memory alloys
have produced actuators with increased strains of up to 32 %
using a braided coil design; however, response speed, dura-
bility, actuation stress, energy consumption, and hysteresis
persist as obstacles to their implementation (Biddiss and
Chau 2008). Recently, the application of bio-inspired actu-
ators with built-in sensors or control strategies has made
the position controlling more accurate (Nguyen et al. 2012).
Although polymer-based dielectric elastomer actuators pro-
vide advantages in terms of very high functioning frequency
and remarkably large force delivered, they need quite high
applied voltage (over 100 V) (Torop et al. 2012). Recently,
much attention has been focused on soft materials that can
directly transform electrical energy into mechanical work
for a wide range of applications including robotics, tactile
123
146
and optical displays, prosthetic devices, medical devices, and
microelectromechanical systems. Low-voltage electroactive
polymer (EAP) actuators, which can function quickly and
softly, are particularly useful because they can be used as
actuators like artificial muscle in various bio-medical and
human affinity applications (Terasawa and Takeuchi 2013).
An efficient actuator for particular application is cho-
sen based on the actuator’s energy density. Phase-change
actuators were identified to have energy densities orders of
magnitude greater than electrostatic actuators (Whalen et al.
2007). Among all types of actuators, pneumatic muscle actu-
ators apparently have the highest power/weight ratio and
power/volume ratio. Because of their superior quality, this
actuator has been used in rehabilitation engineering and also
in robots (Chang 2010). A series damper actuator made of a
DC servo motor and a rotary type MR damper was found to
control effectively the swing phase of a prosthetic leg (Khan
et al. 2012). Actuator and/or component failure could cause
a serious safety problem to engineering systems if no proper
action is taken in time (Fan and Song 2011). The actuator for
prosthetics, therefore, has to be chosen carefully and kept
under constant monitoring during the operation.
3.5 Interfacing unit
The interfacing unit in a control system establishes a
connection between two different units and transfer sig-
nal/commands from one to another. For an artificial prosthe-
sis control system, the interfacing module develops a connec-
tion between amputee and prosthesis. Schultz and Kuiken
(2011) investigated different interfacing methods available
for controlling artificial prostheses. According to them, the
process of signal transmission, and thus the efficiency of
the control system, are largely influenced by the interfacing
unit.
The available interfacing methods using for an artificial
prosthetic limb can be of five different major types. These
are (1) mechanical (body-powered) interfaces; (2) myolec-
tric (EMG-based) interfaces using non-homologous muscles;
(3) myolectric (EMG-based) interfaces using homologous
muscles; (4) non-invasive neural interfaces; and (5) inva-
sive neural interfaces. Devices of the last two classes are
often referred to in the literature as brain–computer inter-
faces (BCI), but are more properly called brain–machine
interfaces (BMI) (Pino et al. 2009). Body-powered inter-
faces mechanically copy the user’s intentions from the move-
ments of a segment of the body by physically linking them
with the prosthesis. Myoelectric interfaces allow the user
to communicate her/his will through voluntary muscle con-
tractions translated into electrical signals by EMG surface
electrodes. Frequently as a consequence of limb amputation,
the muscles used to move the hand are no longer available so
123
Biol Cybern (2015) 109:141–162
the EMG electrodes are located on non-homologous mus-
cles. Recently, an innovative approach called targeted re-
innervation was developed. This surgical procedure allows
the creation of a homologous myoelectric interface even in
a subject with a very proximal amputation (Kuiken et al.
2007). Non-invasive BMIs are gaining increasing popular-
ity as they do not need to be implanted inside the nervous
system. They record microvolt-level extracellular potentials
generated by neurons in the cortical layers. Several imaging
technologies have been adopted for implementing such inter-
faces, i.e. electroencephalography (EEG), electrocorticogra-
phy (ECoG), magnetoencephalography (MEG), functional
magnetic resonance imaging (fMRI), local field potentials
(LFPs), single-neuron action potential recordings (single
units), and near-infrared spectroscopy, allowing easy access
to experiments on human subjects (Schwartz et al. 2006; Jerbi
et al. 2011). However, only a few attempts were made to con-
trol hand prosthetics using noninvasive EEG interfaces due to
the unpleasant experience of producing a very high amount
of undesired movement. With the use of invasive neural inter-
faces (BMIs or BCIs) an electric direct connection between
the nervous system and the prosthesis could be established
at several levels (Schultz and Kuiken 2011). The electrocor-
ticography (ECoG) and local field potentials (LFPs) are two
common types of measure used to develop invasive types of
interfacing unit. A silicon/neuron interface was introduced
to develop a more authentic interface between the prosthetic
hand and neurons (Pizzi et al. 2007). Application of the intra-
neural electrode interface into the nerves of a limb stump has
succeeded in driving an artificial prosthetic hand and thus
provide independent movement to it by enabling the abil-
ity to carry a bidirectional flow of information between the
bionic limb and patient (Pino et al. 2009). They recognized
the change in the CNS as to be a useful factor for prosthesis
control application to evaluate the effectiveness of the pros-
thetic device and its interfaces. Direct interfaces with the
peripheral or central nervous system may provide the solu-
tion to enhanced sensation from bionic hands and ultimately
come closest to restoring the original sensory perceptions of
the hand (Clement et al. 2011).
The application of artificial intelligence to the interfacing
unit has achieved a wide acceptance among control engi-
neers due to its capacity for decision making and adaptabil-
ity. For instance, the fuzzy methods are particularly capable
of predicting possibilities based on the available data, which
is essentially desired to develop a more promising control
system. Joshi et al. (2011) have effectively applied an Adap-
tive Neuro-Fuzzy Inference System (ANFIS), where a least-
squares method and the back propagation gradient descent
method were employed to control the knee angle of pros-
thesis. The ANFIS proved to be quite successful in mapping
out the high nonlinearity between the trajectories of the knee
joint angle while walking at different speeds. The application
Biol Cybern (2015) 109:141–162 147

of artificial intelligence, in this case, has enabled the control Fig. 2 Example of a
body-powered prosthesis for
system to be adaptive by training it with sets of data captured shoulder disarticulation
from the subject. amputees (Schultz and Kuiken
2011)

4 Available control systems for prosthetics

Depending on the type of basic components used for imple-

menting a control system and the utilities they serve, the con-
trol systems for prosthetics are categorized under different
groups. The modern concept of the prosthesis indicates the
functional prostheses, which are classified into three major
types: (1) body-powered (mechanical or cable operated), (2)
myoelectric, and (3) hybrid. Body-powered prostheses are
largely mechanical devices. To control a body-powered upper
limb prosthesis, amputees use remaining shoulder move-
ments to pull on a cable and sequentially operate prosthetic
where operating the ankle-foot joint was the main focus, and
functions such as the elbow, wrist, and terminal device. Myo-
the other, intact, limb was used as a control. The sagittal
electric prostheses are motorized and are controlled via sur-
position of the functional joint center was determined by the
face electromyogram (EMG) signals from residual muscle
three-dimensional kinematics of the foot investigated. The
sites. Control of myoelectric prostheses is generally achieved
motion of the prosthetic foot tested appeared to be unlike than
by recording from two independent muscles or by differen-
that of an intact ankle. As the constraint for kinematics of a
tiating weak and strong contractions of one muscle (Dutta et
prosthetic ankle is based on an intact ankle, some systematic
al. 2011). Currently, it is a common practice to combine myo-
error pertinent to this would impede replication of the real
electric control and body-powered operation in a hybrid pros-
motion of the prosthetic foot. The interaction of the foot with
thesis, such as a body-powered elbow combined with a myo-
the ground during bipedal walking is unavoidable, which is a
electric terminal device (Schultz and Kuiken 2011). Mechan-
major cause of energy loss. A reduction in this friction would
ical and electromechanical switches, locks, and joints, elec-
minimize an appreciable amount of energy loss during walk-
trical motors, transducers, myoelectric sensors, biosensors,
ing; however, developing such a method is really challenging.
etc., are involved as accessory components to develop a con-
Ghorbani and Wu (2009) established a mechanical control
trol system for prostheses.
system to address that problem by developing the idea of
using the adjustable stiffness elastic elements in an artificial
4.1 Mechanical control system
lower limb’s structure. They believed that a control strategy
adjusting elasticity could significantly reduce the energy loss
The mechanical control system is an arrangement of some
during the collision phase and thus improve the energetics of
mechanical components meant to receive input from the
locomotion in bipedal walking robots. They approached two
amputee and thus follows his/her intention to produce desired
conceptual designs and described them with a mathemat-
movement to the prosthesis. Body-powered prostheses are
ical model. Each conceptual design of adjustable stiffness
mainly mechanical devices and the control system it uses
artificial tendons (ASATs) was applied to the ankle joint of
is called a mechanical control system. To control an artifi-
a bipedal walking robot model. To implement the dynamic
cial prosthesis with this controlling method, amputees use
modelling of the bipedal walking robot during the collision
remaining shoulder movements to pull a cable (Bowden
phase, Lagrangian equations and impulsive constraints were
cable) and sequentially operate prosthetic functions such as
used. They were able to reduce the kinetic energy loss up to
the elbow, wrist, and terminal device. To switch between
20 % using their proposed technique for controlling the limb.
functions, users must lock the joints they wish to remain sta-
However, the level of accuracy of the control system was not
tionary by pressing a switch or using body movements to pull
that promising.
a locking cable (Schultz and Kuiken 2011). A mechanically
controlled prosthesis is shown in Fig. 2.
The mechanical control system in controlling prosthe- 4.2 Electrical control system
ses is used for numerous industrial purposes; however, its
use in controlling the bionic prosthesis is no longer admired The more preferable control system to a control engineer
due to poor accuracy. Rusaw and Ramstrand (2010) applied is the electrical control system, which has been extensively
a mechanical control system to an artificial prosthetic foot used in different aspects of everyday life. Functional electri-

123
148
Kp
Trunk Model
∫ Ki or
Real System
Kd
(a)
Fig. 3 a PID controller with biased output, b LQ regulator for the stabilization around a stationary point (Vanoncini et al. 2012)
cal stimulation is a kind of electrical active control system
proposed by Vanoncini et al. (2012) to stabilize paraplegic
subjects who lack trunk stability due to loss of control. A
closed-loop control system has been developed using a PID
controller (Fig. 3a) and then again with a linear quadratic
regulator (LQR) (Fig. 3b) to perform the process of stabi-
lization.
The control system was accomplished in two stages, i.e.
stabilization of the trunk with a fixed level of stimulation,
and improvement of external disturbance rejection using a
PID controller. The best results were obtained for maximum
stimulation of about 60 % with a value of simulation bias
pwbias (θref ) of 265 µs, which was purely a proportional gain.
To ensure no abrupt muscle contractions, a recovery of exter-
nal perturbation was performed using the P/PD controller and
the LQ regulator. An FES-aided closed-loop controller was
proposed by Previdi et al. (2005) for a patient with a leg
problem controlling standing up and sitting down postures.
Because of the lack of stability of the system and its non-
linear dynamic behavior, performing that task became very
hard. A control strategy based on two nested control loops
has been approached. The inner loop comprised a PID con-
troller, whereas the virtual reference feedback tuning (VRFT)
strategy was applied to design the outer loop. The system
was stabilized by the inner loop. The easy and fast tuning of
the controller using a single I/O dataset was carried out by
the outer loop. Both linear and nonlinear control laws were
implemented exploiting the VRFT algorithms to perform the
operation smoothly. Figure 4 depicts all the successive steps
of the control system.
From the figure, stimulation of muscle has been grouped
into three steps: (a) muscle activation, (b) muscle contraction,
and (c) body- weight relief (WR). The forces generated by
the muscles were the input to the body-WR system dynam-
ics (c). The overall output was the time course of knee joint
angle. However, the FES-based controlling has some inher-
ent time delay due to the time elapsed for stimulating and
activating the muscle. It takes some time for the electrical
current to flow from skin tissue to the muscle. In addition,
123
Biol Cybern (2015) 109:141–162
External disturbances
-K Model
Control Output
variable variable
State variable
(b)
the recruitment property of muscles also causes a time delay
between the electrical stimulation and the actuation of muscle
fibers. Generally, the time delay is about 20 ms. In addition,
in real FES application, maximum force cannot be achieved
because the whole muscle cannot be fully stimulated, as is
the case with natural full contraction. It can be only applied
at a macroscopic level (Zhang et al. 2011).
4.3 Electro-mechanical control system
Electro-mechanical controlling is another upgraded control
system developed by combining electrical and mechanical
systems. This system is used widely to control prosthetics.
From an investigation carried out by Matjačić et al. (2003),
the electro-mechanical system has been observed to be used
to control the posture of a lower limb in a paraplegic patient.
They used a mechanical arrangement called a wobbler with a
functional electrical stimulator to control the ankle moment
and angle during the investigation. A nested, closed-loop con-
trol structure was employed to control the ankle moment,
whereas the inclination angle was regulated by the closed
outer loop control structure of Fig. 5a.
Figure 5a shows the nested loop control strategy, whereas
Fig. 5b depicts the total control system in a block diagram.
In Fig. 5a, θ is the inclination angle, m is the ankle moment,
p is the pulse width of the stimulation, Cm is the moment
controller, and C J is the angle controller. The desired values
for ankle moment and inclination angle are m r e f and θr e f ,
respectively. In Fig. 5b, the blocks denoted “left ankle” and
“right ankle” are closed-loop controllers for the left and right
ankle moments, respectively. Their proposed technique was
very simple, however, not accurate enough at very low and
very high frequencies. They have strongly recommended fol-
lowing a design procedure based on the simple linear model
rather than any heuristic rule such as trial-and-error to design
an analytical control system for tuning a PID controller. On
the other hand, it was not very evident that adaptive control
would be a natural solution within the context of standing
control. Garcia et al. (2011) have designed and developed
Biol Cybern (2015) 109:141–162 149

Stimulation Muscle activation dynamics

pulse width Recruitment
characteristic Muscle
activation
Calcium Muscle fatigue Time
X
dynamics & recovery delay 0 ≤ af ≤ 1
Force-frequency
Stimulation characteristic
frequency

(a)

Muscle Muscle contraction dynamics

activation
Muscle
Joint angle Force-length force
X X
relation
Joint angular
velocity Force-velocity
Max isometric
relation
muscle force

(b)

Body-WR dynamics
Muscle Joint angle
force Active Joint + Total Joint Equations
X moments moments of motion Joint angle
+ + velocity

Moment Joint Joint Seat-pelvis

arms elasticity viscosity interaction

(c)
Fig. 4 Block diagram of the simulation model of a single muscle group: a muscle activation, b muscle contraction, and c body-weight reliever
(WR) dynamics (Previdi et al. 2005)

CNS

muscle moment control Upper

mref,l Left m body
l
ankle

- s desired mref,total reference + Lower

(t) stiffness distributio body
ref (t) ref (t)
(t)
mtotal
s
C C +
Right
(t)
mref,r ankle mr

(a) (b)
Fig. 5 a Nested loop control structure, b Block diagram of the ankle stiffness control and standing strategy (Matjačić et al. 2003)

an electro-mechanical control system for controlling loco- of elasticity and magneto-rheological (MR) dampers pro-
motion of a hybrid leg prototype in a project called HADE vided variable compliance to the actuation system to make the
(Hybrid Actuator Development) for achieving agile locomo- motion of the leg more natural and thus to improved interac-
tion while walking dynamically. A mixed series application tion with the ground. A preliminary direct force controlling

123
150 Biol Cybern (2015) 109:141–162

has been applied to the joints of the HADE leg to control of position control system including a PI controller, an
the locomotion. Both the PID and cascade types of control enhanced PID controller, a robust controller H∞ , and a
systems have been employed to evaluate the performance of sliding-mode controller. Each pneumatic muscle was then
different methods. Then again, an active compliance joint tuned up with an internal pressure loop of a position con-
control has been put in practice to control the joints aimed troller. Results with the classic PI and enhanced PID con-
at controlling the locomotion of the leg. Joint application of trollers has been observed to fluctuate widely at different
elasticity and magneto-rheological damping in series could zones in the displacement range and also with the change
possibly reduce 20 % of the energy lost due to contraction of of mass at the tip. A certain level of robustness with regard
the knee during extension of the leg in its stance phase. to the weight of the load was desired by designing the H∞
An electro-mechanical control scheme based on elec- controller. However, the response obtained was not very com-
tromyographic (EMG) information and with a mechanical pelling and found not to remain the same across the move-
setup has been applied by Jiménez-Fabián and Verlinden ment range. In addition, a sliding-mode technique was imple-
(2012) to control an artificial ankle prosthesis. During the mented to achieve a level of control capable of working cor-
investigation, the regenerative kinetics was the focus of the rectly across the range. Nonetheless, the output demanded
control. A neuromuscular model using EMG signals and a designing or tuning the controller in different operating
standard multi-layer feed-forward neural network with back zones.
propagation algorithms was found to be capable of control- A pneumatic controlling has been applied by Jahanabadi
ling the robotic ankle system to a greater extent. From their et al. (2011) to control a two-link planar arm actuator oper-
study, the continuous response of the proportional control can ated by a pair of pneumatic artificial muscles (PAM). They
correspond to the nervous system more competently than a used active force control (AFC) integrated with fuzzy logic
discontinuous response of the foot switch control. No explicit (AFCAFL) and tested the practicality and feasibility of the
comparison on the quantitative results was obtained from methods applied. For the AFC mechanism, the best value of
their work. the inertia matrix of the pneumatic arm was estimated using
the fuzzy logic (FL) technique, which was complemented by
4.4 Hydraulic/pneumatic power-assisted control system a conventional proportional-integral-derivative (PID) control
at the outermost loop (Fig. 6a).
Recently, different types of power-assisted control systems The extra robustness feature could be achieved by apply-
have been using to develop exoskeletons, orthotics, and pros- ing the AFC part through its capability of rejecting distur-
thetics to strengthen people with limb problems and missing bance. The PID system is usually designed prior to the imple-
limbs. A pneumatic power-assisted lower limb orthosis was mentation of AFC.
developed by Yeh et al. (2010) to help elderly or people with
sport injuries or with difficulty walking or climbing stairs.
The tension of the pneumatic muscle and the angle of the Disturbance
orthosis have been controlled by two different feedback sys- + Q
+ e + τ+
PID
tems designed. The required force by the quadriceps muscle - Controller
PAM System 1/s2
Actual
during the stance phase was reduced by tension control of Desired + position
position Estimated inertial
the pneumatic muscle; the knee angle of orthosis was con- τ-1
matrix IN
trolled using trajectory control during swing motion of leg. + -
The control system was integrated with inverse control (LTR
feedback control) and bumpless switch to augment the per-
formance further. A modified Maxwell-slip model was pro- (a)
posed to eliminate the influence of hysteresis by describing Disturbance
and devising the inverse control method. A backpack con- + Q
+ + +
taining CO2 gas of 4 kg weight was necessary to control the PID PAM System 1/s2
- Controller
pneumatic orthosis, which was stable for only 40 min. The Desired +
position
weight of the gas became a burden to the subject; the weight τ-1 Fuzzy
IN
of the prosthesis and its capacity also have some limitations + -
logic

for the user.

Pujana-Arrese et al. (2010) approached the method in a
different way where they implemented a model of a mecha-
(b)
tronic device with 1-DoF that actuated with two opposite
pneumatic or McKibben muscles in the Dymola/Modelica Fig. 6 a Schematic diagram of AFC, b proposed AFCAFL scheme
environment. They designed and developed a different type (Jahanabadi et al. 2011)

123
Biol Cybern (2015) 109:141–162
The rectangular box shown by the dashed line Fig. 6b
depicts the general form of the AFC scheme used in their
study. The effectiveness and robustness of the pneumatically
actuated (PA) AFCAFL scheme was verified by comparing
the results from the experiment; the simulation works, and
was found to remain in good agreement.
Richardson et al. (2005) implemented force (impedance)
and position (admittance) control strategies on a pneu-
matic robotic limb designed for robotic physiotherapy. The
impedance controller was developed based around a joint
space, which develops a congruity between the manipulator
end-point and external forces as a mass, spring, and damper
system. The controller was found to be working appropri-
ately over a range of conditions. Although there was some
limitation in measuring force along multiple axes properly,
the control strategy performed considerably well, enabling
safe interaction between prosthesis and the human body.
4.5 Artificial intelligence-aided control system
Artificial intelligence-aided control systems give the con-
troller the ability to reason and make decisions based on the
available data. Application of the artificial intelligence-aided
control system has provided the bionic limb the capacity to
operate the prosthesis as though it was a biological limb, giv-
ing similar sensation and functionality to its user. The devel-
opments in biomechatronics are now helping amputees to
lead a life with reduced limitations. Employment of artificial
intelligence in controlling prostheses has increased heavily
and thus enables the amputee to operate the prosthesis more
desirably. Adaptive controlling would enable a system to per-
form closer to the desired output by adjusting the input with
the help of a feedback system. Recently, a mind-controlled
Hidden Layer
Training Set w1i,1
Muscle
Burst
Duration
1
Validation Set w i,2
Muscle
Co-
activation
Testing Set
w1i,j
Fig. 7 Workflow of the proposed artificial neural network (ANN)-based technique (Chang et al. 2009)
151
limb (type of myoelectric controlling) was introduced as the
latest advancement in the artificial intelligence-aided control
system. For instance, a joint project between the Pentagon
and Johns Hopkins Applied Physics Laboratory (APL) has
come up with a modular prosthetic limb which would be
fully controlled by sensors implanted in the brain, and would
even restore the sense of touch by sending electrical impulses
from the limb back to the sensory cortex. Chang et al. (2009)
proposed a multilayer artificial neural network (ANN)-based
model to discover the essential correlation between the intrin-
sic impaired neuromuscular activities of people with spina
bifida (SB) and their extrinsic gait behaviors. The correla-
tion found was then applied to develop a control system for
the prosthesis and is shown in Fig. 7.
Before deploying the ANN-based model, the system was
familiarized with the input neuromuscular activity of the tar-
geted individual to enable it to recognize the inherent charac-
teristics and complex correlation. Producing the desired out-
put within a given error threshold depends on systematically
changing the weights and biases of all neurons; a repeated
tuning of input is required to train the network until the min-
imal error is measured. The results showed that the proposed
ANN-based technique could obtain a more precise predic-
tion at different end-point locomotion. It was claimed that
this technique was capable of providing 80 % more accurate
prediction compared to traditional robust regression. A multi-
layer perceptron and an adaptive-network-based fuzzy infer-
ence system (ANFIS) were proposed by Pérez-Rodríguez et
al. (2012) to solve the inverse kinematics (IK) problem of a
defective upper limb. Once the system was trained properly,
it was able to map 3D positions of the end-effector to the
corresponding healthy biomechanical configurations. Both
for the multilayer perceptron and ANFIS-based methods,
Training
Mode
b 11
Output Layer
1
n 1 b21
w2i,1
b 12 n21
. . .
n12 Gait
. . .
b2k
2
w j,k
b1j n 2k
n 1j
123
152
a high mean correlation coefficient and a low root-mean-
square error were found. The results showed that the inverse
kinematics problem could be solved by both the systems
effectively. In addition t, neuro-rehabilitation (a dysfunc-
tional motion prediction system for anticipatory actuation),
the artificial intelligence-based techniques were observed to
be capable of calculating the IK of a human upper limb that
executes specific activities of the daily life (ADL) accurately
and in real time.
The unique properties of artificial neural networks (ANN)
have allowed Lamb et al. (2011) the investigation of fun-
damental and potentially complex and nonlinear patterns
in data, and thus they have deduced to useful information
before developing a control system for a prosthesis. Self-
organizing maps (SOM), a specific type of ANN applied
in the control system, was found to be an impressive tool
for analyzing the biomechanics. Clustering around the phase
of movement was performed using the SOMs, the trajec-
tory of which showed a change in coordination of different
patterns of movement used. These methods explored a solu-
tion technique for coordination dynamics that deal with the
data obtained from discrete high-dimensional movements.
Recently, application of a microprocessor has been added to
the list of artificial intelligence-based control systems. Wolf
et al. (2009) measured pressure from a residual limb at three
different locations under five different walking conditions,
which were then controlled by adapting ankle angle, modi-
fying joint kinetics, and thus the pressure distribution at the
stump. In ramp-up walking with the highest values of 2.4 and
120 kPa/kg, the particulate matter (PM) sensor on average
was found to show the highest values for both peak pressure
and pressure time integrals (PTIs) at different sensor loca-
tions. Based on this sensor’s measurement, a microprocessor-
based controller was developed and deployed to rehabilitate
the subject amputee.
Thanh and Ahn (2006) developed an adaptive control sys-
tem for a neural network-based, two-axis, pneumatic arti-
ficial muscle manipulator. They employed a nonlinear PID
controller to improve the performance of the controller. The
results of their investigation showed that the combination of
a nonlinear PID controller and neural network could learn,
Fig. 8 Block diagram that Musculotendon actuator
shows the interaction of tendon Natural
and muscle with a mechanical
system (Moreno-Valenzuela and
excitaon
u(t)
Salinas-Avila 2011)
Acvaon
dynamics
123
Biol Cybern (2015) 109:141–162
adapt, and tackle the nonlinearity during the controlling of
the pneumatic muscle manipulator, and it provided better
dynamic properties and strong robustness. For controlling
a prosthesis above the knee (AK), the prediction of pros-
thetic knee joint angle was identified to play a crucial role
in obtaining intended posture and movement of amputees. A
first-order Sugeno-type adaptive neuro-fuzzy inference sys-
tem (ANFIS) was employed by Joshi et al. (2011) to pre-
dict the knee angle from contralateral knee angles at differ-
ent walking speeds. The required torque for knee movement
was estimated using the best regression fit applied on the
gait cycle. From the fuzzy method, the average root-mean-
square error (RMSE) was found to remain at 3.4◦ ± 1.4◦
over a range of walking speeds. An adaptive control system
has been developed by Zhang et al. (2011) using functional
electrical stimulation (FES) of muscles to assist a patient
with pathological tremor. A pair of antagonist muscles was
stimulated by a biologically inspired neural oscillator cre-
ating the anti-tremor rhythmic stimulation pattern. The sur-
face electromyographic (EMG) signal was chosen to main-
tain the neural oscillator reciprocally and to shape adaptively
the simulation pattern. The neural oscillator with a feedback
regulator served as an adaptive feed-forward controller. The
feedback regulator comprised a PID controller, and a com-
pensator was exclusively involved in refining the intensity
pattern of stimulation. About 90 % of the limb’s tremor was
successfully attenuated using the proposed controlling setup.
4.6 Biomechanical control system
The biomechanical control system is another type of con-
trol developed by imitating the behaviors of various bio-
logical organs using different mechanical systems. Moreno-
Valenzuela and Salinas-Avila (2011) addressed the problem
of modeling, control, and simulation of a prosthesis structure
using a biomechanical control system, which was actuated
by an agonist-antagonist musculotendon. A new control law
based on switched system theory was proposed to control
the DC motor engaged to operate the actuator. The control
system that applied for controlling the mechanical system is
shown in Fig. 8.
Musculotendon
a(t) FT(t) Mechanical LMT(t)
contracon
dynamics system
VMT(t)
Biol Cybern (2015) 109:141–162
The different parameters of the mechanical system were
input variables to the musculotendon model. The activation
in the musculotendon model only influenced the variables
belonging to the muscle, i.e. u(t) and a(t). Here a(t) is
the muscle activation, with constraint 0 ≤ a(t) ≤ 1. The
value a(t) = 1 indicates that the muscle is fully activated,
and a(t) = 0 means fully deactivated. u(t) is the neural
excitation, which denotes the control input, with constraint
0 ≤ u(t) ≤ 1.
An open-loop proportional derivative (PD) feedback con-
trol system was employed for switching controller input.
Mass position and velocity of mass were considered the
controlling parameters. When considering the limited con-
trolling parameters, the control system could merely provide
partial feedback and thus limit the efficiency of the controller.
A biomechanical control system for knee flexion-
extension was modeled, developed, and verified by
Mohammed et al. (2012). They have also compared the
results obtained to measure the stability and robustness of
the system. A cascaded control system was employed where
model predictive control (MPC) was applied to the input-
output feedback linearized (IOF) system. This allowed lin-
earization by inverting the system dynamics through different
simulation (in Matlab-Simulink environment) scenarios for
knee flexion-extension. From the comparison between the
results from MPC and pole placement (PP), the MPC was
observed to perform controlling and covering to a desired
position in a relatively minimal time. The required optimiza-
tion time to reach the desired position was recorded to be
around 20 ms, which would encourage real-time implemen-
tation.
4.7 Myoelectric control system
Work is currently underway to address many of the priorities
of myoelectric prosthesis users, yet many challenges remain.
There has been a recent increase in the attention paid to the
development of multifunctional prosthetic hands and wrists
(Takeda et al. 2009). However, there remains a gap in meth-
ods to control simultaneous movements and to decrease the
need for visual attention. Traditional myoelectric control sys-
tems take advantage of the neural information provided by
muscle contractions. The electrical signals generated during
muscle contractions are the result of movement commands
generated in the motor cortex and propagated along periph-
eral nerves. Because myoelectric prostheses use these signals
as control commands, they could be considered simple neural
interfaces. Two major limitations of myoelectric control are
the difficulty in recording suitable EMG signals and the short-
age of information for control of multiple functions (Hofstad
et al. 2009). Many patients are unable to produce isolated
EMG signals or have difficulty making repeatable contrac-
tions. In addition, shifting electrode locations and changing
153
skin conditions (such as sweat) alter EMG signals and can
cause control to be unreliable (Schultz and Kuiken 2011).
A major problem in multifunctional prosthesis control via
electromyographic (EMG) signals is filtering the EMG sig-
nals from noise and from other interacting EMG signals,
and of recognizing and separating the EMG signals related
to different limb functions (Graupe and Cline 1975; Gagné
et al. 2009). The recent development in the electromyo-
graphic control systems focus on signal processing and fea-
ture extraction methods.
Based on the signal processing methods, myoelectric con-
trol systems are divided into two groups: pattern recognition
and non-pattern recognition (Asghari Oskoei and Hu 2007).
The state-of-the-art prostheses employ pattern recognition
techniques to extract the neuromuscular control informa-
tion from the surface EMG (Jiang et al. 2009). The pattern
recognition-based control schemes increase the variety of
control functions and improve robustness.
Pattern recognition of EMG signals relies on the gener-
ation of differentiable and repeatable contractions in EMG
signal. Changes in these patterns can erode the performance
of the classifier and may result in an unusable controller
(Fougner et al. 2011). An effective pattern classification can
secure a better performance of the myoelectric control sys-
tem, which typically comprises segmentation, feature extrac-
tion, and classification units (Doerschuk et al. 1983). The
success of any pattern recognition problem depends almost
entirely on the selection and extraction of features. A wide
spectrum of features is reported in the literature for myoelec-
tric classification. All of the features generally fall into any
of the following three categories: time domain, frequency
(spectral) domain, and time-scale (time-frequency) domain
(Zecca et al. 2002). Amplitude and its related features are
often investigated in time domain analysis, while a power
spectrum is usually studied in a frequency domain analysis.
Unlike the time domain and frequency domain features, the
time-scale (time-frequency) domain features are capable of
locating and measuring the intensity of an event simultane-
ously. Figure 9, shows signal processing methods used to
analyze EMG signal.
Using a pattern recognition technique to discriminate mul-
tiple degrees of freedom has shown great promise in the
research, but it has yet to transform to a clinically viable
option (Scheme and Englehart 2011). Figure 10 represents a
schematic diagram of a pattern recognition-based myoelec-
tric control system.
One of the problems is that pattern recognition of the
EMG does not provide simultaneous and proportional con-
trol of multiple functions, but only provides sequential and
on/off activation. Simultaneous and proportional myoelec-
tric control over multiple degrees of freedom (DoFs) can
be achieved with alternative approaches such as synergis-
tic structure-based muscle activation (Karlik et al. 2003;
123
154 Biol Cybern (2015) 109:141–162

Signal processing method

Pattern recognition based Pattern non-recognition based

Classification Feature extraction Classificatio Feature

Classifier: Classifier: Methods:

Multilayer Time domain Frequency domain Time-frequency dom ain Linear mapping, Fourier
perceptron nonlinear mapping, transform (FT)
Cascaded genetic [Ref: Asghari
(MLP), Linear
Methods: Methods: Methods: algorithm with Oskoei and Hu
perceptron (LP), Auto regressive method (AR), Fractural Davies–Bouldin
Fourier transform Wavelet transform ( WT) 2007]
Linier analysis (FA), Bayesian analysis, Deflection (FT), index, Principal Features:
, Stationary wavelet
discriminant measurement [Ref: Asghari Oskoei and Hu Autoregressive component analysis Power spectral
transform (SWT),
analysis (LDA), 2007] model (AR) [Ref: (PCA), Linear
Continuous Wavelet density (PSD),
Gaussian mixture Features: Asghari Oskoei discriminant
Root mean square (RMS), Autoregressive transform (CWT), S hort analysis (LDA), [Ref: Hargrove
model (GMM), and Hu 2007]
coefficient (ARCo), standard deviation (STD), Features: time Fourier transfo rm linear– et al. 2007;
Hidden Markov Asghari Oskoei
Mean absolute value (MAV), Power spectral (STFT) or Gabor nonlinear projection
model (HMM), Mean absolute value method composed of and Hu 2007]
density (PSD), tranform , Wavelet
Gaussian mixture slope (MAVS), zero crossing (ZC), Slope sign PCA and a self-
Linear cepstrum packet transform (W PT)
model (GMM) , changes (SSC), Waveform length (WL), organized feature
coefficients [Ref: Hargrove et al.
Detrended Integrated absolute value map (SOFM), [Ref:
(IAV), Difference absolute mean value (LCC), Adaptive 2007; Asghari Oskoeei Asghari Oskoei and
fluctuation
(DAMV), Variance cepstrum vector and Hu 2007] Hu 2007].
analysis (DFA)
(VAR), Maximum fractal length (MFL), (ACV) Features:
[Ref: Hargrove et Higuchi’s fractal dimension (HFD) [Ref: [Ref: Hargrove et Moving average, Me an
al. 2007; Asghari Hargrove et al. 2007; Ameri et al. 2014;
al. 2007; Asghari square error of spect ral,
Oskoei and Hu Asghari Oskoei and Hu 2007; Park and Lee
Oskoei and Hu Aspect ratio,
2007; 1998; Hudgins et al. 1993; Clancy et al. 2002;
2007; Park and Partitioning ratio [Reef:
Phinyomark et al. Farina and Merletti 2000; Phinyomark et al.
2012; Naik et al. 2008; Oskoei & Hu 2008; Lee, 1998] Asghari Oskoei and Hu
2012]
Parker et al., 2006; Zardoshti-Kermani et al. 2007]
1995; Zecca et al. 2002; Arjunan 2008;
Arjunan & Kumar 2010; Truong Quang Dang
et al. 2012]

Fig. 9 Available signal processing methods for EMG signal analysis

Fig. 10 A myoelectric control

system based on pattern
recognition (Asghari Oskoei and
Hu 2007)

Jiang et al. 2013, 2014). Fougner et al. (2014) have devel- ous proportional control was compared with three other con-
oped a system for simultaneous proportional myoelectric trol strategies implemented on the same prosthesis: mutex
control of a hand prosthesis with two motorized functions proportional control (the same system, but with simulta-
(hand open/close, and wrist pro-/supination). The simultane- neous control disabled), mutex on-off control, and a more

123
Biol Cybern (2015) 109:141–162
traditional, sequential proportional control system with co-
contractions for state switching. They found that the simulta-
neous proportional control strategy and the two mutex-based
pattern recognition strategies performed equally well and
better compared to the more traditional sequential strategy.
Khushaba et al. (2012) captured electromyogram (EMG)
signals from various muscles of an individual to control the
movements of a prosthesis. They used a Bayesian data fusion
approach to enhance the possibility of correct classification
of the EMG data pertinent to different movements. Using the
surface EMG signal and the Bayesian data fusion approach,
individual and combined movement of fingers can be differ-
entiated accurately. The feasibility of the method was tested
from the practical and statistical results, where approximately
90 % accuracy on average was estimated in classifying the
signal from different subjects. Kao and Ferris (2009) have
applied the myoelectric control system to control an exter-
nally powered ankle-foot orthosis (AFO). They employed
the AFO to improve walking ability and prevent tripping of
a patient with very weak dorsiflexion. Because of the quick
adaptability of humans to the plantar flexion, a proportional
myoelectric system was used to control the orthosis. The
results have shown that unlike the rigid AFO, a powered
AFO with dorsiflexion did not impede planter flexion at the
end of stance, and it allowed users to alter their gait to accom-
modate changes in the environment within one or two day(s)
training only.
A new electromyography (EMG)-based control system for
an above-knee (AK) prosthesis was proposed by Wu et al.
(2011) to enable the users to control the prosthesis motion
directly with their muscles activating neural signals. They
obtained the muscle-activating signals by applying electrodes
to the surface of the user’s remaining muscles, which essen-
tially provided a non-intrusive interface to the human CNS.
The active-reactive control algorithm was developed based
on the analysis of a simplified musculoskeletal structure of
the human biological joint where the EMG signal was the
input to the system. The exclusive structure of the controller
provides an active control to the joint motion, where the mod-
ulation of the joint impedance enhanced the system with the
ability of natural interaction with environment. The user was
given an experience similar to a biological lower limb opera-
tion with a unique “active–reactive” control structure, which
combined the “active” actuation of the knee joint and the
“reactive” response to the joint motion. The effectiveness of
this control structure was tested by implementing the pro-
posed control algorithm on an AK prosthesis with powered
knee joint during a set of free-swing and level walking exper-
iments. Experimental results indicated that the controller was
able to generate the desired motion according to the user’s
neural command, and walking gait with the proposed con-
troller was natural and similar to the standard gait of healthy
subjects.
155
Recently, a prosthesis has been developed by Catalan
(2012) and his team at Chalmers University of Technology in
Sweden where intra-muscular EMG signals were collected
by implanted electrodes directly to the bones and nerves of
amputees. Effort was put into developing a control system
capable of regulating multiple joints and movements at a
time, and which could receive corresponding neural feedback
from the respective electronic limb. An ordinary myoelectric
prosthesis is operated by nerve signals picked up from the
skin using the electrode, which are generally produced in the
brain and sent to the limb. The signals are then translated
with an algorithm before sending instructions to motors of
the prosthesis. It was expected that the invasive technique
would provide the amputees more control on their prosthe-
sis, in much the same way as a biological limb of their own
using the person’s nerves and remaining muscles. According
to Catalan, the closed-loop control upgraded the control sys-
tem one step further to obtain natural control of the prosthe-
sis. As each finger was motorized, they could be individually
controlled; this improvement made the electronic limb sys-
tem more like a biological limb. Cipriani et al. (2014) have
tested the feasibility of achieving a real-time, continuous, and
simultaneous controlling of a multi-digit prosthesis by cap-
turing intramuscular EMG signals from a subject’s forearm
muscle. Besides allowing the subjects to achieve continuous
simultaneous and independent control over multiple DoFs,
this technique enables the subject to perform potentially any
arbitrary posture in a physiologically appropriate manner.
Though several events of successful control of a prosthe-
sis are observed using intra-muscular electrodes, no signif-
icant difference is noticed when compared with the classi-
fication accuracy of the surface myoelectric control system.
Both measurement techniques yielded classification accura-
cies between 95 and 99 % (Hargrove et al. 2007). Even though
there is noise due to crosstalk and fatty muscle (which causes
impedance during transmission of the myoelectric signals
from the main nerves to the surface), this could be largely
reduced by using the intra-muscular electrodes; there is a high
possibility of missing more global information contained in
the surface myoelectric signal for pattern recognition-based
myoelectric control.
4.8 Bionic control system
The use of a bionic control system for prosthetics is pro-
gressing aimed at providing realistic control that is more like
a sound biological limb to the users. It is typically performed
using nerve signals, and sometimes stimulating the nervous
system for individuals. For animals and robots, this has been
performed particularly by inspiring the central pattern gen-
erator (CPG) (Ijspeert 2008). Fasano et al. (2012) have clini-
cally treated a patient with lower-limb disability by stimulat-
ing the patient’s brain with medication. Recently, thalamic
123
156
deep-brain stimulation (DBS) on ataxic features has been
carried out in patients with essential tremor (ET), a slowly
progressive movement disorder, during different locomotion
tasks. Thalamic DBS was found to reduce the severity of
tremors significantly (66.1 % ± 15.7 %; p < 0.001), while
intention and postural tremor were observed to improve.
Another bionic control system prior to investigating the EMG
signals captured from different location of muscles of a limb
was developed by Daley et al. (2012). They successfully
classified the movement and activated the distinctive mus-
cle pattern of multiple limbs. The high density of EMG sig-
nals was identified to be useful in locating the most effec-
tive placement of electrodes to classify multiple wrist and
hand movements reliably in transradial amputees. The result
showed that the performance improved radically when the
electrodes were optimally placed rather than increasing the
electrode number or distributing them evenly. Despite the
impressive advancement, bionic prostheses still remain as
an inferior replacement to their biological counterparts. Cur-
rently available bionic hands are able to achieve a better range
of functional grasps including key grip, power grip, precision
grip, index point, and thumb park that confer far wider appli-
cation of the replacements than previously achievable. The
drawback is that extensive training with occupational therapy
input is required to achieve successful function, and despite
this, even the best bionic limbs still do not come close to repli-
cating the complex intrinsic functions of the innate hand.
Clement et al. (2011) have studied the different ways of
replicating the major functions of the human hand in a mod-
ern artificial bionic prosthesis. According to them, as the
nerves that once supplied the amputated limb muscle are sur-
gically anastomosed into the remaining muscles of the ampu-
tation stump, the re-innervation of nervous system of the
remaining muscles could enable the amputee to control the
attached prosthesis independently. In the amputated nerves,
motor command is amplified with the re-innervated muscles;
therefore, the surface electromyogram (EMG) signals could
provide some extra advantage controlling the prosthetic arm.
This technique has shown promising results with the ability
to achieve intuitive control of multiple functions in a bionic
hand.
The ultimate goal is to achieve a “biomechatronic design”
where the mechatronic system of the artificial hand is inspired
and works like the living limb. To achieve this goal, integra-
tion of the prostheses with the CNS is necessary so that the
replacement moves and is perceived as if it were the natural
hand without the requirement for any training or adaptation.
The use of intraneural electrodes is perhaps the most promis-
ing technology that may hold the key to successful integration
of bionic limbs into the biological system. Intraneural elec-
trodes interface directly into the nerves in the limb stump
and have the ability to carry a bidirectional flow of infor-
mation between the bionic limb and patient. There are still
123
Biol Cybern (2015) 109:141–162
many obstacles to overcome before this technology is com-
mercially available, but it is hoped that it will eventually be
advanced enough to allow limb amputees to be given bionic
limbs that act and feel like the innate limb.
4.9 Miscellaneous control systems
Besides the available control systems discussed, there are
some other methods of control for particular application. For
example, Engeberg (2013) developed a control system for
upper limbs that was capable of functioning like a living
limb intuitively. Using a biomimetic sliding mode (BSM)
controller, human muscle signals were mapped into prosthe-
sis motor command signals in a physically expected manner.
The forces were tracked by both controllers to assess the
controllability of the applied force. A nonparametric Mann–
Whitney U test results showed that in terms of level of con-
fidence ( p < 0.01), the BSM controller was about 99 %
different from the PD controller. In cases of position con-
trol ( p < 0.05), the BSM controller was also found to be
statistically different from the PD controller. To imitate the
posture of human hands more accurately, the performance of
the BSM controller was found to supersede that of the PD
force controller. A subjective rating showed that the controls
of the BSM were quite similar to the human hand in terms
of position and force.
A novel control and feedback system based on peripheral
intraneural multielectrodes was proposed by Rossini et al.
(2010) for controlling the multi-movement of a prosthesis.
The LIFE (nerve) signal was recorded from the median and
ulnar nerves from some volunteers by implanting four lon-
gitudinal intrafascicular flexible multi-electrodes (tf-LIFE4)
during three distinct hand movements. An off-line artificial
intelligence with >85 % success rate in classifying LIFE sig-
nal was employed to obtain a real-time control over the motor
for all three actions. Some other various types of the current
stimulation method were identified to provide reproducible
and localized hand/fingers sensations. van der Smagt (2000)
adapted a cerebellar model for trajectory tracking, compli-
ance control, and exteroceptive feedback control of an pros-
thesis. A servo control system made up of a proportional
derivative (PD) controller(s) has been applied to control inde-
pendently every joint. According to them, there still exists an
enormous gap between the applicability of cerebellar mod-
els to prosthesis dynamics control and their current applica-
tions. Nevertheless, the emerging interest in cerebellar mod-
els demonstrates huge possibilities.
A novel technique has been applied by Li and Kuiken
(2008) to upgrade the control of rotation of prosthetic arms. A
magnetic field along the arm rotation was created by implant-
ing a small permanent magnet into the distal end of the resid-
ual bone, which was used as a sensor in the system. The
magnetic-field distribution on the arm’s surface helped to
Biol Cybern (2015) 109:141–162
Table 1 A comparison among different control systems
Type of control system
Mechanical control system
Electrical control system
Electro-
mechanical
control system
Hydraulic/pneumatic
control system
Artificial
intelligence
aided control
system
Biomechanical control system
Myoelectric control system
Bionic control system
Significant characteristics
An appreciable residual limb is
required to be used for
controlling the prosthesis
Efficiency of the control system
influenced by the material
properties used for operating
the artificial limb
Controlling is performed by
stimulating the respective
biological controlling organ
Preferable for the patient with
difficulty in operating the
limb
Not a commonly used
controlling technique for
prosthesis
Both electrical and mechanical
sensors are used to assess the
change within the system
The electrical and mechanical
elements are used together to
control the prosthesis
Usually performed based on the
force controlling unlike to the
electrical and
electromechanical methods,
which are used position
controlling technique
Inherent correlations between
intrinsic impaired
neuromuscular activities are
identified using artificial
neural network
Inverse kinematics is
controlled by training up the
artificial neural network with
variety of input data
Performed by switching the
controller feedback by
actuating the musculotendon
Neural information from
muscle contraction, i.e. EMG
signal is used for controlling
EMG signal is usually captured
from the residual limb
Required substantial analysis of
signal to extract the important
information out
Performed by stimulating the
brain signal
EMG signal and invasive
electrode implantation are
used to innervate muscle and
enhance the control system
Advantages
Cheap and affordable for average
classes of people
Robust in design
Highly regarded in treating the
stroke patient
Preferable for controlling
locomotion of limb
Heavily devoted for the people
with sports injury
Used to assist people moving in
hilly place
Useful in military application
About 80 % more effective
prediction than any other
techniques can be made
Able to successfully control the
tremor and inverse kinematics
Can successfully control the
stability of a patient with
difficulty in knee-flexion
extension
Preferable because of its
non-invasive characteristics
Capable of significantly improve
the tremor problem
157
Disadvantages
Unable to reproduce the real
motion of a biological
prosthesis
Poor accuracy (comparatively)
Damages the muscle in the
long run
Obtained some minimal delay
in transferring electrical
signals from skin to the
muscle
Maximum force of prostheses
cannot be achieved due to its
limitation to stimulate the
whole limb
Not accurate enough
Limited to some particular
types and frequency of
feedback signal
The fluid reservoir is
considered to be a burden to
its user
Required large database and
numerous trials to train the
controller before application
Limited application due to its
only use of mass position and
mass velocity
Inadequate efficiency
Inefficient for the patient with
poor/weak EMG signal
production capacity
Extensive training with
occupational therapy input is
required
Intraneural electrode causes
infection to the patient body
123
158
track the bone rotation. The rotation of the prosthesis was
then manipulated based on the signals from the magnetic
field. During the whole process, the proprioception of the
residual limb skeleton remained intact. It has been observed
from a simulation result of an upper arm model that by decod-
ing the magnetic-field signal, the direction and angle of rota-
tion of the residual humerus could be obtained.
Santos and Matos (2012) have developed a control sys-
tem for a multi-articulated artificial robotic limb by regulat-
ing the walking velocity, orientation, and angular velocity to
provide it the desired movement. Their focus was to develop
a quadruped locomotion controller able to generate omnidi-
rectional locomotion and path in the heading direction. The
modification of control parameters in the heading direction
controller was performed by adapting the visual informa-
tion from the sensory-motor feedback to the limb trajectory
online. Although the proposed control system was famous
for its ability to integrate the sensory-motor feedback and
closed-loop controlling, it has gained little attention so far.
5 Conclusions
The control system varies depending on the nature of the
system to be controlled and the extent to which the control
is to be provided for the user.
• The mechanical control system is considerably robust;
however, it is not sensitive enough to the changes in the
system. The application of this method is preferable for
the amputees who, with a substantial amount of residual
limb, are capable of producing the required driving force
for the prosthesis.
• The electrical system is essentially prompt and highly
regarded for the treatment of stroke patients and the patient
with paraplegia. The major drawback is that it causes dam-
age to muscles in the long run.
• The hybrid system includes an electro-mechanical sys-
tem, which performs better in controlling the locomotion
problem; however, only within a limited range of feedback
frequency.
• The power-assisted control systems such as hydraulic and
pneumatic controllers ensure smooth movement and pro-
vide good adjustability. These types of control systems
are preferable for treating the patient with a sports injury
or difficulty in hiking. But the most deterring backlash
is the weight of the fluid reservoir that one has to carry
always in order to operate the system, and that is certainly
a burden.
• The artificial intelligence aided control system is capa-
ble of treating patients with tremor and inverse kinemat-
ics problems successfully. This technique could develop
more intuitive control systems with impressive adaptabil-
123
Biol Cybern (2015) 109:141–162
ity, but it requires a large dataset and a lot of trials to get
good controlling done.
• Although biomechanical controlling is preferred to con-
trol knee flexion stability during extension, due to its strict
dealing with mass position and velocity of mass, it is not
always efficient.
• The application of electromyography capturing the ner-
vous signals from the brain, remaining muscle, and skin
of different body parts are being groomed due to the out-
standing capability of sensing the nerve’s signals. Nonin-
vasive myoelectric controlling is not favored by the patient
with poor EMG signal production capacity.
• The bionic controlling approach with invasive electrode
implantation into the human nervous system is still a
new method under development and has achieved some
success in improving the tremor problem. This method
demands extensive training with occupational therapy
input, and in some cases the electrode insertion causes
infection to the patient.
• A significant improvement is still required to overcome
the existing setbacks and develop a reliable and com-
petent control system for the patient with prosthetic
disability.
The salient features of the various controlling systems have
been summarized and presented in tabular form (Table 1).
Acknowledgments The authors would like to thank the University of
Malaya for providing the funds, Grant No: UM.C/HIR/MOHE/ENG/28
(D000028-16001) to carry out this work.
References
Akagi T, Dohta S, Kenmotsu Y, Zhao F, Yoneda M (2012) Development
of smart inner diameter sensor for position control of Mckibben
artificial muscle. Procedia Eng 41:105–112. doi:10.1016/j.proeng.
2012.07.149
Alaqtash M, Yu H, Brower R, Abdelgawad A, Sarkodie-Gyan T (2011)
Application of wearable sensors for human gait analysis using
fuzzy computational algorithm. Eng Appl Artif Intell 24(6):1018–
1025. doi:10.1016/j.engappai.2011.04.010
Albertini JN, Barral X, Branchereau A, Favre JP, Guidicelli H, Magne
JL, Magnan PE (2000) Long-term results of arterial allograft
below-knee bypass grafts for limb salvage: a retrospective mul-
ticenter study. J Vasc Surg 31(3):426–435. doi:10.1016/s0741-
5214(00)90301-x
Aliff M, Dohta S, Akagi T, Li H (2012) Development of a simple-
structured pneumatic robot arm and its control using low-cost
embedded controller. Procedia Eng 41:134–142. doi:10.1016/j.
proeng.2012.07.153
Ameri A, Scheme EJ, Kamavuako EN, Englehart KB, Parker PA
(2014) Real-time, simultaneous myoelectric control using force
and position-based training paradigms. IEEE Trans Biomed Eng
61(2):279–287. doi:10.1109/tbme.2013.2281595
Anam K, Al-Jumaily AA (2012) Active exoskeleton control systems:
state of the art. Procedia Eng 41:988–994. doi:10.1016/j.proeng.
2012.07.273
Biol Cybern (2015) 109:141–162
Arjunan SP, Kumar DK (2008) Fractal features based technique to iden-
tify subtle forearm movements and to measure alertness using
physiological signals (sEMG, EEG). In: TENCON 2008–2008
IEEE Region 10 Conference, 19–21 Nov 2008. pp 1–4. doi:10.
1109/TENCON.2008.4766797
Arjunan P, Kumar D (2010) Decoding subtle forearm flexions using
fractal features of surface electromyogram from single and mul-
tiple sensors. J NeuroEng Rehabil 7(1):1–10. doi:10.1186/1743-
0003-7-53
Asghari Oskoei M, Hu H (2007) Myoelectric control systems—a survey.
Biomed Signal Process Control 2(4):275–294. doi:10.1016/j.bspc.
2007.07.009
Biddiss E, Chau T (2008) Dielectric elastomers as actuators for upper
limb prosthetics: challenges and opportunities. Med Eng Phys
30(4):403–418. doi:10.1016/j.medengphy.2007.05.01133
Bryant PR, Pandian G (2001) Acquired limb deficiencies. 1. Acquired
Limb deficiencies in children and young adults. Arch Phys
Med Rehabil 82(3, Supplement 1):S3–S8. doi:10.1016/s0003-
9993(01)80030-7
Burger H, Marinček Č, Jaeger RJ (2004) Prosthetic device provision
to landmine survivors in bosnia and herzegovina: outcomes in 3
ethnic groups. Arch Phys Med Rehabil 85(1):19–28. doi:10.1016/
j.apmr.2003.07.010
Carignan C, Tang J, Roderick S (2009) Development of an exoskeleton
haptic interface for virtual task training. Paper presented at the
proceedings of the 2009 IEEE/RSJ Int Conf Intell Rob Syst. St.
Louis, MO, USA
Catalan MO (2012) First mind-controlled, bone-mounted robotic arms
to be implanted in 2013. http://www.wired.co.uk/news/archive/
2012-11/28/robotic-arm-transplant-operation
Chang C-L, Jin Z, Chang H-C, Cheng AC (2009) From neuromuscular
activation to end-point locomotion: an artificial neural network-
based technique for neural prostheses. J Biomech 42(8):982–988.
doi:10.1016/j.jbiomech.2009.03.030
Chang M-K (2010) An adaptive self-organizing fuzzy sliding mode
controller for a 2-DOF rehabilitation robot actuated by pneumatic
muscle actuators. Control Eng Pract 18(1):13–22. doi:10.1016/j.
conengprac.2009.08.005
Chang M-K, Liou J-J, Chen M-L (2011) T–S fuzzy model-based track-
ing control of a one-dimensional manipulator actuated by pneu-
matic artificial muscles. Control Eng Pract 19(12):1442–1449.
doi:10.1016/j.conengprac.2011.08.002
Cipriani C, Segil JL, Birdwell JA, Weir RF (2014) Dexterous control
of a prosthetic hand using fine-wire intramuscular electrodes in
targeted extrinsic muscles. IEEE Trans Neural Syst Rehabil Eng
99:1–1. doi:10.1109/TNSRE.2014.2301234
Clancy EA, Morin EL, Merletti R (2002) Sampling, noise-reduction
and amplitude estimation issues in surface electromyogra-
phy. J Electromyogr Kinesiol 12(1):1–16. doi:10.1016/S1050-
6411(01)00033-5
Clement RGE, Bugler KE, Oliver CW (2011) Bionic prosthetic hands:
a review of present technology and future aspirations. Surgeon
9(6):336–340. doi:10.1016/j.surge.2011.06.001
Couture M, Desrosiers J, Caron CD (2011) Cognitive appraisal and
perceived benefits of dysvascular lower limb amputation: a lon-
gitudinal study. Arch Gerontol Geriatr 52(1):5–11. doi:10.1016/j.
archger.2009.11.002
Crevecoeur F, Giard T, Thonnard JL, Lefevre P (2011) Adaptive control
of grip force to compensate for static and dynamic torques during
object manipulation. J Neurophysiol 106(6):2973–2981. doi:10.
1152/jn.00367.2011
Cura VO, Cunha FL, Aguiar ML, Cliquet A Jr (2003) Study of the differ-
ent types of actuators and mechanisms for upper limb prostheses.
Artif Organs 27(6):507–516
Daley H, Englehart K, Hargrove L, Kuruganti U (2012) High den-
sity electromyography data of normally limbed and transradial
159
amputee subjects for multifunction prosthetic control. J Elec-
tromyogr Kinesiol 22(3):478–484. doi:10.1016/j.jelekin.2011.12.
012
Dang KTQ, Le Minh H, Van Vo T (2012) Analyzing surface EMG
signals to determine relationship between jaw imbalance and arm
strength loss. BioMed Eng OnLine 11(1):1–14. doi:10.1186/1475-
925X-11-55
del-Ama AJ, Koutsou AD, Moreno JC, de-los-Reyes A, Gil-Agudo
A, Pons JL (2012) Review of hybrid exoskeletons to restore gait
following spinal cord injury. J Rehabil Res Dev 49(4):497–514
Doerschuk PC, Gustafon DE, Willsky AS (1983) Upper extremity
limb function discrimination using EMG signal analysis. IEEE
Trans BME Biomed Eng 30(1):18–29. doi:10.1109/TBME.1983.
325162
Dutta A, Kobetic R, Triolo RJ (2011) An objective method for selecting
command sources for myoelectrically triggered lower-limb neuro-
prostheses. J Rehabil Res Dev 48(8):935–948
Engeberg ED (2013) A physiological basis for control of a prosthetic
hand. Biomed Signal Process Control 8(1):6–15. doi:10.1016/j.
bspc.2012.06.003
Esposito IJ, Beard DJ, Dodd CAF, Shafighian B (1997) Rehabilitation
following patellar tendon or ABC prosthetic ligament reconstruc-
tion for chronic anterior cruciate ligament deficient knees. Knee
4(2):81–86. doi:10.1016/s0968-0160(96)00242-6
Esquenazi A, Meier Iii RH (1996) Rehabilitation in limb defi-
ciency. 4. Limb amputation. Arch Phys Med Rehabil 77(3,
Supplement):S18–S28. doi:10.1016/s0003-9993(96)90239-7
Fan L-L, Song Y-D (2011) Fault-tolerant control and disturbance atten-
uation of a class of nonlinear systems with actuator and compo-
nent failures. Acta Autom Sin 37(5):623–628. doi:10.1016/S1874-
1029(11)60206-1
Farina D, Merletti R (2000) Comparison of algorithms for estimation
of EMG variables during voluntary isometric contractions. J Elec-
tromyogr Kinesiol 10(5):337–349
Fasano A, Herzog J, Raethjen J, Rose FEM, Volkmann J, Falk D,
Deuschl G (2012) Lower limb joints kinematics in essential tremor
and the effect of thalamic stimulation. Gait Posture 36(2):187–193.
doi:10.1016/j.gaitpost.2012.02.013
Fougner A, Scheme E, Chan AD, Englehart K, Stavdahl O (2011)
Resolving the limb position effect in myoelectric pattern recogni-
tion. IEEE Trans Neural Syst Rehabil Eng 19(6):644–651. doi:10.
1109/tnsre.2011.216352935
Fougner A, Stavdahl O, Kyberd P (2014) System training and assess-
ment in simultaneous proportional myoelectric prosthesis control.
J NeuroEng Rehabil 11(1):75
Gagné M, Reilly KT, Hétu S, Mercier C (2009) Motor control over the
phantom limb in above-elbow amputees and its relationship with
phantom limb pain. Neuroscience 162(1):78–86. doi:10.1016/j.
neuroscience.2009.04.061
Gailey R, Allen K, Castles J, Kucharik J, Roeder M (2008) Review of
secondary physical conditions associated with lower-limb ampu-
tation and long-term prosthesis use. J Rehabil Res Dev 45(1):
15–29
Gallagher P, MacLachlan M (2004) The Trinity amputation and prosthe-
sis experience scales and quality of life in people with lower-limb
amputation. Arch Phys Med Rehabil 85(5):730–736. doi:10.1016/
j.apmr.2003.07.009
Gao K, Chen S, Wang L, Zhang W, Kang Y, Dong Q, Zhou H, Li
L (2010) Anterior cruciate ligament reconstruction with LARS
artificial ligament: a multicenter study with 3- to 5-year follow-
up. Arthrosc J Arthrosc Amp Relat Surg 26(4):515–523. doi:10.
1016/j.arthro.2010.02.001
Garcia E, Arevalo JC, Muñoz G, Gonzalez-de-Santos P (2011) Combin-
ing series elastic actuation and magneto-rheological damping for
the control of agile locomotion. Rob Autom Syst 59(10):827–839.
doi:10.1016/j.robot.2011.06.006
123
160
Ghorbani R, Wu Q (2009) Adjustable stiffness artificial ten-
dons: conceptual design and energetics study in bipedal walk-
ing robots. Mech Mach Theory 44(1):140–161. doi:10.1016/j.
mechmachtheory.2008.02.009
Graupe D, Cline WK (1975) Functional separation of EMG signals
via ARMA identification methods for prosthesis control purposes.
IEEE Trans Syst Man Cybern 5(2):252–259. doi:10.1109/TSMC.
1975.5408479
Hargrove LJ, Englehart K, Hudgins B (2007) A comparison of
surface and intramuscular myoelectric signal classification. IEEE
Trans Biomed Eng 54(5):847–853. doi:10.1109/TBME.2006.
889192
Hemami H, Dariush B (2012) Central mechanisms for force and
motion—towards computational synthesis of human movement.
Neural Netw 36:167–178. doi:10.1016/j.neunet.2012.09.008
Hofstad CJ, Weerdesteyn V, van der Linde H, Nienhuis B, Geurts AC,
Duysens J (2009) Evidence for bilaterally delayed and decreased
obstacle avoidance responses while walking with a lower limb
prosthesis. Clin Neurophysiol 120(5):1009–1015. doi:10.1016/j.
clinph.2009.03.003
Hudgins B, Parker P, Scott RN (1993) A new strategy for multifunction
myoelectric control. IEEE Trans Biomed Eng 40(1):82–94. doi:10.
1109/10.20477436
Ijspeert AJ (2008) Central pattern generators for locomotion control in
animals and robots: a review. Neural Netw 21(4):642–653. doi:10.
1016/j.neunet.2008.03.014
Jae-Myung Y, Yong-Myung A (2006) A study on a sensing system for
artificial arm’s control. In: International joint conference on SICE-
ICASE, 2006. 18–21 Oct 2006. pp 3083–3088. doi:10.1109/sice.
2006.314717
Jahanabadi H, Mailah M, Zain MZM, Hooi HM (2011) Active force
with fuzzy logic control of a two-link arm driven by pneumatic
artificial muscles. J Bionic Eng 8(4):474–484. doi:10.1016/S1672-
6529(11)60053-X
Jawhar I, Mohamed N, Agrawal DP (2011) Linear wireless sensor
networks: classification and applications. J Netw Comput Appl
34(5):1671–1682. doi:10.1016/j.jnca.2011.05.006
Jerbi K, Vidal JR, Mattout J, Maby E, Lecaignard F, Ossandon T,
Hamamé CM, Dalal SS, Bouet R, Lachaux JP, Leahy RM, Baillet S,
Garnero L, Delpuech C, Bertrand O (2011) Inferring hand move-
ment kinematics from MEG, EEG and intracranial EEG: from
brain-machine interfaces to motor rehabilitation. IRBM 32(1):8–
18. doi:10.1016/j.irbm.2010.12.004
Jiang N, Englehart KB, Parker PA (2009) Extracting simultaneous and
proportional neural control information for multiple-DOF prosthe-
ses from the surface electromyographic signal. IEEE Trans Biomed
Eng 56(4):1070–1080. doi:10.1109/tbme.2008.2007967
Jiang N, Muceli S, Graimann B, Farina D (2013) Effect of arm position
on the prediction of kinematics from EMG in amputees. Med Biol
Eng Comput 51(1–2):143–151. doi:10.1007/s11517-012-0979-4
Jiang N, Rehbaum H, Vujaklija I, Graimann B, Farina D (2014) Intu-
itive, online, simultaneous, and proportional myoelectric control
over two degrees-of-freedom in upper limb amputees. IEEE Trans
Neural Syst Rehabil Eng 22(3):501–510. doi:10.1109/TNSRE.
2013.2278411
Jiménez-Fabián R, Verlinden O (2012) Review of control algorithms
for robotic ankle systems in lower-limb orthoses, prostheses,
and exoskeletons. Med Eng Phys 34(4):397–408. doi:10.1016/j.
medengphy.2011.11.018
Johnson BF, Singh S, Evans L, Drury R, Datta D, Beard JD (1997) A
prospective study of the effect of limb-threatening ischaemia and
its surgical treatment on the quality of life. EUR J Vasc Endovasc
Surg 13(3):306–314. doi:10.1016/s1078-5884(97)80103-7
Joshi D, Mishra A, Anand S (2011) ANFIS based knee angle predic-
tion: an approach to design speed adaptive contra lateral controlled
123
Biol Cybern (2015) 109:141–162
AK prosthesis. Appl Soft Comput 11(8):4757–4765. doi:10.1016/
j.asoc.2011.07.007
Kaluza P, Cioacă T (2012) Phase oscillator neural network as artificial
central pattern generator for robots. Neurocomput 97:115–124.
doi:10.1016/j.neucom.2012.05.01937
Kao P-C, Ferris DP (2009) Motor adaptation during dorsiflexion-
assisted walking with a powered orthosis. Gait Posture 29(2):230–
236. doi:10.1016/j.gaitpost.2008.08.014
Karlik B, Tokhi MO, Alci M (2003) A fuzzy clustering neural network
architecture for multifunction upper-limb prosthesis. IEEE Trans
Biomed Eng 50(11):1255–1261. doi:10.1109/tbme.2003.818469
Karmarkar AM, Collins DM, Wichman T, Franklin A, Fitzgerald SG,
Dicianno BE, Pasquina PF, Cooper RA (2009) Prosthesis and
wheelchair use in veterans with lower-limb amputation. J Rehabil
Res Dev 46(5):567–576
van Keeken HG, Vrieling AH, Hof AL, Postema K, Otten B (2012)
Principles of obstacle avoidance with a transfemoral prosthetic
limb. Med Eng Phys 34(8):1109–1116. doi:10.1016/j.medengphy.
2011.11.017
Khan MJ, Afzal MR, Naseer N, Koreshi ZU (2012) Control system
design for a prosthetic leg using series damping actuator. In:
2012 international conference on robotics and artificial intelli-
gence (ICRAI), 22–23 Oct 2012. pp 1–6. doi:10.1109/ICRAI.
2012.6413418
Khushaba RN, Kodagoda S, Takruri M, Dissanayake G (2012) Toward
improved control of prosthetic fingers using surface electromyo-
gram (EMG) signals. Expert Syst Appl 39(12):10731–10738.
doi:10.1016/j.eswa.2012.02.192
Klute GK, Kantor C, Darrouzet C, Wild H, Wilkinson S, Iveljic
S, Creasey G (2009) Lower-limb amputee needs assessment
using multistakeholder focus-group approach. J Rehabil Res Dev
46(3):293–304
Kuiken TA, Miller LA, Lipschutz RD, Lock BA, Stubblefield K,
Marasco PD, Zhou P, Dumanian GA (2007) Targeted reinnerva-
tion for enhanced prosthetic arm function in a woman with a prox-
imal amputation: a case study. Lancet 369(9559):371–380. doi:10.
1016/S0140-6736(07)60193-7
Kukke SN, Triolo RJ (2004) The effects of trunk stimulation on
bimanual seated workspace. IEEE Trans Neural Syst Rehabil Eng
12(2):177–185
Lamb PF, Bartlett RM, Robins A (2011) Artificial neural networks for
analyzing inter-limb coordination: the golf chip shot. Hum Mov
Sci 30(6):1129–1143. doi:10.1016/j.humov.2010.12.006
Le F, Markovsky I, Freeman CT, Rogers E (2010) Identification of
electrically stimulated muscle models of stroke patients. Control
Eng Pract 18(4):396–407. doi:10.1016/j.conengprac.2009.12.007
Lee JB, James DA, Ohgi Y, Yamanaka S (2012) Monitoring sprint-
ing gait temporal kinematics of an athlete aiming for the 2012
London Paralympics. Procedia Eng 34:778–783. doi:10.1016/j.
proeng.2012.04.133
Li G, Kuiken TA (2008) Modeling of prosthetic limb rotation control
by sensing rotation of residual arm bone. IEEE Trans Biomed Eng
55(9):2134–2142
Magdalena L (1997) A fuzzy logic controller with learning through the
evolution of its knowledge base. Int J Approx Reason 16:335–358
Maki BE, Sibley KM, Jaglal SB, Bayley M, Brooks D, Fernie GR,
Flint AJ, Gage W, Liu BA, McIlroy WE, Mihailidis A, Perry SD,
Popovic MR, Pratt J, Zettel JL (2011) Reducing fall risk by improv-
ing balance control: development, evaluation and knowledge-
translation of new approaches. J Saf Res 42(6):473–485. doi:10.
1016/j.jsr.2011.02.002
Martins MM, Santos CP, Frizera-Neto A, Ceres R (2012) Assis-
tive mobility devices focusing on smart walkers: classification
and review. Rob Autom Syst 60(4):548–562. doi:10.1016/j.robot.
2011.11.015
Biol Cybern (2015) 109:141–162
Matjačić Z, Hunt K, Gollee H, Sinkjaer T (2003) Control of posture with
FES systems. Med Eng Phys 25(1):51–62. doi:10.1016/S1350-
4533(02)00115-7
Mergner T (2010) A neurological view on reactive human stance con-
trol. Annu Rev Control 34(2):177–198. doi:10.1016/j.arcontrol.
2010.08.001
Mohammed S, Poignet P, Fraisse P, Guiraud D (2012) Toward lower
limbs movement restoration with input–output feedback lineariza-
tion and model predictive control through functional electri-
cal stimulation. Control Eng Pract 20(2):182–195. doi:10.1016/
j.conengprac.2011.10.010
Moreno-Valenzuela J, Salinas-Avila A (2011) Biomimetic control of
mechanical systems equipped with musculotendon actuators. J
Bionic Eng 8(1):56–68. doi:10.1016/S1672-6529(11)60011-5
Muscillo R, Schmid M, Conforto S, D’Alessio T (2011) Early recogni-
tion of upper limb motor tasks through accelerometers: real-time
implementation of a DTW-based algorithm. Comput Biol Med
41(3):164–172. doi:10.1016/j.compbiomed.2011.01.007
Nagano A, Komura T, Fukashiro S, Himeno R (2005) Force, work and
power output of lower limb muscles during human maximal-effort
countermovement jumping. J Electromyogr Kines 15(4):367–376.
doi:10.1016/j.jelekin.2004.12.006
Naik GR, Kumar DK, Palaniswami M (2008) Surface EMG based
hand gesture identification using semi blind ICA: validation of
ICA matrix analysis. Electromyogr Clin Neurophys 48(3–4):169–
180
Nederhand MJ, Van Asseldonk EHF, der Kooij Hv, Rietman HS (2012)
Dynamic Balance Control (DBC) in lower leg amputee subjects;
contribution of the regulatory activity of the prosthesis side. Clin
Biomech 27(1):40–45. doi:10.1016/j.clinbiomech.2011.07.008
Nef T, Guidali M, Riener R (2009) ARMin III—arm therapy exoskele-
ton with an ergonomic shoulder actuation. Appl Bionics Biomech
6(2):127–142. doi:10.1080/1176232090284017939
Newham DJ, Donaldson N de N (2007) FES cycling. Acta Neurochir
Suppl 97(Pt 1):395–402
Nguyen CH, Alici G, Wallace GG (2012) Modelling trilayer conju-
gated polymer actuators for their sensorless position control. Sens
Actuators A Phys 185:82–91. doi:10.1016/j.sna.2012.07.018
Oskoei MA, Hu H (2008) Support vector machine-based classification
scheme for myoelectric control applied to upper limb. IEEE Trans
Biomed Eng 55(8):1956–1965. doi:10.1109/tbme.2008.919734
Park SH, Lee SP (1998) EMG pattern recognition based on artificial
intelligence techniques. IEEE Trans Rehabil Eng 6(4):400–405
Parker P, Englehart K, Hudgins B (2006) Myoelectric signal processing
for control of powered limb prostheses. J Electromyogr Kinesiol
16(6):541–548. doi:10.1016/j.jelekin.2006.08.006
Pérez-Rodríguez R, Marcano-Cedeño A, Costa Ú, Solana J, Cáceres
C, Opisso E, Tormos JM, Medina J, Gómez EJ (2012) Inverse
kinematics of a 6 DoF human upper limb using ANFIS and ANN
for anticipatory actuation in ADL-based physical Neurorehabili-
tation. Expert Syst Appl 39(10):9612–9622. doi:10.1016/j.eswa.
2012.02.143
Phinyomark A, Phukpattaranont P, Limsakul C (2012) Fractal analysis
features for weak and single-channel upper-limb EMG signals.
Expert Syst Appl 39(12):11156–11163. doi:10.1016/j.eswa.2012.
03.039
Di Pino G, Guglielmelli E, Rossini PM (2009) Neuroplasticity in
amputees: main implications on bidirectional interfacing of cyber-
netic hand prostheses. Prog Neurobiol 88(2):114–126. doi:10.
1016/j.pneurobio.2009.03.001
Pizzi R, Cino G, Gelain F, Rossetti D, Vescovi A (2007) Learning
in human neural networks on microelectrode arrays. Biosystems
88(1–2):1–15. doi:10.1016/j.biosystems.2006.03.012
Ponmozhi J, Frias C, Marques T, Frazão O (2012) Smart sen-
sors/actuators for biomedical applications: review. Measurement
45(7):1675–1688. doi:10.1016/j.measurement.2012.02.006
161
Previdi F (2002) Identification of black-box nonlinear models for
lower limb movement control using functional electrical stim-
ulation. Control Eng Pract 10(1):91–99. doi:10.1016/S0967-
0661(01)00128-9
Previdi F, Schauer T, Savaresi SM, Hunt KJ (2004) Data-driven con-
trol design for neuroprotheses: a virtual reference feedback tuning
(VRFT) approach. IEEE Trans Control Syst Technol 12(1):176–
182. doi:10.1109/tcst.2003.82196740
Previdi F, Ferrarin M, Savaresi SM, Bittanti S (2005) Closed-loop con-
trol of FES supported standing up and sitting down using virtual
reference feedback tuning. Control Eng Pract 13(9):1173–1182.
doi:10.1016/j.conengprac.2004.10.007
Pujana-Arrese A, Mendizabal A, Arenas J, Prestamero R, Landaluze J
(2010) Modelling in Modelica and position control of a 1-DoF set-
up powered by pneumatic muscles. Mechatronics 20(5):535–552.
doi:10.1016/j.mechatronics.2010.05.002
Resnik L, Meucci MR, Lieberman-Klinger S, Fantini C, Kelty DL,
Disla R, Sasson N (2012) Advanced upper limb prosthetic
devices: implications for upper limb prosthetic rehabilitation. Arch
Phys Med Rehabil 93(4):710–717. doi:10.1016/j.apmr.2011.11.
010
Richardson R, Brown M, Bhakta B, Levesley M (2005) Impedance
control for a pneumatic robot-based around pole-placement, joint
space controllers. Control Eng Pract 13(3):291–303. doi:10.1016/
j.conengprac.2004.03.011
Rosenfeld A, Dvorachek M, Rotstein I (2000) Bronze single crown-
like prosthetic restorations of teeth from the Late Roman period.
J Archaeolog Sci 27(7):641–644. doi:10.1006/jasc.1999.0517
Rossini PM, Micera S, Benvenuto A, Carpaneto J, Cavallo G, Citi L,
Cipriani C, Denaro L, Denaro V, Di Pino G, Ferreri F, Guglielmelli
E, Hoffmann K-P, Raspopovic S, Rigosa J, Rossini L, Tombini
M, Dario P (2010) Double nerve intraneural interface implant
on a human amputee for robotic hand control. Clin Neurophys-
iol 121(5):777–783. doi:10.1016/j.clinph.2010.01.001
Rusaw D, Ramstrand N (2010) Sagittal plane position of the functional
joint centre of prosthetic foot/ankle mechanisms. Clin Biomech
25(7):713–720. doi:10.1016/j.clinbiomech.2010.04.005
Sagawa Y Jr, Turcot K, Armand S, Thevenon A, Vuillerme N, Wate-
lain E (2011) Biomechanics and physiological parameters during
gait in lower-limb amputees: a systematic review. Gait Posture
33(4):511–526. doi:10.1016/j.gaitpost.2011.02.003
Santos CP, Matos V (2012) CPG modulation for navigation and omnidi-
rectional quadruped locomotion. Rob Autom Syst 60(6):912–927.
doi:10.1016/j.robot.2012.01.004
Sawers A, Hahn ME (2011) Trajectory of the center of rotation in non-
articulated energy storage and return prosthetic feet. J Biomech
44(9):1673–1677. doi:10.1016/j.jbiomech.2011.03.028
Schauer T, Hunt K (2000) Linear modelling and controller design for
the single limb movement of paraplegics using FES. In: Carson
E, Salzsieder E (eds) A proceedings volume from the 4th IFAC
symposium on modell control in biomedical systems. Greifswald,
Germany, March-April 2000. pp 7–12
Scheme E, Englehart K (2011) Electromyogram pattern recognition
for control of powered upper-limb prostheses: state of the art and
challenges for clinical use. J Rehabil Res Dev 48(6):643–659
Scholz MS, Blanchfield JP, Bloom LD, Coburn BH, Elkington M, Fuller
JD, Gilbert ME, Muflahi SA, Pernice MF, Rae SI, Trevarthen JA,
White SC, Weaver PM, Bond IP (2011) The use of composite
materials in modern orthopaedic medicine and prosthetic devices:
a review. Compos Sci Technol 71(16):1791–1803. doi:10.1016/j.
compscitech.2011.08.017
Schultz AE, Kuiken TA (2011) Neural interfaces for control of upper
limb prostheses: the state of the art and future possibilities. PM&R
3(1):55–67. doi:10.1016/j.pmrj.2010.06.016
123
162
Schwartz AB, Cui XT, Moran DW (2006) Brain-controlled interfaces:
movement restoration with neural prosthetics. Neuron 52(1):205–
220. doi:10.1016/j.neuron.2006.09.019
Sinha N, Yeow JTW (2005) Carbon nanotubes for biomedical applica-
tions. IEEE Trans Nanobiosci 4(2):180–195
van der Smagt P (2000) Benchmarking cerebellar control. Rob Autom
Syst 32(4):237–251. doi:10.1016/S0921-8890(00)00090-7
Takeda H, Tsujiuchi N, Koizumi T, Kan H, Hirano M, Nakamura Y
(2009) Development of prosthetic arm with pneumatic prosthetic
hand and tendon-driven wrist. Conf Proc IEEE Eng Med Biol Soc
51(10):5333668
Tan AH (2009) Direction-dependent systems—a survey. Automatica
45(12):2729–2743. doi:10.1016/j.automatica.2009.09.024
Terasawa N, Takeuchi I (2013) Electrochemical and electromechani-
cal properties of high-performance polymer actuators containing
vapor grown carbon nanofiber and metal oxide. Sens Actuators B:
Chem 176:1065–1073. doi:10.1016/j.snb.2012.10.043
Thanh TUDC, Ahn KK (2006) Nonlinear PID control to improve the
control performance of 2 axes pneumatic artificial muscle manip-
ulator using neural network. Mechatronics 16(9):577–587. doi:10.
1016/j.mechatronics.2006.03.011
Timmermans A, Seelen H, Willmann R, Kingma H (2009) Technology-
assisted training of arm-hand skills in stroke: concepts on reacqui-
sition of motor control and therapist guidelines for rehabilitation
technology design. J NeuroEng Rehabil 6(1):1
Tiwana MI, Redmond SJ, Lovell NH (2012) A review of tactile sensing
technologies with applications in biomedical engineering. Sens
Actuators A 179:17–31. doi:10.1016/j.sna.2012.02.05142
Tormene P, Giorgino T, Quaglini S, Stefanelli M (2009) Matching
incomplete time series with dynamic time warping: an algorithm
and an application to post-stroke rehabilitation. Artif Intell Med
45(1):11–34. doi:10.1016/j.artmed.2008.11.007
Torop J, Sugino T, Asaka K, Jänes A, Lust E, Aabloo A (2012)
Nanoporous carbide-derived carbon based actuators modified with
gold foil: prospect for fast response and low voltage applications.
Sens Actuators B Chem 161(1):629–634. doi:10.1016/j.snb.2011.
10.079
Vanoncini M, Holderbaum W, Andrews BJ (2012) Electrical Stimu-
lation for trunk control in paraplegia: a feasibility study. Control
Eng Pract 20(12):1247–1258. doi:10.1016/j.conengprac.2012.06.
007
Varol HA, Sup F, Goldfarb M (2009) Powered sit-to-stand and
assistive stand-to-sit framework for a powered
transfemoral prosthesis. IEEE Int Conf Rehabil Rob
5209582:645–651
Vaugoyeau M, Hakam H, Azulay J-P (2011) Proprioceptive impairment
and postural orientation control in Parkinson’s disease. Hum Mov
Sci 30(2):405–414. doi:10.1016/j.humov.2010.10.006
Wang J (2001) Glucose Biosensors: 40 years of advances and
challenges. Electroanalysis 13(12):983–988. doi:10.1002/1521-
4109(200108)13:12<983:aid-elan983>3.0.co;2-#
Watve S, Dodd G, MacDonald R, Stoppard ER (2011) Upper limb pros-
thetic rehabilitation. Orthopaed Trauma 25(2):135–142. doi:10.
1016/j.mporth.2010.10.003
123
Biol Cybern (2015) 109:141–162
Weir RF (2008) Standard handbook of biomedical engineering and
design. Amazon, United Nation, Chapter 32, 43
Wen-Wei Hsu R, Sim FH, Chao EYS (1999) Reoperation results
after segmental prosthetic replacement of bone and joint for
limb salvage. J Arthroplasty 14(5):519–526. doi:10.1016/s0883-
5403(99)90071-2
Whalen SA, Richards CD, Bahr DF, Richards RF (2007) Characteriza-
tion and modeling of a microcapillary driven liquid-vapor phase-
change membrane actuator. Sens Actuators A Phys 134(1):201–
212. doi:10.1016/j.sna.2006.04.038
Wise KD (2007) Integrated sensors, MEMS, and microsystems: reflec-
tions on a fantastic voyage. Sens Actuators A 136(1):39–50.
doi:10.1016/j.sna.2007.02.013
Wolf SI, Alimusaj M, Fradet L, Siegel J, Braatz F (2009) Pressure
characteristics at the stump/socket interface in transtibial amputees
using an adaptive prosthetic foot. Clin Biomech 24(10):860–865.
doi:10.1016/j.clinbiomech.2009.08.007
Wu S-K, Waycaster G, Shen X (2011) Electromyography-based control
of active above-knee prostheses. Control Eng Pract 19(8):875–882.
doi:10.1016/j.conengprac.2011.04.017
Yap TTV, Ai Hui T, Foo MFL (2007) Modelling of direction-dependent
systems using bilinear models. In: Conference Proceedings IEEE
Instrum Measurement Technology 2007. IMTC 2007. 1–3 May
2007. pp 1–6. doi:10.1109/imtc.2007.379028
Yeh TJ, Wu M-J, Lu T-J, Wu F-K, Huang C-R (2010) Control of McK-
ibben pneumatic muscles for a power-assist, lower-limb orthosis.
Mechatronics 20(6):686–697. doi:10.1016/j.mechatronics.2010.
07.004
Zardoshti-Kermani M, Wheeler BC, Badie K, Hashemi RM (1995)
EMG feature evaluation for movement control of upper extremity
prostheses. IEEE Trans Rehabil Eng 3(4):324–333. doi:10.1109/
86.481972
Zecca M, Micera S, Carrozza MC, Dario P (2002) Control of multi-
functional prosthetic hands by processing the electromyographic
signal. Crit Rev Biomed Eng 30(4–6):459–485
Zhang D, Poignet P, Widjaja F (2011) Neural oscillator based control
for pathological tremor suppression via functional electrical stimu-
lation. Control Eng Pract 19(1):74–88. doi:10.1016/j.conengprac.
2010.08.009
Zhang F, Li P, Hou Z-G, Lu Z, Chen Y, Li Q, Tan M (2012) sEMG-
based continuous estimation of joint angles of human legs by using
BP neural network. Neurocomput 78(1):139–148. doi:10.1016/j.
neucom.2011.05.033
Zhang J, Yin Y (2012) SMA-based bionic integration design of self-
sensor–actuator-structure for artificial skeletal muscle. Sens. Actu-
ators A 181:94–102. doi:10.1016/j.sna.2012.05.017
Ziegler-Graham K, MacKenzie EJ, Ephraim PL, Travison TG, Brook-
meyer R (2008) Estimating the prevalence of limb loss in the
United States: 2005 to 2050. Arch Phys Med Rehabil 89(3):422–
429