Mechatronics 49 (2018) 77–91

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Hybrid impedance control of a robot manipulator for wrist and forearm T

rehabilitation: Performance analysis and clinical results☆
⁎
Erhan Akdoğan ,a, Mehmet Emin Aktana, Ahmet Taha Korua, M. Selçuk Arslana, Murat Atlıhana,
Banu Kuranb
a
Yıldız Technical University, Mechanical Engineering Faculty, Department of Mechatronics Engineering Department, Istanbul, Turkey
b
Yeni Yüzyıl University, Medical Sciences Faculty, Department of Physiotherapy and Rehabilitation, Istanbul, Turkey

A R T I C L E I N F O A B S T R A C T

Keywords: Therapeutic exercises play an important role in the physical therapy and the rehabilitation. The exercises that
Upper limb rehabilitation can be assisted by a physiotherapist are increasingly being performed by the rehabilitation robots partially or
Therapeutic exercise robot fully due to their various merits. This study aims to develop a complete rehabilitation system, which consists of a
Hybrid impedance control rehabilitation robot, an HMI and a hybrid impedance controller that can model all the therapeutic exercises for
an upper limb rehabilitation. The 3-DOF upper limb rehabilitation robot is able to perform the movements of
flexion–extension and ulnar–radial deviation for the wrist, and the movement of pronation–supination for the
forearm. The experimental studies were conducted with healthy subjects and patients. First, the experiments
were done with the healthy subjects to prove the control performance of the robotic system. The results showed
that the hybrid impedance controlled robot can perform the therapeutic exercises very successfully. Then, the
experimental studies were carried out with the real patients in a clinical environment. At the end of the treat-
ment process, remarkable improvements were observed in terms of the limb force in all of the patients.

1. Introduction
Rehabilitation is a treatment process to bring an individual with a
physical disability, which might be congenital or happen due to an
illness, injury, or accident, to the best condition medically, socially and
vocationally, and to reduce the negative results of permanent diseases
to minimum [1]. A limb, which is injured due to the age-associated
muscle disabilities, work or traffic accidents, wars and chronic diseases,
needs rehabilitation to refunction fully or partially. Making a limb
functional and increasing the force of a muscle are crucial problems.
The return of those people to their social life is also highly important for
themselves, their families, and the society they live in.
One of the elements of the rehabilitation is the refunction of the
limbs, such as arms and legs. The therapeutic exercises play a crucial
role in the process of refunction. A physiotherapist can make the patient
perform the therapeutic exercises, which consist of the passive and
active exercises, or the patient can perform by himself or herself de-
pending on his or her physical condition. Especially, in populous
countries, where the number of physiotherapists per patient is not en-
ough (to set an example, in Turkey, physiotherapists are allowed to
accept 16 patients in a day [2]), the transportation problems of
patients, the cost of the rehabilitation process, and the constraints in the
existing devices and equipments of the therapeutic exercises are the
main problems of the rehabilitation process. For these reasons, the re-
searches on the use of robots in the rehabilitation process have in-
creased in the last 15 years [3]. The robots make a significant con-
tribution to the rehabilitation process in terms of the cost, the duration
of therapy, the objective evaluation, the remote control, and enabling
home care. The rehabilitation robots can be classified in four groups
[4]:
• “The assistive robots” supporting the movements of disabled people
in the activities of their daily lives,
• “The prostheses” fulfilling the functions of the severed limb for
amputees,
• The robots used for the gait rehabilitation,
• “The therapeutic exercise robots” help patients perform passive,
active and resistive exercises.
The system of interest in this research, which aims the therapeutic
exercises, is for the rehabilitation of the wrist and the forearm and
belongs to the class of the therapeutic exercise robots.

☆
This paper was recommended for publication by Associate Editor Kong Kyoungchul.
⁎
Corresponding author.
E-mail address: eakdogan@yildiz.edu.tr (E. Akdoğan).

https://doi.org/10.1016/j.mechatronics.2017.12.001
Received 27 September 2016; Received in revised form 6 July 2017; Accepted 3 December 2017
Available online 13 December 2017
0957-4158/ © 2017 Elsevier Ltd. All rights reserved.
E. Akdoğan et al.
The rehabilitation robots developed for the upper-limb rehabilita-
tion can be compared in terms of the capacity of the movement and
exercise, the mechanical properties, and the control methods. The
systems in the literature are capable of performing one or some of the
following exercises: The passive, the active assisted, and the resistive.
The mechanical parts of the systems were mostly developed by either
using already available robotic manipulators or designing robotic ma-
nipulators or exoskeletons from scratch. The widely used control ap-
proaches in the rehabilitation are as follows: The convential control
methods, such as PID or PD, the direct torque control, the admittance
control, and the impedance control.
The most well-known robotic system of the upper-limb rehabilita-
tion studies is the MIT-MANUS (Massachusetts Institute of Technology-
MANUS) developed by Krebs et al. [5,6]. The robot was developed for
the rehabilitation of the shoulder and the elbow and has 3-DOF. The
system can perform the passive, the active assisted, and the resistive
exercises. It can reteach the limb the motion limits of the limb and can
implement applications based on target trajectory. The impedance
control method was used in the control of the system. In their study,
Reinkensmeyer et al. designed a 4-DOF mechanism, called ARM-Guide
(Assisted Rehabilitation and Measurement), for the rehabilitation of the
shoulder and the elbow [7]. The system can perform the passive, active
assisted, and resistive exercises. The PD position control and the direct
torque control methods were used in the control of the system. The
REHAROB by Toth et al. was designed by using a 6-DOF industrial
robot for the rehabilitation of the shoulder, the elbow, and the forearm
[8]. The system has the capacity of performing passive exercises for
decreasing the spasticity. The robot is taught the movements by a
physiotherapist through the direct teaching method using an artificial
intelligence based control algorithm.
An exoskeleton robot called ARMin was developed by Riener et al.
for the rehabilitation of the shoulder and the elbow [9]. The robot has
6-DOF (four active and two passive) and can perform passive and active
assisted exercises. The system can implement applications based on
target trajectory, can feed the audio and visual information back, and
has the ability of gravity compensation. The admittance and impedance
control approaches were used in the control of the system. The use of
such exoskeleton robots in the rehabilitation have generated an im-
mense interest and many studies using the aforementioned control
approaches reported in the literature [10–22].
Lum et al. designed a system, MIME (Mirror Image Movement
Enabler), by using PUMA 360 robot for the rehabilitation of the
shoulder and the wrist [23]. This system has 4-DOF and uses “mirror
therapy” method in implementing the passive, active assisted, and re-
sistive exercises. Additionally, the system can regain the limb the limits
of the range of motion (ROM) and implement the mirror therapy and
the applications based on the target trajectory. The PID position control
and the direct torque control methods were used in the control of the
system. A similar study using the mirror therapy method was presented
by Lewis et al. [24]. In their study, the system was controlled by using
EMG (electromyography) - based admittance control method. One of
the control approaches fitting well to the control of interaction between
the robot and the human is the admittance control and it plays an
important role in the rehabilitation [25–27]. In the admittance control,
the robot adjusts the desired motion based on the measurements of the
interaction forces.
The hybrid impedance control method was developed by Anderson
and Spong [28]. In this technique, the strategies of “the impedance
control” and “the hybrid force-position control” are combined in a
framework. By doing so, both the position and the impedance-based
force controls are implemented within a single control structure. Re-
searchers have used the hybrid impedance control in various applica-
tions. Wang et al. developed a hybrid impedance controlled 3-DOF
system by using the PUMA 562 robot [29]. The robot has a tactile
sensor on its end-effector and designed for massaging. Selection of the
desired impedance value allows massaging in certain levels of the
78
Mechatronics 49 (2018) 77–91
stiffness. Another system designed for localizing tumours in the body
has 7-DOF and is controlled by a hybrid impedance control approach
[30]. A 3-DOF system using both the impedance control and the hybrid
impedance control for the rehabilitation of the wrist, the forearm, and
the shoulder was developed by Wang et al. [31]. The system was de-
signed for the motor recovery in the stroke patients by training the
upper-limb through the predetermined tasks.
Akdogan and his colleagues were developed a 3-DOF system, which
uses the impedance control in the therapeutic exercises, for the re-
habilitation of the lower-limb [32,33]. In this system, to establish the
therapeutic exercise modes, the PID control was used for the position-
based exercises, and the impedance control was used for the force-based
exercises [34]. These two techniques were used by switching between
each other in the exercises requiring both the position control and the
impedance control (active assistive exercises). Two disadvantages ap-
peared in this point. One is that having the trajectory of force by using
the position-based impedance model is not possible, since the trajectory
of force changes depending on the impedance parameters. In the ex-
ercises requiring the PID and impedance control, the numbers of
parameters are three for the impedance control (the coefficients of in-
ertia, stiffness, and damping) and three for the PID control. The six
parameters in total must be set properly. Another disadvantage is the
occurrence of instability due to the disturbances, such as the noise, in
the process of decision to perform switching. Thus, the control of the
system affected unfavorably. For this reason, the hybrid impedance
control method is highly suitable to perform the control of the required
position and force as well as the desired mechanical impedance for the
therapeutic exercises.
Some preliminary studies having more direct relation with this
study should be cited: [35–37]. Mainly, the concept of the discussed
system in this work and some experimental results with healthy sub-
jects are explained in [35] and [36]. These works do not include clinical
results and detailed system identification studies.
There are some commercial therapeutic exercise machines available
in the market, such as Biodex, Cybex and Kincom. These devices are
passive machines and they cannot change the applied position and force
to a patient during the exercise. They are single degree of freedom. For
different type movement, additional apparatus are needed. Therefore,
the higher degree of freedom intelligent robotic systems which can
change the exercise procedure according to patient’s situation are ap-
pear to be more useful than passive exercise machines.
In this study, the hybrid impedance control of this robot manip-
ulator was implemented for the rehabilitation of the wrist and the
forearm. First, the performance of the system was shown by the ex-
perimental studies carried out with healthy subjects. The results in-
dicated that the hybrid impedance control based robotic system can
perform the passive, the active assistive, and the resistive exercises very
accurately. Then, the experimental studies were carried out with real
patients in a clinical environment. At the end of the treatment process,
improvements were observed in terms of limb force in all of the pa-
tients. The results are presented in terms of the ROM (Range of Motion),
the limb force, etc.
The contribution of this study into the literature is that the ther-
apeutic exercises (passive, active assistive, resistive) were performed
under a single control structure using hybrid impedance control and the
effectiveness of this control method was shown by clinical experiments.
In the literature, any study using hybrid impedance control in modeling
all the therapeutic exercises does not exist. Also, another contribution is
the development of an intelligent human–machine interface (HMI).
This powerful HMI was developed by combining knowledge- and rule-
based intelligent techniques and a conventional control technique.
2. System description
The developed rehabilitation support system consists of a phy-
siotherapist, a patient, a robot manipulator, and an HMI. The block
E. Akdoğan et al.
Exercise & Patient Position & Force
Information
HUMAN
MACHINE
INTERFACE
Results
Force & Position
PHYSIOTHERAPIST
Fig. 1. System block diagram.
diagram of the rehabilitation system is shown in Fig. 1. The phy-
siotherapist, who is the main user of the system, determines the type of
the therapeutic exercise and enters the data specific to the determined
exercise as well as personal information of the patient to the system
through the HMI. The HMI uses the entered data and sends the calcu-
lated force and position commands to the robot manipulator. The robot
manipulator, which is attached to the upper limb of the patient, makes
the patient perform the exercises. The position and force information of
the limb are sensed by the position and force sensors placed on the
robot manipulator. This information is fed back to the HMI and then
processed to generate the information related with the therapy, which
are the ROM of the patient, the limb force or reaction force of the pa-
tient, for displaying to the physiotherapist. The detailed explanations
on the robot manipulator and HMI are given in the following sections.
2.1. Robot manipulator and hardware configuration
2.1.1. Robot manipulator
The robot manipulator is of three-degree-of-freedom and designed
according to the design principles to perform the therapeutic exercises
for the forearm and the wrist. Firstly, the functional requirements, and
then, the design parameters, which will meet those functional re-
quirements, are determined:
Functional requirements. The functional requirements of the robot
manipulator are as follows:
• Perform the movements of pronation–supination, ulnar–radial de-
viation, and flexion–extension for the rehabilitation of wrist and
forearm.
• Perform the therapeutic exercises of passive, active assistive, and
resistive (isotonic, isometric, isokinetic).
• Record the position and force information.
• Be light and portable.
• Use in the conditions of home and hospital.
• Adjust its length according to the size of the limb.
• Provide a hybrid safety via hardware and software.
Design parameters. The design parameters, which will meet the
aforementioned functional requirements and the system specifications
of the rehabilitation robot system, are as follows.
• The robot was designed as of three-degree-of-freedom to perform
the movements of pronation–supination, ulnar–radial deviation, and
flexion–extension. The motion limits of the robot manipulator are in
accordance with the motion limits of human. The relevant values
are given in Table 1.
• Servo motors are used as actuators, since the therapeutic exercises of
passive, active assistive, and resistive require the control of position
and force.
• A force sensor is used to measure the force applied to the patient, or
by the patient. Encoders are used to measure the position.
• A 7000 series Aluminium alloy is used in order that the robot ma-
nipulator is light and portable.
• A compact design of the system was preferred to use it in the con-
ditions of home and hospital.
Mechatronics 49 (2018) 77–91
Command Movement Command
ROBOT
MANIPULATOR
Reaction
Feedback PATIENT
Table 1
Range of motion capability of healthy human vs the robot manipulator.
Movement type Human range of motion Manipulator max. range
Pronation 80°–90° 85°
Supination 80°–90° 85°
Radial deviation 15°–30° 30°
Ulnar deviation 30°–45° 45°
Flexion 60°–85° 80°
Extension 50°–80° 80°
• The robot manipulator can be adjusted according to the size of the
limb.
• To prevent the injury of the patient that may occur due to any
problem during exercises, the safety of the patient and the system
was provided; i) mechanically by limiting the moving parts attached
to the shafts of the motors, ii) by motion limiting algorithms in the
software of the system, and iii) by an emergency stop button in the
hardware system.
The reader is referred to [36] for the detailed information on the
design process of the upper limb rehabilitation system.
The general structure of the robot manipulator is shown in Fig. 2.
The arm of the patient is placed in the armrest and fastened. The hand
of the patient is placed between the bars of the handle. There are holes
for stoppers at each axle for safety. The ROM of the mechanism can be
adjusted for any patient by inserting pins into the holes.
2.1.2. Hardware configuration
The block diagram of the hardware system is shown in Fig. 3. The
physiotherapist is the main user of the system, and therefore, enters all
the information relevant to the patient and the therapy to the HMI. The
algorithms of the system were developed in MATLAB and xPC was used
for the real-time prototyping. There are three PC’s in the system: The
Main PC running the algorithms of the system, the Target PC having the
DAQ cards that provide the analog-digital conversions, and the Patient
PC running the patient graphical interface for the game-based exercise
types. The communication between the Main PC and the Target PC is
provided by the TCP/IP. The UDP provides the communication between
the Main PC and Patient PC. The details on the data acquisition cards,
the actuators, and the sensors are as follows.
Actuators and drivers. There are three servo motor/gears/encoder
combinations (Maxon EC-max) and three servo motor drivers (Maxon
EPOS 2 50/5), which are responsible for the actuation in the system.
The specifications of motors, gears, and encoders are given in Table 2.
Force and position sensors. There is a force/torque sensor equipment
(ATI Nano 25) to measure the force and torque applied by the patient.
This sensor is placed under the handle. By using this six-axis sensor,
force is measured for the movements of flexion–extension and
ulnar–radial deviation, whereas torque is measured for the movement
of pronation–supination. The measurement ranges of the sensors for
each axis are given in Table 3. The position information is obtained
from the encoders attached to the couple of motor and speed reducer.
79
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Fig. 2. The general structure of the robot manipulator.

Data acquisition cards. In the system, for the analog inputs, the analog
outputs and the encoder data, NI PCI-6225, NI PCI-6703, and NI PCI-
6601 DAQ cards are used, respectively, at the sampling time of 1 ms.
2.2. Human–Machine interface
The HMI controls the system and establishes the communication
between the patient, robot manipulator, and physiotherapist. The
detailed block diagram of the system is shown in Fig. 4. The HMI
consists of the main controller, the graphical user interface (GUI), the
rule base, the data base, and the hybrid impedance controller. As it is
understood, the HMI is an original software structure developed by
using a knowledge and rule-based approach. It includes intelligence and
a conventional controller, which is the hybrid impedance controller.
The main controller is responsible for the communication between all
the units. The physiotherapist enters the information about the patient

Patient PC
TCP/IP Connection UDP Connection

Target PC Robot Manipulator

Physiotherapist

Main PC – User Interface

Patient

Analog Motor
Torque
Output Board Drivers

Encoder
Position Encoder
Input Board

Analog Input Force

Force
Board Sensor

Fig. 3. The block diagram of the hardware system.

80
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Table 2 experiments with real patients were carried out in a clinical environ-
Technical specifications of the actuator–gearbox–encoder combinations. ment. In these experiments, the system was tested by the isotonic and
isometric exercises depending on the patient profile. The experimental
Axis Motor Reduction Encoder Output Torque
results of healthy subjects and patients are given in Section 5 in detail.
Pronation–Supination Maxon EC- 103:1 256 pulse/rev 20.747 Nm Among the exercise types, the passive exercise requires the position
max 30 control; the isotonic–isometric–robo resistive mode requires the force
Ulnar–Radial dev. Maxon EC- 86:1 500 pulse/rev 3.46 Nm
control; and the active assistive exercise requires both the position
max 30
Flexion–Extension Maxon EC- 128:1 500 pulse/rev 2.88 Nm control and the force control. Therefore, the HMI makes the hybrid
max 22 impedance controller operate in the position-force mode or the hybrid
control mode according to the exercise type. The detailed explanations
on the performed exercise types in terms of the inputs and the control
Table 3 methods are given as follows.
The measurement ranges of the force sensor.

Force X Force Y Force Z Torque X Torque Y Torque Z 2.2.1. Passive exercise

± 150 N ± 150 N ± 500 N ± 3 Nm ± 3 Nm
and the exercise through the GUI. The values of impedance parameters
for the matching exercise type in the rule base and the selection matrix
of the hybrid impedance according to this information are received by
the main controller and sent to the hybrid impedance controller (the
details are given in Section 4). In addition, the trajectory information of
position or force for a certain type of exercise is also sent to the hybrid
impedance controller. The information received from sensors is fed
back to the main controller and the hybrid impedance controller. All
the information is displayed to the physiotherapist in the GUI. The
feedback information is used for control purpose in the hybrid im-
pedance controller. This controller can operate in position and force
modes according to the exercise types. The hybrid control can be
achieved by switching between these modes when needed. At the end of
each therapy, the date, the duration, the type of exercise, the ROM, and
the force values are recorded to the database.
The passive, active assistive, isotonic, isometric, and isokinetic ex-
ercises are performed by the developed interface (the results of iso-
kinetic exercises are not given in this study, since the target audience is
the patients). In addition to these conventional exercise types, another
mode called “robo-resistive exercise” was developed in this study.
Several resistance levels (low, medium, high, highest) by the various
combinations of the parameters of the hybrid impedance controller are
obtained in this type of exercise. In the rule base, there are impedance
control parameters according to the exercise types such as the reference
trajectory, the values of impedance control parameters, and the
switching commands.
The exercises were tested with healthy subjects and the capacity of
implementing the exercise types is shown in this study. Besides,
The passive exercise is performed manually or by the assistance of a
± 3 Nm device within motion limits of the limb. It does not include the co-
ordinated voluntary muscle contraction of the patient. This exercise
requires only the position control, and thus, the hybrid impedance
controller operates only in the position control mode. The inputs are the
type of movement, the ROM, the number of repetition, and the velocity.
The impedance parameters are adjusted so that the patient’s limb is
forced to stretch or not by the robot during the therapy. The details of
adjustment of the impedance parameters are given in Section 4. The
outputs obtained at the end of the exercise are the limb position and the
position error.
2.2.2. Active assistive exercise
In this type of exercise, the patient can move his or her limb to some
extent, but cannot reach the full ROM due to the decreased muscle
strength. The physiotherapist assists the patient manually from the
point where the patient is not able to move his or her limb. The purpose
of this exercise is to enable the start of the active muscle contraction
independently. The robotic system performs this exercise by switching
between force and position control modes. The exercise begins with the
force control mode with low impedance parameters. The patient moves
his or her limb to the extent that he or she is able to move it. After this
point, the control mode is switched to the position control mode. The
robotic system moves the patient’s limb to the target position. The
target position is adjusted by setting its amplitude (target ROM) and
period via the GUI before the exercise. The outputs of the exercise are
the limb position and force.
2.2.3. Isotonic exercise
The patient performs the isotonic exercise by doing work against the
constant force. To model this exercise, hybrid impedance control is used
in the force control mode. The input of the controller is the target force.

Data Rule
Human Machine
Base Base
Interface

Impedance Parameters

Impedance
Parameters Torque Robot
Patient & Exercise Main Controller Hybrid Impedance
Information and GUI Controller Manipulator
Position & Force
Trajectory

Sensors
Position & Force

Fig. 4. The block diagram of the human–machine interface.

81
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

A game, which the patient tries to follow the position of a ball, was System identification
developed. The amplitude and period of sinusoidal trajectory of the ball
and the target force value are set via the GUI. The outputs obtained at Performance of the hybrid impedance control depends on the ac-
the end of the exercise are the limb force, the force error, the limb curacy of the mathematical model of the system, since it includes direct
position and the position error. cancellation of dynamics. Hence, we identified the base parameters
listed in Table 4 to improve the performance of the system.
There are ten inertial parameters per link,
2.2.4. Isometric exercise
Through this exercise, the level of muscle contraction is increased [Ixx,i, Ixy, i , Ixz, i , Iyy, i , Iyz, i , Izz, i , mi rci, x , mi rci, y, mi rci, z , mi],
without causing a change in the length of the muscle. It can be per-
which are called the standard inertial parameters. The first six para-
formed by pressing a stationary object, by opposing the manual act of
meters are the moments of inertia, mi is the mass, and rci, x , rci, y, rci, z are
the physiotherapist or by holding a weight in a static condition. At the
the positions of the center of mass in x, y, and z coordinates. In order to
beginning of the exercise, the robot moves to target position where the
facilitate the identification process, a minimum set of inertial para-
isometric exercise will be performed (position control mode). After
meters, which are called the base parameters such as listed in Table 4,
reaching the target position, the control mode is switched to the force
are calculated. We refer the reader to [39] for the calculation of the
control mode. At that point, the robot applies a force to the subject’s
base parameters.
limb. The task of the subject is to keep the limb constant against this
Let us call the unknown base parameters vector
external force. In the developed game, the subject tries to keep the ball
p = [p1 , p2 , …, p17 , fc1 , fv1 , fc2 , fv2 , fc3 , fv3 ]T . The system dynamics in
at the target position. The constant position, the target force and the
(1) are linear with respect to the system parameters when qi, q̇i and q̈i
number of repetition are set via the GUI. The output of the exercise is
are known:
the limb position and force.
Φ(q, q˙ , q¨) p = τ ,

2.2.5. Robo-Resistive exercise where Φ is the regressor matrix. Consider we have M observations data,
It is aimed in this type of exercise that the patient follows the target we end up with overdetermined set of equations A p = b with
at various difficulty levels and values of the ROM. This exercise is not 1 1 1
similar to other resistive exercises, e.g., isotonic, isometric, etc. In ⎡ Φ(q , q˙ , q¨ ) ⎤ ⎡τ ⎤
1
⎢ Φ(q2 , q˙ 2 , q¨2) ⎥ ⎢ τ2 ⎥
contrast to conventional resistive exercises, this exercise is performed
A=⎢
⎢
· ⎥,
⎥ b = ⎢ · ⎥.
against different resistance levels which is adjusted according to the set · ⎢ ·· ⎥
⎢ · ⎥ ⎢ M⎥
impedance parameters. Because of this, the exercise is called robo-re- ⎢ Φ(q M , q˙ M , q¨ M ) ⎥
⎣ ⎦ ⎣τ ⎦ (3)
sistive exercise in this study. To model this exercise, hybrid impedance
control is used in force control mode with different levels of impedance The condition number of the matrix A determines the sensitivity of
parameters (low, medium, high, highest). The details of the selection of the least square solution of parameters, p, with respect to sensor noises.
impedance parameters are given in Section 4. The task of the patient is The trajectory of qi, q̇i and q̈i determines the condition number of the
to follow the ball representing ROM in the game screen. The inputs are matrix A. Our aim is to find the optimal trajectories for each link, which
the velocity of the ball, the ROM, and the resistance level of the ex- minimize the condition number of A, to obtain more accurate results as
ercise. The outputs are the limb position and the position error. in [40].
Let us represent the trajectories of each link as a finite Fourier
series:
3. Dynamic analysis of the robot manipulator
Ni
ai bli
Obtaining the dynamic model of the robot manipulator is very im- qi (t ) = ∑ ω l l sin(ωf lt ) − ωf l
cos(ωf lt ) + qi0
l=1 f
portant for the control. Particularly, in the impedance control method, Ni
the system parameters must be known with high accuracy to achieve a q˙ i (t ) = ∑ ali cos(ωf lt ) + bli cos(ωf lt )
good control performance. In this section, the dynamic equations of the l=1
system are obtained and its parameters are calculated by using a system Ni

identification method. The MATLAB™Symbolic Toolbox was used to q¨i (t ) = ∑ − ali ωf l sin(ωf lt ) + bli ωf l cos(ωf lt )
l=1
obtain the dynamic equations of the robot manipulator in this study.
The robot dynamic equation is expressed by the following nonlinear with ωf is the fundamental pulsation of the Fourier series, al and bl are
equation: the coefficients and Ni is the number of harmonics. The trajectory is
optimized by the following constrained optimization problem:
τ = M (q) q¨ + C (q, q˙ ) + G (q) + F (q˙ ) − J T (q) Fe , (1)
δ * = arg mincond(δ , ωf )
δ (4)
where M(q), C(q), G(q), J(q) are the inertia, the Coriolis and centrifugal,
the gravitational and the Jacobian matrices, respectively. The vector q subject to
is the angular position of the robot joints. The vector Fe is the force
qmin ≤ q (t ) ≤ qmax (5)
applied by the patient to the end effector, which is the output of the
force sensor. The vector F (q˙ ) contains friction forces. The friction is − qmax ≤ q˙ (t ) ≤ q˙max (6)
assumed to be a simple Coulomb-viscous model for each link i ∈ {1, 2,
3}: − q¨max ≤ q¨ (t ) ≤ q¨max . (7)
fi (q˙i ) = fci sign(q˙ i ) + fvi q˙i , (2) where δ is the vector containing Fourier coefficients as

and F (q˙ ) = [f1 (q˙ 1), f2 (q˙ 2), f3 (q˙ 3)]T where fci ’s are the Coulomb friction
coefficients and fvi ’s are the viscous friction coefficients. Assignment of
the coordinate frames of the links and the corresponding
Denavit–Hartenberg parameters can be seen in Fig. 5. For the calcula-
tion of the matrices M(q), C(q) and G(q), we refer the reader to [38].
δ = [a11, …, a N1 i , q10 , a12, …, a N2i , q20 , a13, …, a N3i , q30 , b11, …, b N1 i ,
b12, …, b N2i , b13, …, b N3i ].
and δ* is the optimal δ which minimizes the condition number of matrix
A. The parameters ωf and Ni are pre-chosen before optimization. We

82
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Link di θi ai αi
1 0 q1 0 90◦
2 0 90◦ + q2 0 90◦
3 0 q3 l1 0◦

x1
x0
z3 y0 z2

z0 y1 z1 Link 1
l1
x3 y3 x2
z3 y2

x3 y3 Link 2

Link 3

Fig. 5. The coordinate frames of the links and the Denavit–Hartenberg parameters of the assigned coordinate frames.

Table 4
Base inertial parameters of the robot manipulator and identification results.

p1 m3 rc3, y 0.0163 kg · m p13 Ixx 2 + Ixx 3 − Izz 2 −0.0045 kg · m2

p2 m3 rc3, x 0.0000 kg · m p14 Ixx 2 + Iyy3 − Izz 2 −0.0053 kg · m2
p3 m1 rc1, x 0.0152 kg · m p15 Ixz 2 0.0001 kg · m2
p4 Ixy3 0.0000 kg · m2 p16 Iyy1 + Izz 2 0.0429 kg · m2
p5 Ixz3 0.0000 kg · m2 p17 Fv2 0.0791 N · m · s / rad
p6 m2 rc 2, x 0.0723 kg · m p18 Fv3 0.2349 N · m · s / rad
p7 Iyz 2 0.0000 kg · m2 p19 Fv1 1.0532 N · m · s / rad
p8 m2 rc 2, y + m1 rc1, z −0.2579 kg · m p20 Iyy2 − Ixx 2 + Izz 2 0.0110 kg · m2
p9 Ixy2 −0.0012 kg · m2 p21 Fc3 0.0267 N · m
p10 Iyz3 0.0002 kg · m2 p22 Fc 2 0.1045 N · m
p11 m2 rc 2, z + m3 rc3, z 0.0820 kg · m p23 Fc1 0.5521 N · m
p12 Izz3 0.0006 kg · m2 p24 m3 0.3744 kg

Fig. 6. The optimal trajectories for the identification of the inertial

Trajectory of Link 1
0.4 base parameters.
rad, rad/s, rad/s 2

position
velocity
0.2
acceleration

-0.2
0 1 2 3 4 5 6 7 8 9 10
Time (sec)
Trajectory of Link 2 Trajectory of Link 3
2
rad, rad/s, rad/s

2
1
1
0 0
-1
-1
-2
-2
0 5 10 0 5 10
Time (sec) Time (sec)

83
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Table 5 Md (x¨ − x¨d ) + Bd (x˙ − x˙ d ) + K d (x − x d ) = −Fe (9)

Control modes of the hybrid impedance control according to the type of exercises.
where Md ∈ 6 × 6,
Bd ∈ 6 × 6,
and K d ∈ 6 × 6
are symmetrical matrices
Exercise type Control mode Switch (s) that denote the desired inertia, damping, and stiffness matrices, re-
spectively. The vector x denotes the end-effector position and orienta-
Passive Position based impedance 1
tion, and xd denotes the desired end-effector position and orientation.
Active-assistive Position and force based impedance (switching) 1/0
Isometric Position and force based impedance (switching) 1/0 The Eq. (9) is arranged as
Isotonic Force based impedance 0
Isokinetic Force based impedance 0
x¨ = x¨d + Md−1 [−Bd (x˙ − x˙ d ) − K d (x − x d ) − Fe]. (10)
Robo-resistive Force based impedance 0
The velocity of the end-effector is
x˙ = J (q) q˙ (11)
chose ωf = 0.1 and Ni = 5. The
and the acceleration is
We solved the optimization problem with fmincon function of
MATLAB. The resulting optimal trajectories for each link can be seen in x¨ = J (q) q¨ + J˙ (q) q˙ . (12)
Fig. 6. In order to identify parameters, the matrix A and vector b in (3)
Let us consider the robot dynamics in (1). If we choose the control
are calculated by using those trajectories.
input as
The matrix A was obtained by using the position, velocity, and ac-
celeration data in these trajectories as we knew qi, q̇i and q̈i during τ = M (q) u + C (q, q˙ ) + G (q) + F (q˙ ) − J (q)T Fe , (13)
motion. It was provided that the robot follows the optimal trajectories
then the dynamic equation becomes q¨ = u. The Eq. (12) yields
by using the tuned PID controllers. The torques applied by the motors
while the robot follows the optimal trajectories were saved during the J (q)† (x¨ − J˙ (q) q˙ ) = u, (14)
experiments. Thus, the matrix b in (3) was obtained. The parameters of
where (·)† denotes Moore Penrose inverse of a matrix. In order to get
the system were identified by using the linear least square estimation
the desired dynamics in (10), u can be written as
method below:
u = J (q)† (x¨d + Md−1 [−Bd (x˙ − x˙ d ) − K d (x − x d ) − Fe] − J˙ (q) q˙ ). (15)
p = (AT A)−1AT b. (8)
The resulting control law after combining the Eqs. (13) and (15)
The dynamic equations and the identified parameters are used to becomes
implement the hybrid impedance control. The estimated parameters
can be seen in Table 4. τ = M (q) J (q)† (x¨d + Md−1 [−Bd (x˙ − x˙ d ) − K d (x − x d ) − Fe] − J˙ (q) q˙ )
+ C (q, q˙ ) + G (q) + F (q) − J (q)T Fe .
4. Control strategy (16)

The impedance control method is widely used in the rehabilitation

4.2. Force based impedance control
systems. The most important reason is that it is the most suitable con-
trol method for the human–robot interaction [32,33]. Different versions
For the force based impedance control case, the desired dynamics
of the impedance control are available in the literature [41]. One of
behaviour of the system can be given as:
those methods is the hybrid impedance control method [28]. In this
method, conventional impedance control and hybrid position or force Md x¨ + Bd x˙ − Fd = −Fe (17)
control methods are combined into a single structure. Thus, the desired which is equal to
mechanical impedance of the robot end-effector can be adjusted while
the robot is tracking the desired force or position trajectory. x¨ = Md−1 (−Bd x˙ + Fd − Fe ) (18)
The therapeutic exercises require the force or position control.
where Fd ∈ 6 × 1
is the desired force. From (13), (14) and (18), we
Furthermore, the active assistive exercise requires both, e.g., the posi-
deduce the following control law:
tion control if the patient needs help, and the force control otherwise.
Instead of the force and position based two separate controllers for τ = M (q) J (q)† (Md−1 [Fd − Fe − Bd x˙ ] − J˙ (q) q˙ ) + C (q, q˙ ) + G (q) + F (q)
modeling the therapeutic exercises, employing a single control struc- − J (q)T Fe . (19)
ture through the hybrid impedance controller is more effective and
flexible. Hence, exercises are modeled by using the hybrid impedance
control approach in this study. 4.3. Hybrid impedance control
These modes are classified according to the exercise types as shown
in Table 5. The details of the hybrid impedance control are explained in The hybrid impedance control includes both force based and posi-
the following sections. tion based impedance control with a switching matrix S. This matrix is a
The hybrid impedance control is a parameter dependent control as it diagonal matrix with entries either 1 or 0 and used to determine which
includes inverse dynamics and cancellation of some terms. degrees of freedom of the end effector is controlled by either position
Uncertainties in the system parameters lead to deterioration between based or force based impedance control, respectively. The desired dy-
the desired and resulting impedances of the system. In the study, the namics behaviour of the system is a combination of (16) and (19) with
system parameters are estimated to achieve a better performance. Note the switching matrix:
that there are robust impedance control schemes in the literature to Md (x¨ − Sx¨d ) + Bd (x˙ − Sx˙ d ) + SK d (x − x d ) + (I − S ) Fd = −Fe . (20)
decrease the effect of parametric uncertainties on impedance control
performance. Readers can refer to [42,43] for further information. The following control law is applied

τ = M (q) J (q)† (Sx¨d + Md−1 [(I − S ) Fd − Fe − Bd (x˙ − Sx˙ d ) − SK d (x − x d )]

4.1. Position based impedance control
− J˙ (q) q˙ ) + C (q, q˙ ) + G (q) + F (q) − J (q)T Fe

The desired dynamic behaviour of the robot manipulator after ap- (21)
plying position based impedance control can be given as: to obtain the desired dynamics. In (21), I denotes the identity matrix.

84
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

4.4. Selecting the controller parameters to model the exercises ∼],

There are four different parameters which are Md, Bd, Kd, S. By
adjusting those parameters, the impedance of the mechanic interaction
between end-effector of the robot manipulator and the patient is de-
termined. These parameters are adjusted with respect to exercise and
wrist movement types.
The impedance parameters are 6 × 6 matrices. All of the matrices
are chosen as diagonal matrices. The first three elements are related to
translational motions along x0, y0 and z0 and the last three elements are
related to rotational motions about x0, y0 and z0 axes, respectively.
• Flexion–Extension: During the flexion–extension movement, the
origin of the coordinate system of the end effector moves in the y0
and z0 axes and rotates about the x0 axis (see Fig. 5). End-effector
does not make a translational movement along x0 axis and does not
rotate around y0 and z0 axes. In order to prevent movement along x0
axis, the desired impedance parameters are chosen as very high
values at the corresponding axis, e.g., 100 kg for desired mass. Also,
to prevent the translational motion along x0 axis, the controller is set
to position-based impedance control with a desired trajectory 0° for
the corresponding axis. The desired parameters for this movement is
selected as:
Md = diag[100, m, m, m ∼, m ∼, 1],
Bd = diag[1000, b, b, b͠ , b͠ , 10],
K d = diag[10000, k, k, k͠ , k͠ , 100],
S = diag[1, s, s, s, 1, 1],
Md = diag[100, 100, 100, 1, 1, m
Bd = diag[1000, 1000, 1000, 10, 10, b͠ ],
K d = diag[10000, 10000, 10000, 100, 100, k͠ ],
S = diag[1, 1, 1, 1, 1, s].
5. Experimental studies
The experimental studies were performed with voluntary subjects
and voluntary patients. The performances of implementing the ther-
apeutic exercises by means of the rehabilitation robotic system and the
developed HMI were exhibited by the tests performed with the healthy
subjects. The performance of the system in therapeutic exercises re-
vealed that the system can be tested by real patients. In the second
stage, the system was tested by performing therapeutic exercises, which
are determined by an attending physician and correct for the patient’s
difficulty, with voluntary patients. The tests were performed in a clin-
ical environment under the supervision of the physician.
The passive, active assistive, isotonic and isometric exercises were
performed by three healthy subjects. The performance of the system
was evaluated by the parameters fit to the exercise type. The attending
physician had four voluntary patients, who have nerve lacerations,
perform the therapeutic exercises. The isometric and isotonic exercises
were chosen according to the conditions of the patients. The procedures
and results of these experiments are discussed in detail in the following
sections.

The operator diag[x1, x2 , …, x n] denotes a block diagonal matrix
whose elements on the main block diagonal are x1, x2, ⋅⋅⋅, xn. Here, m
is the mass, m∼ is the moment of inertia, b is the coefficient of linear
damping, b͠ is the coefficient of rotational damping, k is the stiffness,
k͠ is stiffness for the rotational spring and s is the switch value of the
impedance controller. Those parameters are adjusted to determine
the difficulty level of the exercise whereas s is adjusted to model the
exercise types. Selection of the parameter s with respect to exercise
type can be seen in Table 5. The impedance parameters corre-
sponding to the difficulty levels are presented in Table 6. Non-
parametric values are used to prevent the motion except for the
flexion–extension trajectory.
• Ulnar–Radial deviation: During this movement, the origin of the
coordinate system of the end-effector moves in the x0 and z0 axes
and rotates about the y0 axis. To model this movement, selection of
the parameters are given below:
∼, 1],
Md = diag[m, 100, m , 1, m
Bd = diag[b, 1000, b, 10, b͠ , 10],
K d = diag[k, 10000, k, 100, k͠ , 100],
S = diag[s, 1, s, 1, s, 1].
5.1. Experiments with healthy subjects
The therapeutic exercise implementation capacity of the robotic
rehabilitation system was tested with three voluntary healthy subjects.
The physical properties of the healthy subjects are given in Table 7. In
the selection of the subjects, similarity in the physical properties was
considered. The experiments were performed for the movements of the
flexion–extension, the ulnar–radial deviation, and the pronation–supi-
nation. The passive, active assistive, isotonic, and isometric exercises
were performed through these movements. The subjects grasped the
handle and their forearms were fastened to the robot. The input data
according to the type of exercise was entered through the GUI. All
movements were repeated 10 times. The force and position tracking
errors of the subjects were evaluated by means of the performance
measures: The mean value, the standard deviation and the RMS. The
performed experiments and the results are presented in the following
sections in detail.
5.1.1. Passive exercise
In the passive exercise, the robot manipulator makes the patient’s
limb move within the ROM. The position trajectory of the robot ma-
nipulator is determined by the HMI through the entered information of

• Pronation–Supination: During pronation–supination movement,

the motion limits and the speed. The hybrid impedance controller op-
erates in the position control mode. In Table 8, the result of the passive
end-effector only rotates about the z0 axis.
exercise is given and an example of graphical results is shown in Fig. 7.
As shown in Table 8, the mean values of the manipulator’s position

Table 6
Impedance parameter values used in the experiments for different difficulty levels. Table 7
Physical properties of the healthy subjects.
∼)
Difficulty level (m) (m (b) (b͠ ) (k) (k͠ )
Subject Sex Age Weight Height
Low 1.25 0.0125 10 0.1 100 1
Medium 2.50 0.0250 20 0.2 200 2 A Male 23 75 kg 176 cm
High 5.00 0.0500 40 0.4 400 4 B Male 22 70 kg 180 cm
Very high 7.50 0.0750 80 0.8 800 8 C Male 24 80 kg 178 cm

85
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Table 8
Position tracking error results for passive exercise.

Motion Subject ROM (°) Period (sec) Repetition Position tracking error

Mean(%) Std. Dev.(%) RMS(%)

Extension A 60 2 10 0.677 ± 0.762 1.019

B 0.906 ± 0.870 1.256
C 0.785 ± 0.823 1.138
Flexion A 0.672 ± 0.120 1.307
B 0.364 ± 0.735 0.821
C 0.760 ± 0.841 1.134
Radial deviation A 20 2 10 1.094 ± 0.868 1.397
B 1.136 ± 0.877 1.435
C 1.158 ± 0.849 1.436
Ulnar deviation A 40 1.149 ± 1.095 1.587
B 1.068 ± 0.908 1.402
C 1.074 ± 0.891 1.395
Pronation A 60 4 10 0.672 ± 0.443 0.805
B 0.769 ± 0.541 0.940
C 0.805 ± 0.603 1.006
Supination A 0.188 ± 0.346 0.394
B 0.005 ± 0.445 0.448
C 0.302 ± 0.511 0.594

60 Fig. 7. Passive flexion exercise result for the healthy subject B.

Position ( ° )

40

20
Actual
Desired

1
Error ( ° )

-1
0 2 4 6 8 10 12 14 16 18 20
Time (sec)

Table 9
Position tracking error results for active assistive exercise.

Motion Subject ROM (°) Period (sec) Repetition Position tracking error

Mean(%) Std. Ddev.(%) RMS(%)

Extension A 60 2 10 0.865 ± 2.583 2.725

B 0.332 ± 2.801 2.820
C 0.640 ± 2.951 3.020
Flexion A 0.782 ± 1.523 1.594
B 0.544 ± 1.605 1.695
C 0.483 ± 1.672 1.742
Radial deviation A 20 2 10 0.016 ± 0.370 0.370
B 0.051 ± 0.308 0.312
C 0.027 ± 0.278 0.280
Ulnar deviation A 40 0.229 ± 0.711 0.747
B 0.162 ± 0.620 0.641
C 0.171 ± 0.656 0.679
Pronation A 60 4 10 0.275 ± 0.847 0.891
B 0.153 ± 0.897 0.910
C 0.271 ± 0.875 0.916
Supination A 0.260 ± 0.881 0.919
B 0.214 ± 0.826 0.854
C 0.232 ± 0.819 0.851

86
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Fig. 8. The active assistive pronation exercise result for the healthy
0
Subject C. This figure shows (from top to bottom) the desired and
Position (°)
actual(subject) position trajectory, the force trajectory of subject and
-20 the switching signal of the hybrid impedance controller. The subject
could not move his limb in the second and seventh seconds. The
switching signal was activated in these moments to help the subject
-40 Actual to complete his motion.
Desired
-60

15

10
Force (N)

-5
Assistance

Active
Passive

0 1 2 3 4 5 6 7 8 9
Time (sec)

Table 10
Isometric exercise experiments on the healthy subjects.

Motion Subject Ref. force (N) Repetition Pat. force Force Tracking Error

Mean(%) Std. Dev.(%) RMS(%)

Extension A 10 10 9.660 0.275 ± 0.847 0.891

B 9.638 0.153 ± 0.897 0.910
C 9.480 0.271 ± 0.875 0.916
Flexion A 9.008 0.260 ± 0.881 0.919
B 9.061 0.214 ± 0.826 0.854
C 8.862 0.232 ± 0.819 0.873
Radial deviation A 10 10 10.854 12.632 ± 10.710 16.561
B 10.492 10.212 ± 7.389 12.605
C 10.834 13.604 ± 12.526 18.492
Ulnar deviation A 10.228 5.155 ± 12.771 13.772
B 10.368 6.071 ± 19.583 20.502
C 9.136 8.181 ± 19.627 21.263
Pronation A 10 10 9.675 0.258 ± 4.762 4.769
B 9.879 0.329 ± 5.262 5.272
C 9.837 0.325 ± 4.219 4.231
Supination A 9.866 0.972 ± 5.681 5.763
B 9.451 1.106 ± 7.237 7.320
C 9.327 1.070 ± 4.599 4.722

tracking error are under 0.8% for the extension–flexion, 1.16% for the
ulnar–radial deviation, and 0.8% for the pronation–supination. It is
evaluated by considering all the movements that the robot manipulator
was able to perform the passive exercise under 1.2% performance. In
parallel with this, the robot manipulator was able to track the desired
trajectory with an error under one degree, as shown in Fig. 7. It is
understood that the system can perform the passive exercises.
5.1.2. Active assistive exercise
This exercise is described that the patient moves his or her limb
actively and a physiotherapist assists him or her to complete the
movement from the point where the patient is not able to move his or
her limb further. Modeling the exercise in the system is as follows: The
hybrid impedance controller operates in the force control mode, if the
subject can move his or her limb. The position control mode is switched
from the point where the limb cannot be moved. The decision for
switching between these modes is done by continuously checking the
variation in the position. The switching is performed if the robotic arm
is stationary or started to move in the opposite direction. The entered
information are the target position, the type of movement, the speed,
and the exercise duration. The subject is asked to grasp and move the
handle till a certain point, where the subject is asked not to move his or
her joint. From the point where the limb is stationary, the manipulator
tracks the predetermined trajectory with high accuracy and return to
the start position. This measurement was repeated 10 times. The results
obtained from all the subjects are given in Table 9. The mean values of
the manipulator’s position tracking error are under 0.865% for the
extension–flexion, 0.229% for the ulnar–radial deviation, and 0.275%
for the pronation–supination. To give an example, the experiment result
for the pronation movement of the Subject C, where the ROM was
chosen as 60°, is given in Fig. 8. As it can be seen from the figure, the
subject moved his limb by applying force at the seconds 1–2 and 5–7.
When the subject was not able to move his limb at the seconds 2 and 7,
the decision signal became active and the robot manipulator tracked
the reference trajectory after switching from the force control mode to
the position control mode. It is understood that the system can perform

87
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

5 Fig. 9. The isometric flexion exercise result for the healthy subject A.
This figure shows (from top to bottom) the desired and actual(subject)
force trajectory, the limb position trajectory of the subject. When the
subject stops his limb (3rd and 12th seconds), the robot applies an
0
external force (10 N) to the limb of the subject. When the subject stops
his limb at 40° (3rd and 12th seconds), the robot applies an external
Force (N)

force (10 N) to the limb of the subject. The subject tries to keep his
-5
limb position constant.

-10

-15 Actual
Desired

-20

40
Position (°)

20

0
0 2 4 6 8 10 12 14 16 18 20
Time (sec)

Table 11
Isotonic exercise experiments on healthy subjects.

Motion Subject Ref. force (N) Repetition Force (N) Force tracking error

Mean(%) Std. Dev.(%) RMS(%)

Extension A 10 10 11.468 14.688 ± 15.984 21.708

B 11.333 13.333 ± 15.496 20.442
C 11.670 16.709 ± 18.757 25.120
Flexion A 11.893 18.941 ± 24.176 30.712
B 11.680 16.808 ± 17.434 24.216
C 11.506 15.062 ± 18.482 23.842
Radial deviation A 10 10 10.598 5.982 ± 17.847 18.822
B 10.744 7.446 ± 16.995 18.555
C 10.483 4.832 ± 16.954 17.629
Ulnar deviation A 11.095 10.952 ± 20.055 22.851
B 11.605 16.053 ± 24.060 28.923
C 12.151 21.518 ± 27.692 35.069
Pronation A 10 10 11.881 18.818 ± 18.829 26.620
B 11.822 18.226 ± 21.722 28.355
C 12.050 20.509 ± 21.019 29.367
Supination A 11.910 19.103 ± 19.190 27.077
B 12.289 22.890 ± 21.790 31.603
C 11.955 19.557 ± 21.307 28.921

the active assistive exercises.
5.1.3. Isometric exercise
The purpose of the isometric exercise is to cause a change in the
muscle contraction and not in the length of the muscle. This exercise is
modeled as follows: The hybrid impedance controller is in the force
control mode. The subject starts to move his or her limb. When the
motion of the limb stops, the robot manipulator applies an opposite
force to the limb and the subject tries to oppose this force. The subject
was asked to grasp and move the handle in the isometric exercise. The
robot applied a predetermined and constant force to the subject in the
opposite direction at the point where the limb becomes stationary. This
movement was repeated 10 times.
In Table 10, the results for the force tracking errors of the subjects
are given according to the types of movements. The mean values of the
manipulator’s position tracking error are under 0.275% for the ex-
tension–flexion, 13.6% for the ulnar–radial deviation, and 1.1% for the
pronation–supination. It is meaningful that the errors are high in the
movement of the ulnar–radial deviation for all the patients, since the
experiments exhibited that the relevant muscles are difficult to control
against constant forces. To give an example, the experiment result with
the Subject A is given in Fig. 9. In that experiment, the Subject A was
asked to reach 40° ROM and to oppose the predetermined 10 N force
generated by the robot manipulator. As shown in the bottom plot in
Fig. 9, the subject moved his wrist from 0° to 40° together with the
manipulator and stopped at 40°. His stop was detected by the HMI and
an opposing 10 N force was applied by the manipulator at the seconds 3
and 12. In this time interval, the subject performs the isometric exercise
against 10 N force at a constant position. It is understood that the
manipulator can perform the isometric exercises.
5.1.4. Isotonic exercise
In this exercise, the subject moves his or her limb while the robot
manipulator generates an opposite force. The subject does work by

88
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

15 hand trauma and completing the post-traumatic immobilization period

10 and post-operative time more than 6 weeks and not longer than 3
Force (N)

months were included in the experiments. Cases involving the forearm,

5
wrist or finger fractures, non-healing wounds, sensation of pain pre-
0 Actual venting the exercise, were excluded from the experiments.
Desired
After the patients arrived at the clinic, first, the attending physician
-5
measured the squeezing force by the dynamometer and the ROM. The
0 dynamometer data of the patients before and after the therapy are given
Position ( ° )

in Table 14. Two pictures from the treatment process are shown in
-10
Fig. 11. All patients had full ROM and needed to strengthen their
-20 muscles. Therefore, isotonic and isometric exercises were applied to
patients for the flexion–extension movement.
-30
A four-step procedure, whose flowchart given in Fig. 12, were ap-
plied in each session (two sessions per week) of the rehabilitation. The
Decision

Active
procedure is explained in detail as follows:
Passive

0 2 4 6 8 10 12 14 16 Step 1-(Measurement of the ROM): In order to measure the ROM,

Time (sec) the patient moves the robot manipulator to his or her maximum
ROM. The maximum ROM in a session is measured by the position
Fig. 10. The isotonic radial deviation exercise result for the healthy subject B.
sensors of the robot and the data is saved to the database. If the ROM
is not enough, the treatment continues with the passive exercise.
Table 12 However, this exercise was not needed since the values of the ROM
Personal information of patients.
of the patients in the experiment group were normal.
Patient 1 Patient 2 Patient 3 Patient 4 Step 2-(Measurement of the isometric joint force): The patient
applies the maximum force 10 times for the isometric measure-
R.O.M.(deg) Full Full Full Full ments. In this case, the robot manipulator is fixed at zero degree
Sex Female Male Male Male
position. The maximum forced applied by the patient in a session is
Job Housewife Security staff Driver Unemployed
Age 46 33 33 36 calculated by taking the average of 10 force measurements and the
Treated hand Right Left Left Right data is saved to the database.
Step 3-(Isometric exercise): This exercise is performed in three
stages: In the first stage, the patient performs the isometric exercise
resisting that opposite force. The hybrid impedance controller operated 10 times by applying the 50% of the maximum force, which was
in the force mode. The subject was asked to grasp and move the handle. calculated in the previous step. In the second and third stages, the
The robot resisted to the motion of the subject by applying the pre- exercise repeated 10 times for both 75% and 100% of the maximum
determined force. The exercise repeated 10 times. In Table 11, the re- force, respectively.
sults for the force tracking errors of the subjects are given according to Step 4-(Isotonic exercise): This exercise is performed in three
the types of movement. The mean values of the force tracking error are stages: The isotonic exercise is performed according to the values of
under 19% for the extension–flexion, 21.5% for the ulnar–radial de- the maximum ROM and the maximum isometric joint force in the
viation, and 23% for the pronation–supination. To give an example, the current session. These values are entered from the GUI manually. In
isotonic experiment result with the Subject B for the movement of ra- the first stage, the patient performs the isotonic exercise 10 times by
dial deviation is given in Fig. 10. It is understood that the manipulator applying the 50% of the maximum force, which was calculated
can perform the isotonic exercises. previously, and trying to reach the maximum ROM. In the second
and third stages, as in the first stage, the exercises repeated 10 times
5.2. Experiments with patients for both 75% and 100% of the maximum force. The patient tries to
follow the red ball in the game screen.
The experiments with four voluntary patients, who had muscle force
deficiency problem due to peripheral nerve lacerations, were performed The clinical test results of the patients are given in Table 13. In this
in Sisli Hamidiye Etfal Training and Research Hospital, The Clinic of table, the maximum joint forces exerted by the patients in the directions
Physical Therapy and Rehabilitation, Istanbul, Turkey. The experiments of flexion and extension are given. Treatments were ended after 13th
were approved by the Ethical Committee of Istanbul University. The session for Patient 1, 7th session for both patients 2 and 4, and after 5th
experimental procedures conform with the Declaration of Helsinki. The session for the Patient 3. All the treatments ended by doctor when she
information about the patients are given in Table 12. Cases involving

Fig. 11. The pictures from the treatment process.

89
E. Akdoğan et al. Mechatronics 49 (2018) 77–91

Patient comes to hospital Save

Measurements

Isometric Exercises with Isotonic Exercises with

New NO Measurement of ROM YES Isometric
50%, 75% and 100% of 50%, 75% and 100% of
Patient ? Range of Motion enough? Measurements
saved measurements saved measurements

YES
NO

Passive Exercises
Enrollment Need to Physiotherapist YES
Exercise Finish completed
NO continue? checks results
succesfully ?

YES NO
Scheduling to
next session

Fig. 12. The flowchart of the treatment process.

Table 13
Clinical test results.

Session Patient 1 Patient 2 Patient 3 Patient 4

Flexion force(N) Extension force(N) Flexion force(N) Extension force(N) Flexion force(N) Extension force(N) Flexion force(N) Extension force(N)

1 20 10 18 30 33 20 15 25
2 18 9 23 38 35 28 19 26
3 20 16 23 37 55 30 16 25
4 22 18 35 39 50 40 23 31
5 24 20 30 35 55 35 24 28
6 27 18 36 40 40 36
7 28 18 38 44 44 38
8 30 17
9 32 20
10 30 21 The patient has reached the sufficient level of the muscle activity
11 34 23
12 33 23
13 30 20
Mean 26.8 17.9 29.0 37.6 45.6 30.6 25.9 29.9
Std Dev. 5.4 4.3 7.7 4.4 10.8 7.5 11.6 5.3
Max 34 23 38 44 55 40 44 38

Table 14 which were carried out with healthy subjects. Then, experiments were
The dynamometer data of the patients.a carried out with patients in a clinical environment. As a result of these
experiments, the increase in muscle strength was observed in all the
Patient 1 Patient 2 Patient 3 Patient 4
patients. In the next study, an adaptive impedance controller will be
B.T. A.T. B.T. A.T. B.T. A.T. B.T. A.T. developed for individuals and this controller will be tested with pa-
tients.
Right hand 9.0 18.0 35.0 36.0 39.0 38.0 2.0 9.5
Left hand 20.0 19.0 7.0 16.5 9.5 17.0 27.0 28.0
Acknowledgment
a
B. T. = Before Treatment, A. T. = After Treatment. The units are in kilogram.
This work was supported by the Scientific and Technological
decided that the force exerted by the patients will not increase. As can Research Council of Turkey (TUBITAK) under grant number 111M603.
be seen from Table 13, at the end of the exercises, it was observed that
the flexion–extension isometric muscle force values of all the patients References
increased. The grasping forces of the patients before and after treatment
are given in Table 14. Significant improvements were observed in the [1] Akdemir N, Akkuş Y. Rehabilitasyon ve hemşirelik (derleme). Hemşirelik
Yüksekokulu Dergisi 2006;13(1):82–91.
grasping forces of the problematic hands of all patients, which are [2] Sosyal Güvenlik Kurumu TC. Fizik tedavi ve rehabilitasyon tanı listesi. 2010 Yılı
highlighted in the corresponding table. The clinical results show that Sağlık Uygulama Tebliği 2010;EK-9/A.
the developed robotic rehabilitation system had a positive contribution [3] Krebs HI. An overview of rehabilitation robotic technologies. in: American spinal.
Injury association symposium. 2006.
to the treatment process. [4] Akdoğan E. Rehabilitasyon amaçl bir robot kolunun kuvvet ve konumunun zeki
kontrol. Ph.D. thesis. Marmara University, İstanbul, Turkey; 2007.
6. Conclusions [5] Krebs HI, Hermano I, Hogan N, Aisen ML, Volpe BT. Robot-aided neurorehabilita-
tion. IEEE Trans Rehabil Eng 1998;6(1):75–87.
[6] Krebs HI, Volpe BT, Williams D, Celestino J, Charles SK, Lynch D, et al. Robot-aided
In this study, a hybrid impedance controller was developed to model neurorehabilitation: a robot for wrist rehabilitation. IEEE Trans Neural Syst Rehabil
the therapeutic exercises. A 3-DOF upper limb rehabilitation robot was Eng 2007;15(3):327–35.
[7] Kahn LE, Averbuch M, Rymer WZ, Reinkensmeyer DJ. Comparison of robot-assisted
controlled by the developed controller. The capacity of robot in per- reaching to free reaching in promoting recovery from chronic stroke. Proceedings of
forming the therapeutic exercises was presented through the tests,

90
E. Akdoğan et al.
the international conference on rehabilitation robotics. 2001. p. 39–44.
[8] Toth A, Fazekas G, Arz G, Jurak M, Horvath M. Passive robotic movement therapy
of the spastic hemiparetic arm with reharob: report of the first clinical test and the
follow-up system improvement. Rehabilitation robotics, 2005. ICORR 2005. 9th
international conference on. IEEE; 2005. p. 127–30.
[9] Nef T, Riener R. Armin-design of a novel arm rehabilitation robot. Rehabilitation
robotics, 2005. ICORR 2005. 9th international conference on. IEEE; 2005. p. 57–60.
[10] Sanchez Jr R, Wolbrecht E, Smith R, Liu J, Rao S, Cramer S, et al. A pneumatic robot
for re-training arm movement after stroke: rationale and mechanical design.
Rehabilitation robotics, 2005. ICORR 2005. 9th international conference on. IEEE;
2005. p. 500–4.
[11] Denève A, Moughamir S, Afilal L, Zaytoon J. Control system design of a 3-dof upper
limbs rehabilitation robot. Comput Methods Programs Biomed 2008;89(2):202–14.
[12] Moubarak S, Pham MT, Pajdla T, Redarce T. Design and modeling of an upper
extremity exoskeleton. World congress on medical physics and biomedical en-
gineering, September 7–12, 2009, Munich, Germany. Springer; 2009. p. 476–9.
[13] Culmer PR, Jackson AE, Makower S, Richardson R, Cozens JA, Levesley MC, et al. A
control strategy for upper limb robotic rehabilitation with a dual robot system.
Mechatron IEEE/ASME Trans 2010;15(4):575–85.
[14] Colizzi L, Laneve L, Savino N, Martini A, Potenza A, Cirillo P, et al. Aramis: a virtua-
mechatronic approach for neuro-rehabilitation purposes. Information technology
and applications in biomedicine (ITAB), 2010 10th IEEE international conference
on. IEEE; 2010. p. 1–4.
[15] Kiguchi K, Kose Y, Hayashi Y. An upper-limb power-assist exoskeleton robot with
task-oriented perception-assist. Biomedical robotics and biomechatronics (BioRob),
2010 3rd IEEE RAS and EMBS international conference on. IEEE; 2010. p. 88–93.
[16] Kiguchi K, Hayashi Y. An emg-based control for an upper-limb power-assist exos-
keleton robot. Syst Man Cybern Part B: Cybern IEEE Trans 2012;42(4):1064–71.
[17] Mao Y, Agrawal SK. Design of a cable-driven arm exoskeleton (carex) for neural
rehabilitation. Rob IEEE Trans 2012;28(4):922–31.
[18] Hsu L-C, Wang W-W, Lee G-D, Liao Y-W, Fu L-C, Lai J-S. A gravity compensation-
based upper limb rehabilitation robot. American control conference (ACC), 2012.
IEEE; 2012. p. 4819–24.
[19] Frisoli A, Loconsole C, Leonardis D, Banno F, Barsotti M, Chisari C, et al. A new
gaze-bci-driven control of an upper limb exoskeleton for rehabilitation in real-world
tasks. Syst Man Cybern Part C: Appl Rev IEEE Trans 2012;42(6):1169–79.
[20] Ren Y, Kang SH, Park H-S, Wu Y-N, Zhang L-Q. Developing a multi-joint upper limb
exoskeleton robot for diagnosis, therapy, and outcome evaluation in neuroreh-
abilitation. Neural Syst Rehabil Eng IEEE Trans 2013;21(3):490–9.
[21] Kim H, Miller LM, Fedulow I, Simkins M, Abrams GM, Byl N, et al. Kinematic data
analysis for post-stroke patients following bilateral versus unilateral rehabilitation
with an upper limb wearable robotic system. Neural Syst Rehabil Eng IEEE Trans
2013;21(2):153–64.
[22] Wei W, Guo S, Zhang W, Guo J, Wang Y. A novel vr-based upper limb rehabilitation
robot system. Complex medical engineering (CME), 2013 ICME international con-
ference on. IEEE; 2013. p. 302–6.
[23] Lum PS, Burgar CG, Van der Loos M, Shor PC, Majmundar M, Yap R. The mime
robotic system for upper-limb neuro-rehabilitation: results from a clinical trial in
subacute stroke. Rehabilitation robotics, 2005. ICORR 2005. 9th international
conference on. IEEE; 2005. p. 511–4.
[24] Lewis GN, Perreault EJ. An assessment of robot-assisted bimanual movements on
upper limb motor coordination following stroke. Neural Syst Rehabil Eng IEEE
Trans 2009;17(6):595–604.
[25] Loureiro R, Amirabdollahian F, Topping M, Driessen B, Harwin W. Upper limb
robot mediated stroke therapy gentle/s approach. Auton Robots 2003;15(1):35–51.
[26] Sukal TM, Ellis MD, Dewald J. Dynamic characterization of upper limb dis-
coordination following hemiparetic stroke. Rehabilitation robotics, 2005. ICORR
2005. 9th international conference on. IEEE; 2005. p. 519–21.
[27] Stienen AH, McPherson JG, Schouten AC, Dewald J. The act-4d: a novel re-
habilitation robot for the quantification of upper limb motor impairments following
brain injury. Rehabilitation robotics (ICORR), 2011 IEEE international conference
on. IEEE; 2011. p. 1–6.
[28] Anderson RJ, Spong MW. Hybrid impedance control of robotic manipulators. Rob
Autom IEEE J 1988;4(5):549–56.
[29] Wang J, Li Y. Massaging human feet by a redundant manipulator equipped with a
tactile sensor. Advanced intelligent mechatronics (AIM), 2010 IEEE/ASME inter-
national conference on. IEEE; 2010. p. 7–12.
[30] Talasaz A, Patel RV, Naish MD. Haptics-enabled teleoperation for robot-assisted
tumor localization. Robotics and automation (ICRA), 2010 IEEE international
conference on. IEEE; 2010. p. 5340–5.
[31] Wang J, Li Y. Hybrid impedance control of a 3-dof robotic arm used for
91
Mechatronics 49 (2018) 77–91
rehabilitation treatment. Automation science and engineering (CASE), 2010 IEEE
conference on. IEEE; 2010. p. 768–73.
[32] Akdoğan E, Taçgın E, Adli MA. Knee rehabilitation using an intelligent robotic
system. J Intell Manuf 2009;20(2):195–202.
[33] Akdoğan E, Adli MA. The design and control of a therapeutic exercise robot for
lower limb rehabilitation: Physiotherabot. Mechatronics 2011;21(3):509–22.
[34] Yoshikawa T. Foundations of robotics: analysis and control. Mit Press; 1990.
[35] Atlıhan M. Design and hybrid impedance control of an upper limb rehabilitation
robot. Master’s thesis. Yıldız Technical University; 2013.
[36] Atlıhan M, Akdoğan E, Arslan MS. Development of a therapeutic exercise robot for
wrist and forearm rehabilitation. Methods and models in automation and robotics
(MMAR), 2014 19th international conference on. IEEE; 2014. p. 52–7.
[37] Akdoģan E.. Ust uzuv rehabilitasyon robotu, Turkish Patent Institute. 2013. 2013/
07494.
[38] Spong MW, Hutchinson S, Vidyasagar M. Robot modeling and control. 3. Wiley New
York; 2006.
[39] Gautier M. Numerical calculation of the base inertial parameters of robots. Robotics
and automation, 1990. proceedings., 1990 IEEE international conference on. 2.
1990. p. 1020–5. http://dx.doi.org/10.1109/ROBOT.1990.126126.
[40] Swevers J, Ganseman C, Schutter JD, Brussel HV. Experimental robot identification
using optimised periodic trajectories. Mech Syst Signal Process 1996;10(5):561–77.
[41] Zeng G, Hemami A. An overview of robot force control. Robotica
1997;15(05):473–82.
[42] Valency T, Zacksenhouse M. Accuracy/robustness dilemma in impedance control. J
Dyn Syst Meas Control 2003;125(3):310–9.
[43] Ibeas A, De la Sen M. Robust impedance control of robotic manipulators. Decision
and control, 2004. CDC. 43rd IEEE conference on. 2. IEEE; 2004. p. 1258–63.
Erhan Akdoğan received the B.S. degree in electronics and communication engineering
from Yildiz Technical University, in 1999, and the M.S. and Ph.D. degrees from Marmara
University, Turkey, in 2001 and 2007, respectively. From 2008 to 2009, he was a Visiting
Researcher at Hiroshima University Graduate School of Engineering’s Biological Systems
Engineering Laboratory. He is currently an Associate Professor and Biomechatronics
Research Lab. Director at the Mechatronics Engineering Department, Yildiz Technical
University, Istanbul, Turkey. His research interests are biomechatronics, rehabilitation
robotics, cybernetic interface design and artificial intelligence.
Mehmet Emin Aktan received the B.S. degree in mechatronics from Suleyman Demirel
University, in 2010, and the M.S. degree from Marmara University, Turkey, in 2012. He is
currently a Research Assistant at the Mechatronics Engineering Department, Yildiz
Technical University, Istanbul, Turkey. His research interests are rehabilitation robotics,
biomedical device design and manufacturing.
Ahmet Taha Koru received the B.S. and the M.Sc. degrees in Electrical and Electronics
Engineering from Bilkent University, in 2009 and 2012, respectively. He is currently a
Ph.D. candidate in Control and Automation Engineering at Yildiz Technical University.
His research interests are time–delay systems, switching control and haptic systems.
M. Selçuk Arslan received the B.Sc. degree in mechanical engineering from Kirikkale
University, Kirikkale, Turkey, in 2000, the M.Sc. degree in mechanical engineering from
the Middle East Technical University, Ankara, Turkey, in 2005, and the Ph.D. degree in
mechanical sciences and engineering from Tokyo Institute of Technology, Tokyo, Japan,
in 2009. He worked as a post–doctoral researcher at Tokyo Institute of Technology from
2009 to 2011. Since 2011, he is an Assistant Professor in the department of mechatronics
engineering at Yildiz Technical University, Istanbul, Turkey. His research interests are in
the areas of optimal control, vehicle dynamics and control, steer–by–wire systems and
control, control of autonomus underwater vehicles.
Murat Atlıhan received the B.S. degree in Mechatronics Eng. from Kocaeli University and
M.Sc. degree in Mechanical Engineering from Yildiz Technical University, in 2010 and
2013, respectively. He is currently a Ph.D. candidate in Mechatronics Engineering at
Yildiz Technical University. His research interests are control, robotic and magnetic le-
vitation systems.
Banu Kuran received the B.S. degree in medicine from Istanbul University, in 1984, and
the expertise degree in medicine from Sisli Hamidiye Etfal Hospital, in 1989. He is cur-
rently a Professor at the Physical Medicine and Rehabilitation Department, Yeni Yuzyil
University, Istanbul, Turkey.