Content uploaded by Gholamreza Khademi
Author content
All content in this area was uploaded by Gholamreza Khademi on Nov 07, 2017
Content may be subject to copyright.
Robotics and Prosthetics at Cleveland State
University: Modern Information,
Communication, and Modeling Technologies
Yuriy Kondratenko
1,2(&)
, Gholamreza Khademi
1
, Vahid Azimi
1
,
Donald Ebeigbe
1
, Mohamed Abdelhady
1
, Seyed Abolfazl Fakoorian
1
,
Taylor Barto
1
, Arash Roshanineshat
1
, Igor Atamanyuk
2
,
and Dan Simon
1
1
Department of Electrical Engineering and Computer Science,
Cleveland State University, Cleveland, OH, USA
y_kondrat2002@yahoo.com, d.j.simon@csuohio.edu
2
Department of Intelligent Information Systems, Petro Mohyla Black Sea
National University, Mykolaiv, Ukraine
Abstract. This chapter concentrates on the correlation between research-based
education, government priorities and research funding. Special attention is paid
to an analysis of the role of modern information and communication technology
(ICT) in the education of engineering students. Successful cases with specific
description of computer modeling methods for the implementation of prosthesis
and robotics research projects are presented based on experiences in the
Embedded Control Systems Research Laboratory of Cleveland State University.
Keywords: Robotics Prosthetics Modeling Research-based education
1 Introduction
Information and communication technologies (ICT), mathematical modeling, and
computer simulation play a significant role in higher education. Most advanced edu-
cational systems in the world are oriented toward the implementation of educational
processes of modern ICT and software for modelling and simulation in various fields of
human activity, including science, engineering, and technology. This approach is
required for the efficient training of students at various levels: undergraduates, grad-
uates, and doctoral students. Many international conferences on ICT and its applica-
tions for education are devoted to the use of computer modeling, open-source software,
pedagogical e-learning, web-based e-learning, course-centered knowledge management
and application in online learning based on web ontology, on-online learning in
enterprise education, simulation languages, modeling and simulation for education and
training, improving education through data mining, 3D software systems, 3D visual-
ization, wireless communication, experimental teaching of program design, different
approaches in teaching programming, web-based computer-assisted language learning,
and so on.
©Springer International Publishing AG 2017
A. Ginige et al. (Eds.): ICTERI 2016, CCIS 783, pp. 133–155, 2017.
https://doi.org/10.1007/978-3-319-69965-3_8
It is important that university and IT-industry participants of conferences try to find
efficient solutions for the abovementioned computer-modeling-based educational
problems. For example, participants from 178 different academic institutions, including
many from the top 50 world-ranked institutions, and from many leading IT corpora-
tions, including Microsoft, Google, Oracle, Amazon, Yahoo, Samsung, IBM, Apple,
and others, attended the 12th International Conference on Modeling, Simulation and
Visualization Methods, MSV-2015, in Las Vegas, Nevada, USA.
If IT industry today supports higher education, then tomorrow’s IT-based com-
panies, government research agencies, and national laboratories will obtain the
high-quality graduates that they need. New achievements in ICT require continuous
tracking by educators, and implementation in education.
Successful introduction of ICT to higher education based on research-oriented
education and training is considered and analyzed in this chapter. The focus is on the
role of computer modeling and simulation in prosthesis and robotics research for
increasing student quality, including grading their practical skills, and including effi-
cient professor-student interactions.
This chapter is organized as follows. Section 2reviews related literature and dis-
cusses the challenge of integrating research and education. Section 3summarizes
research-based education at Cleveland State University (CSU). Section 4summarizes
seven research-based education projects at CSU. Section 5briefly discusses the com-
mon training, skills, and educational program at CSU that enables the success of
research-based education. Section 6concludes the chapter.
2 Related Works and Problem Statement
Many publications are devoted to teaching methods and approaches based on ICT and
computer modelling, for increasing the efficiency of their interrelation: qualitative
modeling in education [3], computer simulation technologies and their effect on
learning [24], opportunities and challenges for computer modeling and simulation in
science education [34], web-based curricula [4] and remote access laboratories, com-
puter‐based programming environments as modelling tools in education and the
peculiarities of textual and graphical programming languages [17], interrelations
between computer modeling tools, expert models, and modeling processes [44], effi-
cient science education based on models and modelling [9], educational software for
collective thinking and testing hypotheses in computer science [26], and others.
A lot of publications deal with improving teaching efficiency for specific courses by
introducing modern ICT and computer modelling technologies. In particular, modelling
supported course programs, computer-based modelling (AutoCAD, Excel, VBA, etc.)
and computer system support for higher education in engineering [8]; software to
enhance power engineering education [35]; computer modeling for enhancing
instruction in electric machinery [23]; computer modelling in mathematics education
[40]; GUI-based computer modelling and design platforms to promote interactive
learning in fiber optic communications [45]; RP-aided computer modelling for archi-
tectural education [36]; teaching environmental modelling; computer modelling and
134 Y. Kondratenko et al.
simulation in power electronics education [27]; and a virtual laboratory for a com-
munication and computer networking course [22].
Special attention in the literature [5,15,16] is paid to the role of ICT and modeling
technology in education and training in the framework of research-based curricula. This
educational approach deals first with educational directions such as robotics, mecha-
tronics, and biomechanics (RMBM) [12,33,41]. The correlation of RMBM with ICT
and modeling are underlined by results such as: a multidisciplinary model for robotics
in engineering education; integration of mechatronics design into the teaching of
modeling; modelling of physical systems for the design and control of mechatronic
systems [41]; biomechanical applications of computers in engineering education [33];
computerized bio-skills system for surgical skills training in knee replacement [6];
computer modelling and simulation of human movement [25]; computer modelling of
the human hand [19]; and design and control of a prosthesis test robot [29,30].
This chapter builds upon, and extends, the references discussed above. The basic
classroom teaching methods, approaches, and specific courses at CSU are similar to
those at universities across the world. However, those characteristics are not the pri-
mary determinants of research-based education. This chapter presents the features of
research-based education at CSU by reviewing seven specific graduate student-led
research projects. As the research projects are discussed in the following sections, the
reader will note their commonalities, including common tools, research approaches,
motivation, and societal focus. The main aims of this chapter are given as follows.
(a) Description and analysis of research-based education based on the experience in
the Embedded Control Systems Research Laboratory at the Electrical Engineering
and Computer Science Department at the Washkewicz College of Engineering at
Cleveland State University (CSU), USA, with a focus on undergraduate, graduate,
and doctoral student participation in prosthesis and robotics research, which is
funded by the US National Science Foundation (NSF);
(b) Analysis of applied ICT and modeling technologies and advanced software, as
well as their implementation in student research, including course work, diploma
projects, and Doctoral, Master’s, and Bachelor’s theses;
(c) Focus on the correlation between student research and government science pri-
orities based on successful cases of ICT and advanced modelling implementation
in US government-funded prosthesis research, with particular focus on under-
graduate, graduate, and doctoral student participation in prosthesis and robotics
research.
3 Research-Based Education and Government Priority
Project
CSU’s research project “Optimal prosthesis design with energy regeneration”
(OPDER) is funded by the US NSF (1.5 M USD). Professors and students from the
Department of Electrical Engineering and Computer Science, and the Department of
Mechanical Engineering, are involved in research according to the project goals, which
deal with the development of: (a) new approaches for the simulation of human limb
Robotics and Prosthetics at Cleveland State University 135
control; (b) new approaches for optimizing prosthetic limb control, capturing energy
during walking, and storing that energy to lengthen useful prosthesis life; (c) prosthesis
prototype development.
The human leg transfers energy between the knee, which absorbs energy, and the
ankle, which produces energy. The prosthesis that results from this research will mimic
the energy transfer of the human leg. Current prostheses do not restore normal gait, and
this contributes to degenerative joint disease in amputees. This research will develop
new design approaches that will allow prostheses to perform more robustly, closer to
natural human gait, and last longer between battery charges.
This project forms a framework for research-based education. Doctoral, graduate,
and undergraduate students are involved in research such as: the study of able-bodied
gait and amputee gait; the development of models for human motion control to provide
a foundation for artificial limb control; the development of electronic prosthesis con-
trols; the development of new approaches for optimizing prosthesis design parameters
based on computer intelligence; the fabrication of a prosthesis prototype and its test in a
robotic system; the conduct of human trials of the prosthesis prototype.
The role of student participation in all aspects of the research is significant for
increasing their qualifications for their careers, for presentations at conferences, for
publishing in journals, and for research with professors who can help them be more
successful in building their future careers in industry or academia. In the next section
we describe the student contribution to prosthesis and robotics research at CSU.
4 Student Contributions to Prosthesis and Robotics Research
Successful cases of student research in the framework of the OPDER project are
described in this section.
Evolutionary Optimization of User Intent Recognition (UIR) for Transfemoral
Amputees. Powered prostheses are being developed to help amputees handle several
different activities: standing, level walking, stepping up and down, walking up and
down a ramp, etc. For each walking mode, a different control policy is required to
control the prosthesis. User intent recognition system plays an important role to infer
the user’s activity mode while transitioning from one walking mode to another one, and
then to activate the appropriate controller. Pattern recognition techniques are used to
address such problems.
In this research, mechanical sensor signals are experimentally collected from an
able-bodied subject, and comprise the training inputs to the UIR system. Signals are
processed and filtered to eliminate noise and to handle missing data points. Signals
reflecting the state of the prosthesis, user-prosthesis interactions, and prosthesis-
environment interactions are used for user intent recognition. Hip and ankle angles,
ground reaction force (GRF) along three axes, and hip moment are chosen as relevant
input signals that reflect various gait modes. Principal component analysis is used to
convert data to a lower dimension by eliminating the least relevant features. We pro-
pose the use of correlation analysis to remove highly correlated observations from the
training set.
136 Y. Kondratenko et al.
The main component of the UIR system is its classifier. We use K-nearest neighbor
(KNN) as a classification method for this purpose. KNN is modified and optimized
with an evolutionary algorithm for enhanced performance. We also modify KNN so
that the contribution of each neighbor is weighted on the basis of its distance to the test
point, and on the basis of the history of previously classified test points. This modi-
fication leads to better performance than standard KNN. Optimization techniques can
be used to tune the KNN parameters and obtain a classification system with the highest
possible accuracy. We choose biogeography-based optimization (BBO) as the evolu-
tionary optimization algorithm for this purpose. The optimization problem is to min-
imize the classification error. The UIR system can then be used to identify unknown
walking activities. The architecture of the UIR system is illustrated in Fig. 1. We use
MATLAB to implement user intent recognition. BBO is a stochastic algorithm, so it
requires several runs to optimize the parameters. We use parallel computing to reduce
the optimization time from 7.8 days to about 20 h [11]. To test the proposed method,
multiple sets of experimental data are collected for various gait modes: standing (ST),
slow walking (SW), normal walking (NW), and fast walking (FW). Figure 2illustrates
the experimental setup for able-bodied subjects. Future work will extend the proposed
approach to amputee gait data. Figure 3shows an example of test data for a walking
trial lasting approximately 18 s, which includes different walking modes.
Fig. 1. Architecture of user intent recognition system: an evolutionary algorithm (not shown) is
used to optimize the system components
Fig. 2. Experimental setup: data
collection for user intent recogni-
tion for able-bodied subjects
-100
-80
Ankle
Flexion
(deg)
0
20
40
Hip
Flexion
(deg)
-100
0
100
Hip
Moment
(Nm)
0
50
100
Fx
(N)
0
500
1000
Fy
(N)
-200
0
200
Fz
(N)
0 2 4 6 8 12 14 16 18
ST
SW
NW
FW
Walking
Mode
time (s)
Fig. 3. Sample test data showing four different gait modes
and transitions: ST (standing), SW (slow walk), NW
(normal walk), and FW (fast walk)
Robotics and Prosthetics at Cleveland State University 137
Table 1shows the performance of different versions of KNN. The first row of
Table 1shows simple KNN, which uses K¼7 nearest neighbors and results in 12.9%
test error. Test error reduces to 11.5% if we use weighted KNN with K¼7 nearest
neighbors. Test error is 8.06% when weighted KNN is used in addition to previously
classified gait modes to inform the current classification mode. The fourth row of
Table 1shows that the optimized weighted KNN with information from previously
classified modes provides the minimum classification error. Figure 4shows the per-
formance of the classifier using both simple KNN and optimized KNN. Classification
error for optimized KNN is 3.59% compared to 12.9% with standard KNN.
In conclusion, KNN was modified to enhance the performance of a user intent
recognition system. An evolutionary algorithm was applied to optimize the classifier
parameters. Experimental data was used for training and testing the system. It is shown
that the optimized system can classify four different walking modes with an accuracy of
96%. The code used to generate these results is available at http://embeddedlab.
csuohio.edu/prosthetics/research/user-intent-recognition.html. Further details about this
research can be found in [11].
Stable Robust Adaptive Impedance Control of a Prosthetic Leg. We propose a
nonlinear robust model reference adaptive impedance controller for a prosthesis test
robot. We use an adaptive control term to compensate for the uncertain parameters of
the system, and a robust control term to keep the error trajectories in a boundary layer
so the system exhibits robustness to variations of ground reaction force. The algorithm
not only compromises between control chattering and tracking, but also limits
tracking-error-based (TEB) parameter adaptation to prevent unfavorable drift. The
Table 1. Performance of KNN for user intent recognition with different classifier parameters
Method Train error Test error
Simple KNN (K¼7) 7.41% 12.9%
Weighted KNN (K¼7) 3.81% 11.5%
Weighted KNN and Recent Modes (K¼7) 3.44% 8.06%
Optimized KNN (K¼12) 3.22% 3.59%
0 2 4 6 8 10 12 14 16 18
ST
SW
NW
FW
Time (s)
Walking
Mode
Actual
Classified
0 2 4 6 8 10 12 14 16 18
ST
SW
NW
FW
Time (s)
Walking
Mode
Actua l
Classified
Fig. 4. User intent classifier results for optimized KNN is 3.59% error (right), which improved
from 12.9% with standard KNN (left)
138 Y. Kondratenko et al.
acceleration-free regressor form of the system obviates the need to measure joint
accelerations, which would otherwise introduce noise in the system. We use particle
swarm optimization (PSO) to optimize the parameters of the controller and the adap-
tation law. The PSO cost function is comprised of torque optimality and tracking
performance.
The prosthesis is an active transfemoral (above-knee) prosthesis. The complete
system model has a prismatic-revolute-revolute (PRR) joint structure. Human hip and
thigh motion are emulated by a prosthesis test robot. The vertical degree of freedom
represents vertical hip motion, the first rotational axis represents angular thigh motion,
and the second rotational axis represents prosthetic angular knee motion [1,2]. The
three degree-of-freedom model can be written as follows:
M€
qþC_
qþgþR¼uTe;ð1Þ
where qT¼q1q2q3
½is the vector of generalized joint displacements (q1is the
vertical displacement, q2is the thigh angle, and q3is the knee angle); uis the control
signal that comprises the active control force at the hip and the active control torques at
the thigh and knee; and Teis the effect of the GRF on the three joints. The contribution
of this research is a nonlinear robust adaptive impedance controller using a boundary
layer and a sliding surface to track reference inputs in the presence of parameter
uncertainties. We desire the closed-loop system to provide near-normal gait for
amputees. Therefore, we define a target impedance model with characteristics that are
similar to those of able-bodied walking:
Mr€
qr€
qd
ðÞþBr_
qr_
qd
ðÞþKrqrqd
ðÞ¼Teð2Þ
where qrand qdare the trajectory of the reference model and the desired trajectory
respectively. Since the parameters of the system are unknown, we use the control law
u¼b
TeKdsatðd
s=diagðuÞÞ þ b
M_
vþb
Cvþb
gþb
Rð3Þ
where the diagonal elements of uare the widths of the saturation function; s¼_
eþke
and v¼_
qrkeare the error and signal vectors respectively; e¼qqr
denotes tracking error, k¼diag k1;k2;...;kn
ðÞ;where ki[0; Kd¼diag Kd1;ð
Kd2;...;KdnÞ;where Kdi [0; nis the number of rigid links; and b
M;b
C;b
g;b
R, and b
Teare
estimates of M;C;g;R, and Terespectively. The control law [36] of Eq. (3) comprises
two different parts. The first part, b
TeKdsatðs=diagðuÞÞ, satisfies the reaching con-
dition (sgn sðÞ
_
sc,c¼c1c2... cn
½
Tand ci[0) and handles the variations
of the external inputs Te. The second part, b
M_
vþb
Cvþb
gþb
R, is an adaptive term that
handles the uncertain parameters, which are estimated via the following adaptation
mechanism:
_
^
p¼l1YTq;_
q;v;_
vðÞsD;ð4Þ
Robotics and Prosthetics at Cleveland State University 139
where Yq;_
q;v;_
vðÞis an acceleration-free regressor for the left side of Eq. (1); sDis the
boundary layer trajectory; and lis an rrdesign matrix with positive diagonal
elements. To trade off control chattering and tracking accuracy, and to create an
adaptation dead zone to prevent unfavorable parameter drift, we define a trajectory sD
as follows [1,2,39]:
sD¼0;s
jj
diagðuÞ
susatðs=diagðuÞÞ;s
jj
[diagðuÞ
ð5Þ
where sDis an n-element vector; the region s
jj
diagðuÞis the boundary layer and the
inequality is interpreted element-wise; and the diagonal elements of uare the boundary
layer thicknesses and the widths of the saturation function so that u¼
diag u1;u2;...;un
ðÞand ui[0.
To perform a stability analysis of the controller, the following positive-definite
Lyapunov function is considered:
Vs
D;~
pðÞ¼
1
2sT
DMsD
þ1
2~
pTl~
p
:ð6Þ
Let us assume that b
TeiTei
FiFm,cm¼maxðciÞ, and ais a positive scalar.
Given the Lyapunov function of Eq. (6), the control law of Eq. (3), and the TEB
adaptation mechanism of Eq. (4) in conjunction with the boundary layer trajectory of
Eq. (5), if Kdi a_
qmaxuiþFmþcm, then _
Vs
D;~
pðÞ!0ast!1, which means that
sD!0 and the controller guarantees the convergence of the error trajectories to the
boundary layer after the adaptation period.
We use PSO to tune the controller and estimator parameters. PSO decreases the
cost function (a blend of tracking and control costs) by 8%. We suppose the system
parameters can vary by ±30% from their nominal values. Figure 5compares the states
of the closed-loop system with the desired trajectories when the system parameters
vary. The MATLAB code used to generate these results is available at http://
embeddedlab.csuohio.edu/prosthetics/research/robust-adaptive.html [2].
Hybrid Function Approximation-Based Impedance Control for Prosthetic Legs.
In our previous research [7] we developed a process to combine the different control
schemes of a prosthesis test robot and a prosthetic leg to yield a stable system. We
assumed that the prosthesis test robot was controlled with Slotine and Li’s
regressor-based controller while the prosthesis was controlled with a regressor-free
controller. We addressed this problem by defining a framework within which two
controllers could be systematically combined by maintaining their indirect dependence
on each other, and we developed a theorem that proved that the combined robotic
system was stable, and we showed the efficacy of the system using simulation results.
However, the goal of the controllers in [7] was pure motion tracking in the presence of
external disturbances. As a result, we obtained good reference trajectory tracking but
relatively high control signal magnitudes.
In an effort to reduce the effects of external disturbances, or ground reaction forces
(GRFs), we augment impedance control, which is a form of environmental interaction
140 Y. Kondratenko et al.
control, to our previously developed hybrid control scheme. Impedance control gives
the modified control scheme the ability to trade off trajectory tracking with control
signal magnitude, depending on the nature of the GRFs.
The combination of the prosthesis test robot and the prosthetic leg can be described
by the dynamic equations of a rigid robot under the influence of external forces
[29,39]. We use pure motion tracking as the goal of the prosthesis test robot and
impedance control as the goal of the prosthetic leg. For impedance control we design a
controller such that the closed-loop system behaves like the target impedance
Mi€
qr€
qd
ðÞþBi_
qr_
qd
ðÞþIiqrqd
ðÞ¼Teð7Þ
where qr2Rnand qd2Rnare the reference and desired trajectory respectively,
Mi2Rnn,Bi2Rnn, and Ii2Rnnare the apparent inertia, damping, and stiffness
respectively, and Te2Rncaptures the external torques and forces applied to the
coupled robotic system.
We use MATLAB/Simulink to simulate the system’s behavior with the proposed
controller, which is a combination of regressor-free environmental interaction control
and regressor-based pure motion tracking control; see Figs. 6and 7. Figure 6shows
good tracking of the reference trajectories for hip displacement, hip angle, and knee
angle. The controller for the knee angle gives reasonable trade-offs in tracking and
0 1 2 3 4
-0.04
-0.02
0
0.02
0.04
time(s)
hip displacement(m)
0 1 2 3 4
-0.4
-0.2
0
0.2
0.4
time(s)
hip velocity(m/s)
0 1 2 3 4
0.5
1
1.5
2
time(s)
thigh angle(rad)
0 1 2 3 4
-4
-2
0
2
4
time(s)
thigh angular velocity(rad/s)
0 1 2 3 4
-0.5
0
0.5
1
1.5
time(s)
knee angle(rad)
0 1 2 3 4
-10
-5
0
5
10
time(s)
knee angular velocity(rad/s)
Fig. 5. Joint displacements and velocities with stable robust adaptive impedance control
Robotics and Prosthetics at Cleveland State University 141
control signal magnitude when under the influence of GRFs, and then resumes tight
reference trajectory tracking when there is no GRF.
In Fig. 7we see that the control signal magnitudes for the hip displacement, hip
angle, and knee angle are relatively low, and the vertical GRF is comparable to that
experienced during able-bodied walking.
In conclusion, the simulation results show the combination of two different robotic
systems with different control schemes. The simulation results show that impedance
control is valuable in reducing control signal magnitudes, and hence preventing
damage to equipment and reducing strain on amputees.
System Identification and Control Optimization of a Prosthetic Knee. In this
research, an EMG-30 geared DC motor is installed in a Mauch SNS knee to create an
active prosthesis. The Mauch SNS knee is a widely used passive prosthesis; we adapted
it here by detaching the damper connection and driving it with a DC motor.
Our work provides a basic framework for system identification, control optimiza-
tion, and implementation of an active prosthetic knee during swing phase. To apply
velocity control to the system, proportional-integral-derivative control (PID) is utilized
due its applicability to an extensive variety of systems, its simplicity, and its ease of use
with embedded systems technology. The objective of this research is parameter
investigation for PID with respect to prosthetic leg shank length. To accomplish this
objective, we first need to develop a prosthetic leg model. We use heuristic algorithms
and gradient descent algorithms to identify model parameters and tune the PID
controller.
Particle swarm optimization, biogeography-based optimization (BBO) and
sequential quadratic optimization (SQP) [18,20,32,37] are selected for identification
and tuning. The reason for using more than one optimization algorithm is to avoid local
minimum solutions, to discover which algorithm is superior for this task, and to dis-
cover how sensitive each heuristic algorithm is to its own parameters.
Fig. 6. Joint angle trajectories with hybrid
function approximation-based impedance
control
Fig. 7. Control signals and vertical ground
reaction force with hybrid function
approximation-based impedance control
142 Y. Kondratenko et al.
Hardware setup includes a servo system composed of a desktop PC connected to a
Quanser©DAQ card, Matlab with Quanser Quarc software for real-time connectivity,
and DAQ hardware; see Fig. 8. The DAQ system delivers an analog control signal to a
servo amplifier to drive the EMG30 DC motor. An axial quadrature encoder sends
signals through two digital channels. A Mauch SNS knee is attached to an EMG-30
geared DC motor to comprise our active leg prosthesis.
Numerical differentiation is usually used to obtain angular velocity by differenti-
ating the encoder signal [42]. This technique often leads to a highly distorted signal due
to encoder resolution. Therefore, a Kalman filter is designed to estimate the angular
velocity. The DC geared motor and the Mauch SNS are described mathematically in
[10]. Matlab Simulink is used to build simulation models. In order to identify model
parameters and obtain average performance metrics, optimization algorithms execute
20 times each. The DC motor model and Mauch knee joint model are combined to form
the active prosthetic leg model.
In order to see how sensitive BBO and PSO performances are, a sensitivity analysis
test is carried out for each algorithm. We say that an algorithm’s sensitivity to one of its
parameters is “High,”“Medium,”or “Low,”if a given percentage parameter variation
leads to a deviation from the best solution by less than 10%, 10–25%, or more than
25%, respectively. Tables 2and 3show the algorithm parameter values and their
sensitivities.
Servo Amplifier
Active Prosthetic
Leg DAQ System
Encoder Data
Fig. 8. Hardware setup for system identification and control optimization
Table 2. BBO algorithm parameter sensitivity
Lowest
value
Highest
value
Test
increment
Best
value
Sensitivity
Number of
generations
50 125 25 100 Medium
Population size 50 100 10 60 Medium
Mutation
probability
5% 20% 5% 10% High
Number of elites 1 5 1 2High
Robotics and Prosthetics at Cleveland State University 143
The active prosthetic knee model and PID are used to build a closed-loop feedback
system. To investigate PID controller parameter behavior with respect to shank length,
we use the optimization algorithms to tune the controller parameters (Kp;Kiand Kd).
Results show that for model parameter identification, particle swarm optimization gives
the best optimization results, and BBO gives better average overall performance than
SQP. For PID tuning, BBO achieves the best average overall performance, but PSO
shows the fastest average convergence. Finally, we see that increasing shank length
results in an increase in the optimal proportional gain, and a decrease in the optimal
differential and integral gains; see Fig. 9.
Ground Reaction Force Estimation with an Extended Kalman Filter. A method to
estimate GRF in a robot/prosthesis system is presented. The system includes a robot
that emulates human hip and thigh motion, and a powered prosthesis for transfemoral
amputees, and includes four degrees of freedom: vertical hip displacement, thigh angle,
Table 3. PSO algorithm parameter sensitivity
Lowest
value
Highest
value
Test
increment
Best
value
Sensitivity
Number of
generations
50 250 50 100 Medium
Population size 50 100 10 60 Low
Correction factor 0.5 3 0.5 2Medium
Acceleration
constant
15 1 0.5 High
Cognitive
parameter
0.05 1 0.05 0.1 High
Social parameter 0.1 0.5 0.1 0.3 High
1 2 3 4 5 6 7 8 9
0
5
10
15
20
Extension (cm)
Gains
K
p
K
i
K
d
Fig. 9. PID parameter values with respect to shank length
144 Y. Kondratenko et al.
knee angle, and ankle angle. A continuous-time extended Kalman filter (EKF) [38]
estimates the states of the system and the GRFs that act on the prosthetic foot.
The system includes eight states: q1is vertical hip displacement, q2is thigh angle,
q3is knee angle, q4is ankle angle, and their derivatives. Horizontal and vertical GRF is
applied to the toe and heel of a triangular foot. The ground stiffness is modeled to
simulate GRF. The initial state xð0Þis obtained from reference data, and we randomly
initialize the estimated state b
x0ðÞto include estimation error. The diagonal covariance
matrices of the continuous-time process noise and measurement noise are tuned to
obtain good performance. Results are shown in Fig. 10. Even with significant initial
estimation errors, the EKF converges to the true states quickly.
The performance of the EKF may deteriorate significantly with modeling uncer-
tainties. The H1filter [38] was designed to improve the robustness of state estimation
in the presence of modeling errors. Here we assume that the robot/prosthesis system
parameters vary by ±10% from their nominal values. In this test the initial value of the
state vector xð0Þand the measurement and process noise covariances Rand Qare
identical to those used in the EKF. We also set the initial value of the estimated state
vector b
x0ðÞto provide an arbitrary but nonzero initial estimation error. The tuning
parameters in the H1filter are chosen by the trial and error to obtain good perfor-
mance. The RMS estimation errors of the two filters are compared in Tables 4and 5
when 10%uncertainty on the system parameters is imposed. We see that the H1
filter generally performs better than the EKF. However, the EKF still works well.
Table 4. Comparison between EKF and H1filters in terms of RMSE when the system
parameters (SPs) deviate from their nominal values by +10%
SPs x1;mðÞ x2;ðradÞx3;ðradÞx4;ðrad Þx5;ðm=sÞx6;ðrad=sÞ
EKF 0.005 0.007 0.03 0.02 0.06 0.12
H10.001 0.005 0.01 0.01 0.03 0.12
SPs x7;ðrad=sÞx8;ðrad=sÞx9;ðNÞx10 ;ðNÞx11 ;ðNÞx12;ðNÞ
EKF 0.47 1.09 4.1 19 7.8 33
H10.44 0.36 1.5 7.5 5.7 25
Table 5. Comparison between EKF and H1filters in terms of RMSE when the system
parameters (SPs) deviate from their nominal values by −10%
SPs x1;mðÞ x2;ðradÞx3;ðradÞx4;ðrad Þx5;ðm=sÞx6;ðrad=sÞ
EKF 0.007 0.005 0.03 0.02 0.07 0.14
H10.002 0.005 0.01 0.03 0.03 0.11
SPs x7;ðrad=sÞx8;ðrad=sÞx9;ðNÞx10 ;ðNÞx11 ;ðNÞx12;ðNÞ
EKF 0.45 1.10 5.5 19 6.2 31
H10.44 0.44 2.8 9.3 4.5 22
Robotics and Prosthetics at Cleveland State University 145
Electronic Energy Converter Design for Regenerative Prosthetics. Prosthetic
models use ideal electromechanical actuators for knee joints, which do not include
energy regeneration. In order to focus on energy regeneration, a voltage source con-
verter is designed in this research to interface an electric motor to a supercapacitor.
The converter is designed to resemble a typical H-bridge motor driver. The voltage
converter control system allows power to flow from the motor to the capacitor (motor
mode) and from the capacitor to the motor (generator mode). During motor mode, the
voltage converter’s control system modulates the voltage applied to the motor using
two circuits; one with the capacitor connected (powering the motor from the capacitor)
and one with the capacitor disconnected (shorting the motor connection through the
H-bridge). During generator mode, the voltage converter control system changes the
impedance connected to the motor using two circuits; one with the capacitor connected
(charging the capacitor) and one with the capacitor disconnected (allowing the motor to
move with less resistance from the electronics). The circuit and motor are modeled with
state space equations using MATLAB and Simulink software.
The converter is augmented to a previously developed mechanical prosthesis model
[21]. The model includes the mechanical dynamics of the prosthesis, a ground contact
model, and a robust tracking/impedance controller. The controller calculates desired
joint torques to achieve trajectory tracking of abled-bodied reference data. The con-
verter replaces the ideal knee motor actuator in this prosthesis model. Since torque and
current are proportionally related through the motor dynamics, the tracking/impedance
controller is modified to command a desired current that the converter applies through
the knee motor to create the desired torque at the knee joint.
A neural network creates an inner control loop for the converter to generate the
commanded motor current while the tracking/impedance controller determines torques
to meet the tracking and impedance goals for the prosthesis. The neural network
(a) (b)
(c) (d)
00.2 0.4 0.6 0.8 1
-50
0
50
100
150
200
time(sec)
external force in x-direction acting on heel(N)
actual
estimated
Heel
strike
Midstance
00.2 0.4 0.6 0.8 1
-50
0
50
100
150
time
(
sec
)
external force in x-direction acting on toe(N)
actual
estimated
Toe- o f f
00.2 0.4 0.6 0.8 1
-200
0
200
400
600
800
1000
time
(
sec
)
external force in z-direction acting on heel(N)
actual
estimated
Heel
strike
Midstance
00.2 0.4 0.6 0.8 1
-200
0
200
400
600
800
time
(
sec
)
external forc e in z-direction acting on to e(N)
actual
estimated
Toe- o f f
Fig. 10. Horizontal and vertical ground force (GRF) estimation
146 Y. Kondratenko et al.
includes an input node which compares the motor current generated by the converter to
the desired motor current commanded by the tracking/impedance controller. In addition
to the error signal of the motor current, a measured ground reaction force input node is
used to determine if the prosthesis is in a stance phase or swing phase, a measured
velocity input node is used to determine the direction of motor rotation, and a measured
torque input node is used to indicate whether the prosthesis is operating in the motoring
or generating mode. The neural network contains a single output node which com-
mands a change in duty cycle for the converter to modulate the power flow between the
capacitor and the knee motor to achieve the motor torque commanded by the
tracking/impedance controller.
The controller gains for the tracking/impedance controller and neural network
controller, as well as the physical parameters such as the capacitance of the capacitor
and the length of the prosthesis transmission links, are optimized with BBO. The
system is optimized using a single set of desired trajectories (hip displacement, hip
rotation, knee rotation, and ankle rotation). The control gains and physical parameters
selected by BBO achieve the optimization objective of knee angle tracking with a root
mean square (RMS) error of 0.13° during five seconds of simulated walking. The
selected control gains and physical parameters are then used with test sets of walking
data to ensure that the system can maintain tracking with different trajectories. The
prosthesis maintains knee angle tracking with an RMS error of 1° with seven different
sets of data during five seconds of walking. One representative set of data is shown in
Fig. 11. Some data sets show a loss of energy in the capacitor, but the best case results
in an increase of 67 Joules of energy during five seconds of walking. The transfer of
energy between the knee and capacitor is shown in Fig. 12. It is observed that the
capacitor charges as the knee motor produces excess energy and the capacitor dis-
charges as the knee motor consumes energy.
Fig. 11. Reference and simulated trajectories of hip displacement, hip rotation, knee rotation,
and ankle rotation during five seconds of walking
Robotics and Prosthetics at Cleveland State University 147
Fuzzy Logic for Robot Navigation. This research uses fuzzy logic to find a path for a
mobile robot to navigate in an environment with both static and dynamic obstacles
when the robot does not have any prior information about the obstacle locations. The
robot stores the coordinates of previously visited locations in memory to avoid getting
stuck in dead ends.
The robot radar returns a fuzzy set based on the distance L
i
from obstacle i(see
Fig. 13): lu
iui
ðÞ¼
Li
Lmax. The robot finds the angle between its position and the target
position, which we call a. If the robot moved in the adirection in an obstacle-free
environment it would follow a direct line to the target. However, there are obstacles in
the path. To find a safe path around the obstacles, we introduce a Gaussian fuzzy set
[13,14,28,43] which has a maximum value at a:
la
iui
ðÞ¼eððuiaÞ2
2r2Þ:
We combine lu
iui
ðÞand la
iui
ðÞto obtain a new fuzzy set, lw
iwi
ðÞ, shown in
Fig. 13.
lw
iui
ðÞ¼minðla
iui
ðÞ;lu
iðuiÞÞ
The movement direction is the maximum point in lw
iwi
ðÞ, which we call uA. If the
robot moves with angle of uA, it will touch the obstacles. We therefore introduce a new
fuzzy set that has the value 1 in a range of 120° around uA:
Fig. 12. Comparison between energy produced at the knee and energy stored in the capacitor
148 Y. Kondratenko et al.
lh
1ðuiÞ¼ 1juiuAj\60
0 otherwise
In the next step we defuzzify lwui
ðÞlh
1ðuiÞusing center of mass defuzzification
[31], which is shown in Fig. 14.
Simulations confirm that the proposed approach provides reliable navigation.
However, the robot is only able to get to the target point if it does not enter a dead end
zone. Examples of dead end zones include rooms, single-entry areas, and other situ-
ations where the robot needs to move backward to find the path to the target. In
practical applications, we cannot guarantee that a map won’t have dead end zones.
The robot therefore saves visited paths in memory. The data in the robot’s memory
is a list of coordinates which we call Memory Points (MPsÞ. The robot should avoid
visiting the same place multiple times and it should be able distinguish between
locations that were visited in the recent and in the distant past. To achieve this goal, we
assign weights to each coordinate in memory (w). Weights change based on the dis-
tance to the robot. Their values are calculated with a derivative of a Gaussian distri-
bution as shown in Fig. 15 and as described by the following:
wiðMPiÞ¼ ded2
2r2
r
;where d¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
xMPixRPi
ðÞ
2þyMPiyRPi
ðÞ
2
qi¼1;2;...;N
(;
Nis the number of MPs in memory and RP is the current position of the robot. With
this weighting function, the robot is more likely to visit very recent and very old
locations and tends to avoid coordinates that are in between.
In different layouts and different robot and target positions, the robot can find a path
to the target point successfully; see Fig. 16.
(a) (b)
Fig. 13. (a) A polar radar map in the presence of an obstacle, and (b) its transformation to
Cartesian coordinates
Robotics and Prosthetics at Cleveland State University 149
5 Education and Research Integration
The seven research projects described in the preceding section have a common edu-
cational core which enables their success. This section reviews those common
elements.
First, most of the students involved in research come to CSU with the promise of
multi-year research funding. Most funding comes from external agencies, such as the
US government or industry. Some funding comes from the limited resources within
CSU. In order to be productive researchers, students need to be devoted full-time to
their studies and research, and they need to be worry-free with regards to finances.
Successful research depends on funding, and funding depends on faculty.
Second, research students are involved not only in coursework and research, but
also in teaching-assistant duties. Some students assist with labs, tutoring, and grading,
while other students are given sole responsibility for an entire undergraduate course,
depending on their experience level. Superficially, this detracts from their research.
However, in the long run, it enhances their research by providing them with training in
the area of communication skills, problem solving, and networking.
Fig. 14. Shaded area is lw
iwi
ðÞ Fig. 15. Weight distribution of Memory
Points for robot path planning
Fig. 16. Fuzzy robot path planning results: the red line is the robot path from start to target, and
the green circles are dynamic obstacles. (Color figure online)
150 Y. Kondratenko et al.
Third, most research students are recruited as doctoral students, who are more
serious about research than master’s students. Some students enter CSU as master’s
students, but always with the intent of ultimately progressing to doctoral studies.
Fourth, research students are required to attend and present at weekly research
seminars at CSU. These seminars provide the students with opportunities to learn from
each other and from faculty, to network with each other, and to practice their com-
munication skills. This indicates the need for a critical mass of research students in
order to ensure a successful research enterprise.
Fifth, research students are required to publish and present their research at one or
more international conferences each year. This provides them with similar benefits as
the weekly seminars discussed above, but at a larger scale.
Sixth, research students with varying levels of experience and disciplinary focus all
work together. Research participants include faculty, post-doctoral scholars, visiting
scholars, doctoral students, master’s students, and undergraduate students. Diversity
also intentionally includes gender, ethnicity, and nationality. This diversity allows
research students to be involved in both receiving and providing mentoring, and in
learning to work across comfortable boundaries.
Seventh, research students are given as much responsibility as possible in the
conduct of their research, and this responsibility gradually increases as the students
gain experience. Faculty advisors fill the role of advising, but generally try to keep a
hands-off approach in the daily conduct of student research. This approach teaches
students to be proactive in solving problems, to take the initiative in seeking advice,
and to take responsibility for their research. Faculty advisors are responsible to help
students find the right balance so that the students don’t go down the wrong research
path or stall in their research efforts.
Many educational factors are involved in successful research. The above factors are
just a few. Many more could be incorporated at CSU and other universities. But the
most important consideration here is that in order to be successful, faculty must take an
intentional approach to integrating education and research, and to graduating research
students who are prepared to take the lead in the next generation.
6 Conclusions
The authors have described university student training. The description has focused on
student participation in the US NSF project “Optimal prosthesis design with energy
regeneration”and the application of ICT and modelling technologies.
Several factors play an important role in the results of this chapter. Student research
requires skill in programming and software, and a broad theoretical knowledge in
computer science, and mechanical, electrical, and control engineering. Students used
MATLAB, Simulink, and toolboxes (Optimization, Fuzzy Logic, etc.), and program-
ming in C and C++. The software used for robot trajectory planning research was
designed and written by students in C++, and the GUI was designed using Qt and
OpenGL. Standard libraries were used to make the software cross-platform.
The most important foundation for student research is theoretical knowledge in
fundamental and elective disciplines such as Circuits, Linear Systems, Control
Robotics and Prosthetics at Cleveland State University 151
Systems, Nonlinear Control, Machine Learning, Artificial Intelligence, Intelligent
Controls, Optimal State Estimation, Optimal Control, Embedded Systems, Robot
Modeling and Control, Probability and Stochastic Processes, Population-Based Opti-
mization, and Prosthesis Design and Control, which provides a basic understanding of
human biomechanics and lower-limb prosthesis design and control. These courses
played a vital role in the proper grounding of basic and advanced ICT and control
theory for robotic and prosthetic leg research. The facilities at CSU and funding from
the NSF significantly helped in furthering student research-based education.
Finally, student participation in government-sponsored research, student exchanges
of research experiences with each other, and publication of research results in
high-caliber journals and conferences [1,2,7,11,18,29], provide students with effective
training and self-confidence in their higher education. Research-based education also
allows students to obtain practical experience as research assistants, with corresponding
responsibilities in the development and implementation of research projects.
Student participation in real-world research significantly influences their engi-
neering and research qualifications by: (a) giving them a strong understanding of ICT
and engineering concepts that are covered in corresponding courses; (b) giving them
practical experience and the ability to apply theoretical knowledge; (c) giving them the
opportunity to learn technical material independently; (d) helping them improve fun-
damental skills to apply in other research in their future; (e) providing them with a rich
interdisciplinary research environment; and (f) providing them with an understanding
of concepts both familiar and unfamiliar. Through extensive literature review and
actively seeking ways to solve research problems, students are prepared to make
meaningful future contributions to the field of ICT and control engineering.
Acknowledgements. The authors thank the Fulbright Program (USA) for supporting Prof.
Y.P. Kondratenko with a Fulbright 2015-2016 scholarship and for making it possible for this
team to conduct research in together in the USA. This research was partially supported by
US NSF Grant 1344954.
References
1. Azimi, V., Simon, D., Richter, H., Fakoorian, S.A.: Robust composite adaptive transfemoral
prosthesis control with non-scalar boundary layer trajectories. In: Proceedings of the
American Control Conference (ACC), Boston, MA, USA (2016)
2. Azimi, V., Simon, D., Richter, H.: Stable robust adaptive impedance control of a prosthetic
leg. In: Dynamic Systems and Control Conference, Columbus, Ohio, USA, October 2015
3. Bredeweg, B., Forbus, K.D.: Qualitative modeling in education. AI Mag. 24(4), 35–46
(2003). American Association for Artificial Intelligence
4. Chou, C., Tsai, C.-C.: Developing web-based curricula: issues and challenges. J. Curriculum
Stud. 34(6), 623–636 (2002)
5. Clements, D.H.: Curriculum research: toward a framework for “Research-Based Curricula”.
J. Res. Math. Educ. 38(1), 35–70 (2007)
6. Conditt, M.A., Noble, P.C., Thompson, M.T., Ismaily, S.K., Moy, G.J., Mathis, K.B.: A
computerized bioskills system for surgical skills training in total knee replacement. Proc.
Inst. Mech. Eng. Part H J. Eng. Med. 221(1), 61–69 (2007)
152 Y. Kondratenko et al.
7. Ebeigbe, D., Simon, D., Richter, H.: Hybrid function approximation based control with
application to prosthetic legs. In: IEEE International Systems Conference, Orlando, FL,
USA, April 2016
8. Gáti, J., Kártyás, G.: Computer system support for higher education programs in
engineering. In: Proceedings of 4th International Conference on Emerging Trends in
Engineering and Technology (ICETET), pp. 61–65 (2011)
9. Gilbert, J.K.: Models and modelling: routes to more authentic science education. Int. J. Sci.
Math. Educ. 2(2), 115–130 (2004)
10. Gonçalves, J., Lima, J., Costa, P., Moreira, A.P.: Modeling and simulation of the EMG30
geared motor with encoder resorting to SimTwo: the official Robot@Factory simulator. In:
Azevedo, A. (ed.) Advances in Sustainable and Competitive Manufacturing Systems,
pp. 307–314. Springer, Heidelberg (2013)
11. Khademi, G., Mohammadi, H., Simon, D., Hardin, E.C.: Evolutionary optimization of user
intent recognition for transfemoral amputees. In: IEEE Biomedical Circuits and Systems
Conference (BioCAS), Atlanta, Georgia, USA, pp. 1–4 (2015)
12. Kondratenko, Y.P., Kondratenko, G.V.: Robotic system with myoelectric adapting for
rehabilitation tasks. ACTA Bioeng. Biomech. 3,1–6 (2001)
13. Kondratenko, Y., Kondratenko, V.: Soft computing algorithm for arithmetic multiplication
of fuzzy sets based on universal analytic models. In: Ermolayev, V., Mayr, H., Nikitchenko,
M., Spivakovsky, A., Zholtkevych, G. (eds.) Information and Communication Technologies
in Education, Research, and Industrial Application. Communications in Computer and
Information Science, vol. 469, pp. 49–77. Springer, Cham (2014)
14. Kondratenko, Y.P., Klymenko, L.P., Al Zu’bi, E.Y.M.: Structural optimization of fuzzy
systems’rules base and aggregation models. Kybernetes 42(5), 831–843 (2013)
15. Kondratenko, Y., Khademi, G., Azimi, V., Ebeigbe, D., Abdelhady, M., Fakoorian, S.A.,
Barto, T., Roshanineshat, A.Y., Atamanyuk, I., Simon D.: Information, communication, and
modeling technologies in prosthetic leg and robotics research at Cleveland State University.
In: Ermolayev, V., et al. (eds.) Proceedings of 12th International Conference on Information
and Communication Technologies in Education, Research, and Industrial Application, 21–
24 June 2016, Kyiv, Ukraine, ICTERI 2016, vol. 1614, pp. 168–183. CEUR-WS (2016)
16. Kondratenko, Y., Simon, D., Atamanyuk, I.: University curricula modification based on
advancements in information and communication technologies. In: Ermolayev, V., et al.
(eds.) Proceedings of 12th International Conference on Information and Communication
Technologies in Education, Research, and Industrial Application, 21–24 June 2016, Kyiv,
Ukraine, ICTERI 2016, vol. 1614, pp. 184–199. CEUR-WS (2016)
17. Louca, L.T., Zacharia, Z.C.: The use of computer-based programming environments as
computer modelling tools in early science education: the cases of textual and graphical
program languages. Int. J. Sci. Educ. 30(3), 287–323 (2008)
18. Ma, H., Simon, D.: Biogeography-based optimization with blended migration for
constrained optimization problems. In: Proceedings of the 12th Annual Conference on
Genetic and Evolutionary Computation, New York, NY, USA, pp. 417–418 (2010)
19. McKee, N.H., Agur, A.M., Tsang, W., Singh, K.S.: Possible uses of computer modeling of
the functioning human hand. Clin. Plast. Surg. 32(4), 635–641 (2005)
20. Mo, H., Xu, Z.: Research of biogeography-based multi-objective evolutionary algorithm. In:
Khosrow-Pour, M. (ed.) Interdisciplinary Advances in Information Technology Research,
pp. 125–135 (2013)
21. Mohammadi, H., Richter, H.: Robust tracking/impedance control: application to prosthetics.
In: American Control Conference, Chicago, Illinois, July 2015 (2015)
Robotics and Prosthetics at Cleveland State University 153
22. Musa, A., Al-Dmour, A., Fraij, F., Gonzalez, V., Al-Hashemi, R.: Developing a virtual
laboratory for a communication and computer networking course. Int. J. Continuing Eng.
Educ. Life Long Learn. 20(3–5), 390–406 (2010)
23. Nehrir, M.H., Fatehi, F., Gerez, V.: Computer modeling for enhancing instruction of electric
machinery. IEEE Trans. Educ. 38(2), 166–170 (1995)
24. Nemanjic, B., Svetozar, N.: Computer Simulations: Technology, Industrial Applications and
Effects on Learning. Nova Science Publishers Inc., New York (2012)
25. Neptune, R.R.: Computer modeling and simulation of human movement. Sci. Principles
Sports Rehabil. 11(2), 417–434 (2000)
26. Panselinas, G.E., Komis, V.: Using educational software to support collective thinking and
test hypotheses in the computer science curriculum. Educ. Inform. Technol. 16(2), 159–182
(2011)
27. Patil, L.S., Patil, K.D., Thosar, A.G.: The role of computer modeling and simulation in
power electronics education. In: Proceedings of the 2nd International Conference on
Emerging Trends in Engineering and Technology (ICETET), pp. 416–419 (2009)
28. Piegat, A.: Fuzzy Modeling and Control. Springer, Heidelberg (2001). doi:10.1007/978-3-
7908-1824-6
29. Richter, H., Simon, D., Smith, W., Samorezov, S.: Dynamic modeling, parameter estimation
and control of a leg prosthesis test robot. Appl. Math. Model. 39, 559–573 (2015)
30. Richter, H., Simon, D.: Robust tracking control of the prosthesis test robot. J. Dyn. Syst.
Measur. Control 136(3), Paper No. DS-13–1052 (2014)
31. Runkler, T.: Selection of appropriate defuzzification methods using application specific
properties. IEEE Trans. Fuzzy Syst. 5(1), 72–79 (1997)
32. Sayed, M., Saad, M., Emara, H., El-Zahab, E.A.: A novel method for PID tuning using a
modified biogeography-based optimization algorithm. In: Proceedings of the 24th Chinese
Control and Decision Conference (CCDC), pp. 1642–1647 (2012)
33. Schonning, A.: Biomechanical applications of computers in engineering education. In:
ASME 2007 International Design Engineering Technical Conferences and Computers and
Information in Engineering Conference, Las Vegas, Nevada, USA, pp. 389–393. American
Society of Mechanical Engineers (2007)
34. Schwarz, C.V., Meyer, J., Sharma, A.: Technology, pedagogy, and epistemology:
Opportunities and challenges of using computer modeling and simulation tools in
elementary science methods. J. Sci. Teacher Educ. 18(2), 243–269 (2007)
35. Shaalan, H.: Using software to enhance power engineering education in a technology
program. IEEE Power Eng. Rev. 21(6), 62–63 (2001)
36. Shih, N.-J.: RP-aided computer modeling for architectural education. Comput. Graph. 30(1),
137–144 (2006)
37. Simon, D.: Evolutionary Optimization Algorithms: Biologically Inspired and
Population-Based Approaches to Computer Intelligence. Wiley, Hoboken (2013)
38. Simon, D.: Optimal State Estimation: Kalman, H-infinity, and Nonlinear Approaches. Wiley,
Hoboken (2006)
39. Slotine, J.-J.E., Coetsee, J.A.: Adaptive sliding controller synthesis for non-linear systems.
Int. J. Control 43(6), 1631–1651 (1986)
40. Teodoro, V.D., Neves, R.G.: Mathematical modelling in science and mathematics education.
Comput. Phys. Commun. 182(1), 8–10 (2011)
41. Van Amerongen, J., Breedveld, P.: Modelling of physical systems for the design and control
of mechatronic systems. Ann. Rev. Control 27(1), 87–117 (2003)
42. Yang, W.: Applied Numerical Methods Using MATLAB. Wiley-Interscience, Hoboken
(2005)
43. Zadeh, L.A.: Fuzzy sets. Inf. Control 8, 338–353 (1965)
154 Y. Kondratenko et al.
44. Zhang, B.H., Liu, X.F., Krajcik, J.S.: Expert models and modeling processes associated with
a computer-modeling tool. Sci. Educ. 90(4), 579–604 (2006)
45. Zhang, Q.: GUI based computer modeling and design platform to promote interactive
learning in fiber optic communications. In: Proceedings of the 37th Annual Frontiers in
Education Conference - Global Engineering: Knowledge Without Borders, Opportunities
Without Passports, FIE 2007, pp. S3H-14–S3H-19 (2007)
Robotics and Prosthetics at Cleveland State University 155