You are on page 1of 8

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882

Volume 4 Issue 2, February 2015

Intelligent Adaptive Control of an One Link Manipulator


Using Self-Tuning Regulator
Abd-El meged mohamed1, Gaber elsaady2, Ashraf Hemeda3, Asmaa Fawzy4
1engineering faculty, Aswan university, Egypt
2engineering faculty, Assiut university, Egypt
3,4energy engineering faculty, Aswan university, Egypt

Abstract
For overcomes several shortcomings of the inverse
control design to controlling nonlinear systems using the
neural networks as the controller based self tuning
regulator. The one link manipulator parameters are
estimated online and are used to update the weights of
the RBFNN. The weight update equations are derived
based on the least mean squares principle. The RBFNN
virtually models the inverse of the plant and thus the
output tracks the reference trajectory. The proposed
algorithm is successfully verified using simulations.
Then, this paper compared its result with the outcome of
using proportional-plus-integral feedback (PI) self
tuning regulator.
Keywords: STR, RBFNN, ARX, LMS, arm robot
1 INTRODUCTION
Adaptive inverse control method has received much
attention in recent years. But its main problem, when
applied to controlling nonlinear systems, is how to adapt
the inverse controller. The adaptive inverse control
system based neural network can achieve good
performance. The basic idea of adaptive inverse control
is to drive the plant with a signal from a controller
whose transfer function is the inverse of that of the plant
itself. The objective of this system is to cause the plant
output to follow the command input. Since the plant is
generally unknown, it is necessary to adapt or to adjust
the parameters of the controller in order to create a true
plant inverse. An error signal, the difference between the
plant output and the command input, is used by an
adaptive algorithm to adjust the controllers parameters
to minimize the mean square of this error [1].
In many practical applications it is, however, difficult
to determine the parameters of the controller, since the
dynamics of the process and its disturbances are
unknown [2, 3]. The parameters of the process thus have
to be estimated. For stationary processes it is possible to
determine
the
unknown
parameters
through
identification. The experiments and their evaluations
can, however, be rather time consuming. It is thus
desirable to have a regulator which tunes its parameters
on-line [4]. The purpose of self-tuning regulators (STR)

is to control systems with unknown but constant


parameters. The regulators can also be applied to
systems with slowly varying parameters.
K. Amini [5] their research addresses fuzzy-logic
method as an appropriate approach to control of
mechanical robot manipulators. The method is explained
and a review on applications of method in control of
mobile robot and flexible links manipulators is
presented. Ismail H. [6] utilized PID and State feedback
control methods which have been studied for Single Link Flexible Joint Robot Manipulator. The obtained
results compared due to various input signals for each
control methods. The results showed that state feedback
control slightly outperforms the PID control for singlelink flexible manipulator.
Neural network (NN) is a good structure for control
the nonlinear plants and has many types [7, 8]. Kumar
[9] used neural network for modeling the retention
process and as controller. In this paper, we used the
RBFNN as a controller. This type is faster one and uses
least number of neurons at hidden layer [10, 11]. The
inverse control means that the controller (RBFNN) acts
the inverse of the plant (dc motor) so the output tracks
the reference input [12]. Enzeng [13] present a neural
network based self tuning PID controller for autonomous
underwater vehicle, the control system consists of neural
network identifier and neural network controller, and the
weights of neural networks are trained by using Davidon
least square method, also[14].
Sabahi [15] used a new adaptive and nonlinear
control based on neural network approaches, this method
has been named feedback error learning (FEL)
approaches, that classical controller is used for training
of neural network feedforward controller. Pal [16]
proposed a simple self-tuning scheme for PI-type fuzzy
logic controllers (FLCs) for a real time water pressure
control system. This scheme is improved performance of
the system even at load change and set point variations.
Kota [17] used PID controller and fuzzy logic controller
for control separately excited dc motor. Fuzzy selftuning PID has better dynamic response curve, shorter
response time, small overshoot, and small steady state

www.ijsret.org

69

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

error compared to the conventional PID controller. Saad


[18] proved that the proposed Neural Network (NN)
self-tuning PID controller is more efficient to control the
robot manipulator to follow the desired trajectory
compared to classical tuning method of PID controller.
Alfonso [19] introduced a new self-tuning algorithm is
developed for determining the Fourier Series Controller
coefficients with the aim of reducing the torque ripple in
a Permanent Magnet Synchronous Motor (PMSM), thus
allowing for a smoother operation. This algorithm
adjusts the controller parameters based on the
component's harmonic distortion in time domain of the
compensation signal.
In this paper a technique is proposed that gives a good
control for the one link manipulator. An online control
algorithm is structured using the radial basis function
neural network (RBFNN). The one link manipulator
parameters are estimated on line and are used to update
the weights of the RBFNN. The weight update equations
are derived based on the least mean squares principle.
The RBFNN virtually models the inverse of the one link
manipulator and thus the output tracks the reference
trajectory. The self tuning regulator (STR) inspects with
the various disturbances.

DERIVATION OF EQUATIONS OF
MOTION FOR ACTUATED ONE-LINK
MANIPULATOR

Consider the single-link manipulator shown in Fig. 1.


It has one rotational joint. The mass is considered to be
located at a point at the distal end of the link, and so the
moment of inertia is ml 2 [21].

of gravitational potential energy possessed by an


elevated object is equal to the work done against gravity
in lifting it. Thus, for an object at height h , the
gravitational potential energy E p is defined by its mass

m , and the gravitational constant g :


E p mgh
(2)
Lagrange's equations employ a single scalar function,
rather than vector components. In classical mechanics,
the natural form of the Lagrangian is defined as
L Ek E p
(3)

L 12 mv 2 mgh
L 1 ml 2 2 mgl sin
2

(4)
(5)

By using Euler Lagrange equation:

d L L

dt

(6)

We shall now compute both sides of the Euler-Lagrange


equation

L
mgl cos

d L
ml 2

dt

L
ml 2

Now that we have both sides of the Euler-Lagrange


Equation
ml 2 mgl cos
(7)
Linearization for one-link manipulator
The linearization of system is based on the Taylor
series expansion and on knowledge of nominal system
trajectories and nominal system inputs [22]. Assume that
the motion of the nonlinear system is in the
neighborhood of the nominal system trajectory, that is

(t ) n (t ) (t )

(t ) n (t ) (t )
Where (t ) represents a small quantity. It is natural to

Fig. 1 An inverted pendulum or a one-link manipulator


The kinetic energy is defined as the work needed to
accelerate a body of a given mass from rest to its stated
velocity. In classical mechanics, the kinetic energy E k of
a point object is defined by its mass m and velocity v that
is a vector representing the change in position:
Ek 12 mv 2
(1)
The potential energy is the energy of an object or a
system due to the position of the body or the
arrangement of the particles of the system. The amount

assume that the system motion in close proximity to the


nominal trajectory will be sustained by a system input
which is obtained by adding a small quantity to the
nominal system input

(t ) n (t ) (t )
(t ) n (t ) (t )

Abstract

n f ( n ,n , n ,n )

The righthand side can be expanded into a Taylor series about the
nominal system trajectory and input, which produces

(t ) a1 (t ) a 0 (t ) b1 (t ) b0 (t )

Where
the corresponding coefficients are evaluated at the
nominal points as

www.ijsret.org

70

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

f ( n ,n , n ,)
0

f ( n ,n , n ,)
g
a0
sin n

l
f ( n ,n , n ,)
b1
0

f ( n , n , n ,)
1
b0
2

ml

Where g r is the diagonal matrix of reduction ratio. The


applicable control input for driving robot arm is the
armature voltage of the motors, here. So, by using
equations (8)-(12) and neglecting the inductance L ,
because of its tiny amount, the following equation is
achieved.

a1

1
1
V RKm ((J m g r g r ml 2 )
1
1
( Bm g r K m R 1 K b g r g r k 2 )

This system is linearized at the nominal points


n (t ) n 0 , n (t ) 2 producing a 0 g l
the linearized equation is given by

g
l

ml 2

Actuated one-link manipulator


The mathematical equations describing electrical and
mechanical dynamics of a permanent magnet DC motor
are as follows [24]:
(9)
(10)

K mi
(11)
Where V is the armature voltage of the motor, R and
L are armature equivalent resistance and inductance,
respectively,

K b is the back electromotive force

constant, i is the armature current and r denotes the


rotor position, J m is the total moment of inertia, Bm is
the damping coefficient, m and represent the
generated motor torque and the load torque, respectively,
and K m is the diagonal matrix of motor torque constant.
With the purpose of increasing motion speed of the
manipulators, motors are equipped with the high
reduction gears as follows:
q g r r
(12)

m g r

k1 , k2 are closed-loop poles that determine the value of


K R1n .

(8)

The linear system is unstable and controllable.


Therefore, the pole placement method is used to
determine a value of K R1n that will produce a
desired set of closed-loop poles. Ackermann's formula
can be used for pole placement. Ackermann's formula
(1972) is a direct evaluation method. It is only
applicable to SISO systems [23].
Generally the actuators used to move the joints of any
industrial robot are motors, usually DC permanent
magnet (PM) motors or AC (PM) motors. Other motors
can be used, including pneumatic or hydraulic servo
motors.

di
d
V Ri L Kb r
dt
dt

J m r Bm r m

g r (k1 mLg) )

(13)

PROPOSED STRUCTURE

Fig. 2 is the proposed structure. Autoregressive


with exogenous input (ARX) is used to identify the onelink manipulator and found the model. The model
coefficients are updated online depending on the onelink manipulator parameters variation. These
coefficients are fed the weight update block which trains
the controller whether RBFNN or PI controller using the
least mean square LMS algorithm.

Fig. 2: Proposed self-tuning robot arm regulator


structure
3.1 ARX model
The process is modeled by an ARX model [20],
whose output is given by
n

y (t ) ai y (t i ) b j x(t j ) (14)
i 1

j 1

Bq b1q b2 q bm q m
1

Aq 1 a1q 1 a2 q 2 an q n

Or in terms of q 1 operator

y (t )

B ( q 1 ) d
q x (t )
A(q 1 )

(15)

3.2 radial basis functions neural networks


A single input single output radial basis function
neural network (SISO RBFNN) is shown in Fig. 3. It
consists of an input node r (t ) , a hidden layer with n
neurons and an output node x(t ) . Each of the input

www.ijsret.org

71

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

nodes is connected to all the nodes in the hidden layer


through unity weights (direct connection). While each of
the hidden layer nodes is connected to the output node

w , , w

n0
through some weights 1
.
Each neuron finds the distance d of the input and its
center and passes the resulting scalar through
nonlinearity. So the output of the hidden neuron is given
by [10, 20]

(d ) exp( 12 d 2 )
exp( r (t ) ci )
2

1
2

(16)

( K 1) ( K )

I
( K )

the nonlinear basis function. Normally this function is


taken as a Gaussian function of width . The output
x(t ) is a weighted sum of the outputs of the hidden
layer, given by

(20)

I
W ( K 1) W ( K )
W ( K )
PI ( K 1) PI ( K )

r (t ) cn0
r (t ) c1

1
2

th
ci is the center of i
hidden layer node
where i 1,2,, n0 , is the norm matrix and (.) is

(21)

I
PI ( K )

(22)
Where a1 an b1 bm is the parameter
vector,
RBFNN,

W w1 w2 wn0

PI k p

is the weight vector for

k i vector for the parameters

is the learning

proportional-integral (PI) and

parameter. The variable K is used to show the iteration


number of training.
Keeping the regressions of the variables in the
system
in
a
regression
vector
as

(t ) (t 1) (t n) V (t d ) V (t m d )

n0

x(t ) wi ( r (t ) ci )

Where r (t ) is the reference input signal and (t ) is


the output joint position of the one-link manipulator. The
coefficients of the ARX model and the weights of the
RBFNN/parameters of the PI are updated in the negative
direction of the gradient as,

(17)

i 1

and

finding

partial derivatives.

As we see the radial basis function (RBF) network


utilized a radial construction mechanism. This gives the
hidden layer parameters of RBF networks a better
interpretation than for the multilayer perceptron network
MLP, and therefore allows new, faster training methods.

I 1 e 2 (t )

(23)
2

(24)
e(t ) (r (t ) (t ))

1
n
r (t ) a1q an q (t )
e(t )
b1q 1 bm q m q d V (t )
I
(25)
e(t ) (t )

The final parameter update equation will be,

( K 1) ( K ) e(t ) (t ) (26)

The partial derivatives for the weights are derived as


follows,
Fig. 3: A general RBF network
3.3 parameters estimation for controller
The parameters of the one-link manipulator model
are estimated online and are used to update the
coefficients of the controller (weights of the RBFNN /
parameters of PI). The weight/coefficient update
equations are derived based on a recursive scheme (least
mean squares principle). This previous parameters are
updated by minimizing the performance index I given
by [11]

I 12 e 2 (t )

e(t ) r (t ) (t )

(18)

I
1 e 2 (t )

(27)
W 2 W


B(q 1 ) d
r (t )
(28)
e(t )
q V (t )
1
W
A(q )

I
e(t ) B(q 1 )q d (t )
W

(29)

The final weight update equation will be,

W ( K 1) W ( K ) e(t ) B(q 1 )q d (t )
But the final coefficients update equation of PI will be,

PI ( K 1) PI ( K ) e(t ) B(q 1 )q d
(V (t 1) (t s 1)V (t ))

(19)

sample time.
www.ijsret.org

ts

is the

72

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

SIMULATION RESULTS
arm robot input & output

The proposed self-tuning regulator (STR) structure


designed to achieve trajectory tracking with minimum
steady state error and improving the dynamic behavior
(overshoot).
When utilizing the RBFNN, the actuated one link
manipulator has got an excellent output with a square
wave reference and small estimated parameters values,
Fig. 4. Secondly, the system gives a good tracking of
various sinusoidal reference trajectories and the model
parameters value fluctuates as seen; Fig. 5. Finally, the
output controller meets disturbance at specific time
period 250 t 300 and the output of self-tuning
regulator (STR) structure mimics the track at different
disturbances Fig. 6.

1
0.5
0
-0.5
-1

50

100

-5

-1

250

300

400

450

50

100

200

250

300

350

400

450

50

100

position (q)

150

450

500

200

250
time(sec)

300

350

400

450

500

350

400

450

500

350

400

450

500

350

400

450

500

-0.5
0

50

100

150

200

250

300

the error between the input and the output

x 10

0.5

500

500

50

100

150

the error between the input and the output

x 10

400

-1
150

350

0.5

100

300

arm robot input & output

50

250

-0.5

-10

200

(a)

150

10

robot arm output

-2

500

15

-1

350

450

20

1
200

400

parameters evolution

150

350

the error between the input and the output

x 10

-5

100

300

-0.5

a square wave input

50

250

200

0.5

Do it another way, the PI self tuning actuated one


link manipulator has got a good output with a square
wave reference and small estimated parameters values,
Fig. 7. Then, the system gives a good tracking of various
sinusoidal reference trajectories and the model
parameters value hardly change; Fig. 8. At last, the
output controller meets disturbance at specific time
period 250 t 300 and the output of self-tuning
regulator (STR) structure mimics the track at two
statuses low and high disturbances; the model
parameters value hardly change after the end of noise
Fig. 9.

-2

150

200

250

300

parameters evolution
20

0.5
0
-0.5

15
0

50

100

150

200

250

300

350

400

450

500

10

parameters evolution
50

0
-50

50

100

150

200

250
time(sec)

300

350

400

450

50

100

150

200

250
time(sec)

300

500

Fig. 4 the output of radial basis self-tuning actuated one


link manipulator for a square wave reference

(b)
Fig. 5 the output of radial basis self-tuning actuated one
link manipulator system for a different sinusoidal wave
reference

www.ijsret.org

73

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

a square wave input

a square wave input


2

-2

50

100

150

200

250

300

350

400

450

-1

500

50

100

150

position (q)

position (q)

250

300

350

400

450

500

300

350

400

450

500

350

400

450

500

350

400

450

500

2
0
-2

200

robot arm output

robot arm output

50

100

-10

150

200

250

300

350

400

450

0
-1

500

50

100

150

200

250

the error between the input and the output

the error between the input and the output

x 10

0.5
0

-0.5
0

50

100

150

200

250

300

350

400

450

-2

500

50

100

150

50

10.012

10.01

-50

50

100

150

200

250
time(sec)

300

200

250

300

parameters evolution

parameters evolution

350

400

450

10.008

500

(a)

50

100

150

200

250
time(sec)

300

Fig. 7 Simulation Results for PI self-tuning actuated one


link manipulator system

a square wave input


2

arm robot input & output


1

0.5
-2

50

100

150

200

250

300

350

400

450

500

robot arm output

-0.5

position (q)

-1
0

50

100

150

200

250

300

350

400

450

500

350

400

450

500

350

400

450

500

the error between the input and the output


-2

0.05
0

50

100

-10

150

200

250

300

350

400

450

500

the error between the input and the output

x 10

0.5
0
-0.5

-0.05
0

50

100

150

200

250

300

350

400

450

500

50

100

150

parameters evolution

200

250

300

parameters evolution

200

10.0106

10.0104
10.0102

-200

50

100

150

200

250
time(sec)

300

350

400

450

500

10.01
10.0098

(b)
Fig. 6 Simulation Results for radial-basis self-tuning
actuated robot system output disturbance (a) 0.05 sin(t )
(b) 0.5 sin(t )

www.ijsret.org

50

100

150

200

250
time(sec)

(a)

300

74

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

arm robot input & output

arm robot input & output

0.5

0.5

-0.5

-0.5

-1

50

100

150

200

250

300

350

400

450

-1

500

50

100

-3

the error between the input and the output


0.05

150

200

250

300

350

400

450

500

350

400

450

500

350

400

450

500

the error between the input and the output

x 10

0.5
0

0
-0.5

-0.05

50

100

150

200

250

300

350

400

450

-1

500

50

100

150

parameters evolution

200

250

300

parameters evolution

10.012

11
10

10.011

9
10.01
10.009

8
0

50

100

150

200

250
time(sec)

300

350

400

450

500

(b)
Fig. 8 the output of PI self-tuning actuated one link
manipulator system for a sine wave reference with
different frequency

100

150

200

250

300

350

400

450

the error between the input and the output

x 10

0.5
0
-0.5
-1

50

100

150

200

250

300

350

400

450

250
time(sec)

300

CONCLUSIONS

500

-0.5
50

200

In nonlinear and unstable system, the controller


construction mimics the trajectory of the reference. As is
noticed in aforementioned figures, the radial basis
function neural network is the best method and the
parameters evolution are smaller value than its
counterpart at the another method (PI) in most cases.

-3

150

500

0.5

100

This paper introduces a very simple structure for


control the actuated one link manipulator that updates
itself online. The exact model of the actuated one link
manipulator needs not to be known and just the
estimates are enough to drive the RBFNN as the process
inverse.

-1

50

Fig. 9 (b) Simulation Results for PI self-tuning actuated


one link manipulator system output disturbance
0.5 sin(t ) at 250 t 300

arm robot input & output

parameters evolution

The RBFNN is a fast neural network compared with


others type due to using least mean squares principle as
training algorithm. Its structure has 2 neurons in hidden
layer.

10.015
10.01
10.005
10
0

50

100

150

200

250
time(sec)

300

350

400

450

REFERENCES

500

Fig. 9 (a) Simulation Results for PI self-tuning actuated


one link manipulator system output disturbance
0.05 sin(t ) at 250 t 300

www.ijsret.org

[1] QIAOGE LIU, MENGYIN FU, AND


ZHIHONG DENG "adaptive inverse control for
nonlinear system". International journal of
information and systems sciences Volume 1,
Number 3-4, Pages 253-263, 2005.
[2] K. J. Astrom and B.Wittenmark. Adaptive
Control. Addison Wesley, 1995.
[3] Won Seok Oh; Kim Sol ; Kyu Min Cho ;
Kyungsang Yoo ; Young Tae Kim. Self-tuning

75

International Journal of Scientific Research Engineering & Technology (IJSRET), ISSN 2278 0882
Volume 4 Issue 2, February 2015

speed controller for induction motor drives.


Power
Electronics,
Electrical
Drives,
Automation and Motion (SPEEDAM), 2012
International Symposium.
[4] K.J. ASTROM, U. BORISSON, L. LJUNG, B.
WITTENMARK '' Theory and Applications of
Self-Tuning Regulators '' Automatica. Vol.13,
pp. 457-476. Printed in Great Britain
[5] K. Amini Khoiy, F. Davatgarzadeh, M. Taheri
"A Review on Fuzzy-Logic Method to Control
Robotic Manipulator Systems" Universal
Journal of Computational Analysis 1 (2013), 4047
[6] smail H. AKYUZ, Ersin YOLACAN, H.
Metin ERTUNC, Zafer BINGUL "PID and State
Feedback Control of a Single-Link Flexible
Joint Robot Manipulator" Mechatronics (ICM),
2011 IEEE International Conference on 13-15
April 2011 pp 409 - 414
[7] Rahmouni, A.; Lachiver, G. Optimal speed
tracking control of induction motor using
artificial intelligence techniques . Power
Electronics Specialist Conference, 2003. PESC
'03. 2003 IEEE 34th Annual.
[8] K. S. Narendra and K. Parthasarathy.
Identification and control of system using neural
networks. IEEE Transactions on Neural
Networks, 1:427, 1990.
[9] Kumar Rajesh, A.K.Ray. Artificial neural
network modeling and control of Retention
process in the wet end, International Journal of
Information Technology and Knowledge
Management July-December 2010, Volume 2,
No. 2, pp. 259-264.
[10]
S.
Haykin.
Neural
Networks:A
Comprehensive Foundation II. Macmillan/IEEE
Press, 1994, 1999.
[11]
Syed S., Hhssain A.,Muhamad M.
Radial Basis Functions Neural Network Based
Self-Tuning Regulator. WSEAS Transaction On
Systems, Issue 9, Volume 3, November 2004.
[12]
Vali U.,M .Yasir. Multiple Layer
Perceptron for Direct Inverse Control of a
Nonlinear System. Computer, Control and
Communication, 2009. IC4 2009. 2nd
International Conference.
[13]
Enzeng Dong, Shuxiang Guo, Xichuan
Lin, Xiaoqiong Li, Yunliang Wang A neural
network-based self-tuning PID controller of an
autonomous underwater vehicle, Mechatronics
and Automation (ICMA), 2012 International
Conference
on
5-8 Aug. 2012, pp 898 903.
[14]
Zhao Ximei , Sun Xianfeng. NeuralNetwork-Based Self-Tuning. PI Controller for
Permanent. Magnet Synchronous Motor,
www.ijsret.org

Electrical Machines and Systems (ICEMS),


2011 International Conference
[15]
K.sabahi.
Application of ANN
Technique for DC-Motor Control by Using FEL
Approaches.
pp.131-134,
2011
Fifth
International Conference on Genetic and
Evolutionary Computing, 2011.
[16] A.K.Pal, I.Naskar.
Design of Self-Tuning
Fuzzy PI controller in LABVIEW for Control of
a Real Time Process. International Journal of
Electronics and Computer Science Engineering.
Volume 2, Number 2, P.P 538-545, 2013.
[17] Kota, B.V.S Goud. Fuzzy PID Control for
Networked Control System of DC Motor with
Random Design.
International Journal of
Computer Applications (0975 8887) Volume
52 No. 7, August 2012.
[18] Saad Zaghlul.
Tuning PID Controller by
Neural Network for Robot Manipulator
Trajectory Tracking Al-Khwarizmi Engineering
Journal, Vol. 8, No. 2, P.P. 91-28 (2013).
[19] Alfonso, Vctor, Manuel, Hugo, Edgar,
Juvenal, Gilberto. A New Adaptive Self-Tuning
Fourier Coefficients Algorithm for Periodic
Torque Ripple Minimization in Permanent
Magnet Synchronous Motors (PMSM). Sensors,
Vol. 13, Pages 3831-3847, 2013.
[20] Oliver. Nonlinear System Identification From
Classical Approaches To Neural Networks And
Fuzzy
Models,Springer-Verlage
Berlin
Heidelberg 2001.
[21] Simpson Colin D. "Introduction to Robotics"
March 3, 2008.
[22] Zoran Gajic .Linear Dynamic Systems and
Signals,
Prentice
Hall
2003.
www.ece.rutgers.edu/~gajic/psfiles/linearization
.pdf
[23] Roland S. Advanced control engineering. first
published 2001
[24] S. Ehsan, advanced strategies for robot
manipulators. 2010

76