Professional Documents
Culture Documents
Terminal Sliding Mode Control With Evolutionary Algorithms For Finite-Time Robust Tracking of Nonholonomic Systems
Terminal Sliding Mode Control With Evolutionary Algorithms For Finite-Time Robust Tracking of Nonholonomic Systems
ITC 1/47 Terminal Sliding Mode Control with Evolutionary Algorithms for
Journal of Information Technology
Finite-Time Robust Tracking of Nonholonomic Systems
and Control
Vol. 47 / No. 1 / 2018 Received 2017/05/11 Accepted after revision 2018/01/22
pp. 26-44
DOI 10.5755/j01.itc.47.1.15031
© Kaunas University of Technology http://dx.doi.org/10.5755/j01.itc.47.1.15031
This paper deals with utilizing a recursive fast terminal sliding mode control method for finite-time robust
tracking in a class of nonholonomic systems described by an extended chained form of differential equations.
To enhance the performance of the proposed method, the constrained parameters of the controller are exactly
tuned using evolutionary algorithms such that the tracking error reaches zero in a short time while chatter-
ing is significantly reduced. A comparative study is also presented among the applied evolutionary algorithms,
namely, differential evolution, bat optimization, cuckoo optimization and bacterial foraging optimization. Ap-
plying the proposed design method leads to a considerable reduction in convergence time of the states as well
as the chattering phenomenon. It is shown that the method is robust against disturbance in the input of the sys-
tem. Numerical simulations for a well-known nonholonomic system, i.e., wheeled mobile robot demonstrate
effective improvement in the results compared with conventional terminal sliding mode control method.
KEYWORDS: Nonholonomic systems, Terminal sliding mode control, Finite-time tracking, Evolutionary al-
gorithms, Chained form.
1. Introduction
An important class of general nonlinear systems is These systems have many applications in the fields
nonholonomic (NH) systems which have been stud- of mechanics, mobile robotics, electro-magnetics
ied as classical mechanical systems for many years. and electromechanics. In such systems, there exists
Information Technology and Control 2018/1/47 27
a special type of conditions which restricts their mo- sliding surfaces and also some improvements by using
tions. Some examples of the NH systems are sledg- adaptive or intelligent methods have been proposed in
es that slide on a plane, wheels and spheres that roll the literature [13, 22, 23]. Such mechanisms are useful
without slipping on a plane, mobile cars and car-like for improving the finite-time convergence of the con-
vehicles, wheeled mobile robots (WMRs), knife-edge, trol systems. In addition, nonsingular versions of the
under-actuated satellites, surface vessel and space TSMC method have also been studied in [7, 11] to tack-
robots [1, 3, 10, 14, 16]. le the problem of singularity in TSMC.
In recent years, issues related to the control of the NH Utilizing SMC for controlling the NH systems have
systems have received much attention. Due to the fact been reported in several papers [2, 6, 9, 21, 24, 34, 36].
that Brockett’s necessary condition cannot be satis- In these works, backstepping-based second-order
fied for the NH systems [4], such systems cannot be SMC [9], backstepping-based adaptive SMC [6] have
stabilized by smooth or even continuous stationary been used to control the NH systems in presence of
feedback control laws while they are controllable. In uncertainties. Moreover, in [2, 21, 24, 34, 36], TSMC
addition, such systems are highly nonlinear and they has been proposed for stabilizing the NH systems
are also generally not invertible, and thereby design- with chained form of state equations. In addition, the
ing controller for stabilizing or tracking is not an easy intelligent control strategies such as fuzzy and neural
task. Therefore, different control methods based on controllers and their combinations with SMC-based
nonlinear control strategies such as discontinuous methods have also been widely proposed for trajecto-
control methods, time varying control methods, the ry tracking problems [5, 8, 15, 25-27, 32, 42].
methods based on system conversion, sliding mode Despite the advantages of the SMC-based control
control methods, hybrid control methods and so on, methods developed for the NH systems, most of them
have been devoted to the problem of stabilizing or possess low speed of convergence and tracking error
tracking in the NH systems [12, 16, 17, 20, 33, 37, 38]. takes a long period of time to reach zero. Even in some
One of the important methods for controlling the NH cases, tracking error does not reach zero, especially in
system is based on system conversion into subsys- presence of uncertainty and disturbance. Therefore,
tems. For instance, Ploeg et al. in [29] have proposed a the problem of robust finite-time tracking of the NH
position control method based on feedback lineariza- systems with chained form in presence of disturbanc-
tion, in which multi-cycle robots have been converted es and uncertainties is still an open challenge in the
to a set of identical unicycles. Whenever each unicy- literature. Moreover, making improvements in the
cle is controlled, as a consequence, the whole NH sys- SMC-based control methods in the NH systems, by
tem position is controlled. Even though their method using some supervisory mechanisms such as intel-
has some advantages, uncertainty in the dynamics ligent control algorithms, can be an open challenge
has not been considered. which motivates the current research. To address
NH systems can also be described in a chained form, these problems, this paper uses a TSMC method, in
and many control strategies have been proposed based which the main targets are to minimize the tracking
on this form [16, 20, 21, 34]. WMR is a well-known ex- error and to maximize the convergence speed.
ample of the NH systems that can be stated in chained In this paper, we utilize a recursive TSMC for con-
form with an appropriate coordinate transformation. trolling the chained-form NH systems in order to min-
Such transformations are useful tools for making the imize the tracking error in these systems. Despite the
system suitable for applying control strategies in order effectiveness of the recursive TSMC, there are some
to force the system to move along a desired trajectory. constrained parameters in the structure of recursive
Sliding mode control (SMC) method, especially, ter- terminal sliding surface that should be determined
minal sliding mode control (TSMC) is a discontinuous accurately in order to reach such control objectives as
control technique that has attracted the attention of tracking and the convergence of the states. Exact ad-
many researchers in recent years due to its superior fea- justment of such parameters leads to better tracking
tures like robustness, easy implementation, finite-time results and increases the convergence speed. The idea
convergence and high precision performance [18, 19, studied in this paper is to determine such parameters
35, 41]. Recently, new types of TSMC with different by using the intelligent algorithms. For this purpose,
28 Information Technology and Control 2018/1/47
an intelligent TSMC is proposed, in which the TSMC where u1 and u2 are control inputs of the system and
is combined with evolutionary algorithms. By using x = [ x1 ,..., xn ]T is the state vector of the system.
such intelligent optimization methods, the recursive The desired dynamics, which is to be tracked by the NH
terminal sliding structure is improved by reducing system, is considered as:
the convergence time of tracking error for the NH
systems described in an extended chained form. Four x1d = u1d
intelligent algorithms, namely, differential evolution x2 d = u2 d
(DE), bat optimization (BO), cuckoo optimization
x3d = x2 d u1d (2)
(CO) and bacterial foraging optimization (BFO) are
studied and compared in this paper. Such strategies
are separately used to adjust the TSMC in order to xnd = x( n −1) d u1d ,
minimize the tracking error of the NH system. Input
disturbance is also considered in this paper, and it is where u1d and u2d are reference control inputs. Now,
shown that the convergence is guaranteed in pres- consider the following dynamic extension of system
ence of the disturbance. The simulation results show (1) in order to solve the tracking control problem:
the effectiveness of the proposed method and it will
be shown that by using the proposed method, a bet- u1 = f1
(3)
ter tracking error can be achieved compared with the u2 = f 2 ,
conventional TSMC.
The paper is organized as follows. In Section 2, the where u1 and u2 are dynamic extension of the sys-
dynamic model and problem formulation are de- tem and f1 , f 2 are adjustable inputs of the dynamical
scribed. The main proposed controller is introduced model. Let xe= x − xd be the tracking error. The track-
in Section 3, in which a recursive TSMC is applied to ing error dynamics can be obtained as:
the NH system in order to achieve finite-time track-
ing control of the desired output in presence of ex- x1e= u1 − u1d
ternal disturbances. The intelligent optimization x 2=
e u 2 − u 2d
algorithms are also utilized in this section to adjust x 3e = x 2e u1d + x 2 (u1 − u1d )
the coefficients of the TSMC in order to improve the
(4)
performance of the closed loop system. Section 4 il-
lustrates the application of the proposed method to x ne = x ( n −1)e u1d + x n −1 (u1 − u1d )
a nonholonomic WMR as a well-known benchmark u1 = f 1
example. Simulation results are illustrated in Section u2 = f 2 .
5 to show the applicability and effectiveness of the
proposed method. Finally, Section 6 gives some con- The objective is to design control inputs, f1 and f 2 ,
clusion remarks. such that the states of the system track the desired dy-
namics of the reference system (2) and control strat-
egy makes the tracking error to reach zero in a short
2. The Problem Statement period of time.
is considered as two subsystems. The first subsystem ry starting from any initial state will reach the sliding
can be described by: mode s1 = 0 in finite-time and u1 = u1d. Therefore, the
convergence of tracking error for the subsystem (5)
x1e= u1 − u1d will be guaranteed.
(5)
u1 = f 1 . In order to overcome the singularity problem in
this recursive method, the initial value should be
The first subsystem (5) has been described in the ca- defined carefully to avoid trajectory from reaching
nonical form of nonlinear SISO system. To design the si 0=
= (i 1,..., j − 1) before s j = 0 is reached. Hence,
fast terminal sliding control law for this subsystem, if the initial value x1e (0) and u1 (0) is defined such
according to the conventional procedure of TSMC that s0 > 0 and s1 > 0 , then the switching manifolds
studied in the existing references such as [2, 21, 34- s1 and s0 reach zero sequentially and the singularity
36], recursive sliding surfaces should be defined as: problem will not occur.
The second subsystem can be described as follows:
s0 = x1e
s= s0 + β s0 q / p , (6)
1
x 2=
e u 2 − u 2d
where p and q are odd positive integers satisfying x 3e = x 2e u1d + x 2 (u1 − u1d )
q < p , and β is a positive constant. For the first sub- (10)
system, the recursive sliding surfaces are selected as
x ne = x ( n −1)e u1d + x n −1 (u1 − u1d )
(6) and the control law is taken as:
u2 = f 2 .
q q q′
q −1
f1 = u1d − β x p
1e (u1 − u1d ) − k1 (u1 − u1d + β x ) ,
p
1e
p′
(7) By changing variables of the second subsystem, the
p
system can be written as classical form so that the
where k1 > 0 and p ′ and q ′ are odd positive integers TSMC design can be applied.
such that q ′ < p ′. Therefore: Considering y1 = xne , y2 = x( n −1) e , … , yn −1 = x2 e ,
y=
n u 2 − u 2d , and assuming u u
1 = 1d , the subsystem (10)
d can be transformed into the following form:
s=
1 s0 + ( β s 0q / p )
dt
q y1 = u1d y 2
= s0 + β s 0q / p −1s0
p y 2 = u1d y 3
q
= x1e + β s 0q / p −1x1e (8) (11)
p
d q y n −1 = y n
= (u1 − u1d ) + β s 0q / p −1 (u1 − u1d )
dt p y n = f 2 .
q q / p −1
=f 1 − u1d + β s 0
(u1 − u1d ).
p For this system, the following recursive fast terminal
sliding surface structure is used to design the control
Thus, using the control input in the form of (7), we law f 2 [35]:
have:
s0 = y1
s1 + k 1s1q ′ / p ′ =
0. (9)
s=
1 s0 + β1 s0 q1 / p1
(12)
It can be proven that the solution s1 of this equation
will reach zero in finite-time, i.e., the equilibrium s= sn − 2 + β n −1 snq−n−21 / pn−1 ,
n −1
s1 = 0 is a fast terminal attractor. Based on the recur-
sive structure of (6), s0 = 0 will be reached in finite where pi and qi are odd positive integers satisfying
time and thereby the state errors, x1e will reach zero qi < pi and β i are positive constants
= (i 1,..., n − 1) .
in finite time. It means that the system state trajecto- Equation (12) can be rewritten as:
30 Information Technology and Control 2018/1/47
s nq n−1/ p n n −1
d n − i q i / pi
M1 s=
n −1 s 0( n ) + ∑ β i (s i −1 ). (27)
k1 = q ′/ p ′
+ η1 , η1 > 0 i =1 dt n −i
s1
n −1 Considering f 2 as (17), it can be obtained that:
M 2 u1d (22)
k=
2 qn / pn
+ η2 , η2 > 0
s n −1
−k 2 s nq n−1/ p n + u1nd−1 (Df 2 ).
sn −1 = (28)
p ′/q ′
M
=Ω1 y : s n −1 ≤ 1 Thus,
k1
M u n −1
pn /qn
sn −1 = −γ 2 s nq n−1/ p n . (29)
=Ω 2 y : s n −1 ≤ 2 1d .
k2
To be able to have the fast terminal convergence, we
have to prove that γ 2 > 0 . For this, according to the
Proof. For the first subsystem described by (20), tak- condition (22), we have:
ing derivate of s1 defined in (6), similar to (8), we can
obtain: M u
n −1
(Df )u n −1
η2 + 2 q1dn / pn − q2n / pn1d
γ2 =
d
q
s n −1 s n −1
s=
1 s0 + (β s 0 p ) (30)
dt q M 2 u1d
n −1
Df 2 u1nd−1
d q −1
≥ η2 + − ≥ η 2 > 0.
= (u1 − u1d ) + β s 0 (u1 − u1d ) p
(23) s n −1
qn / pn
s n −1
qn / pn
dt p
q
q p −1
= f 1 − u1d + β s 0 (u1 − u1d ) + Df 1 . It means that the conditions in (22) guaran-
p
tee that γ 2 ≥ η 2 > 0 . Thus, it can be seen that the
Considering f1 as (7), it can be obtained that error will reach the region Ω 2 constrained by
n −1
sn −1 ≤ ( M 2 u1d / k2 ) qn / pn in finite time. Hence the
−k 1s1q ′ / p ′ + Df 1 .
s1 = (24) proof of theorem has been completed.
In the proposed controllers in (7) and (17), instead of
Thus:
sign function, k1 and k2 have to be selected as (22) in
Theorem 2 in order to guarantee the robustness. Sim-
s1 = −γ 1s1q ′ / p ′ . (25)
ilar to switching control inputs like sign functions,
it has been proven in Theorem 2 that the proposed
To prove the fast terminal convergence, we must
method can compensate the effects of disturbances
prove that γ 1 > 0 . According to the condition (22), we
and the state errors will reach the neighborhood of
have:
zero in finite time. In fact, by replacing k1 and k2 in
(22), into (7) and (17), it can be seen that these con-
M1 Df 1
γ1 =
η1 + q ′/ p ′
− trollers act similar to sign function and can suppress
s1 s1q ′ / p ′ the disturbances. In this method, the existence of η1
(26)
M1 Df 1 and η 2 plays an important role to reduce the chatter-
≥ η1 + q ′/ p ′
− q ′/ p ′
≥ η1 > 0. ing. On the other hand, increasing the parameters k1
s1 s1
and k2 leads to increasing the gains γ 1 and γ 2 which
causes faster convergence of terminal attractors
Therefore, the conditions in (22) guarantee that
s1 = −γ 1 s1q′ / p′ and sn −1 = −γ 2 snq−n 1/ pn .
γ 1 ≥ η1 > 0 . Thus, the error will reach the region Ω1
32 Information Technology and Control 2018/1/47
3.3. Intelligent Tuning of Parameters netic algorithm (GA). However, the main difference is
its exclusive way to produce new population.
The error dynamics of the NH system reach zero in
a finite period of time if the parameters are designed Three main operators in this algorithm are crossover,
properly. Therefore, the design parameters should mutation and selection. DE algorithm changes the or-
be chosen carefully, otherwise, it may take very long dering of operators used in this algorithm. In addition,
time for tracking error to reach zero or tracking er- the way of using mutation is exclusive. The algorithm
ror may fluctuate with the amplitude around zero is explained in the following order:
without reaching it, as it will be shown in simulation Parameter Selection. The problem formulation in-
results presented in Section 5. We separately utilize cludes the objective function, the unknown variables
four intelligent optimization algorithms, namely, DE, or parameters and decision parameters which have to
BO, BFO and CO to design these parameters such that be defined at the first stage. The ranges of unknown
an enhanced tracking error can be achieved. Figure 1 parameters have to be also determined. In addition,
shows the block diagram of the main problem. the parameters of the algorithm such as the popula-
tion size and scaling as well as crossover factors have
Figure 1 to be chosen.
A block diagram of the proposed method. Initialization. The initial population with Np vectors
is produced randomly in the acceptable range defined
for parameters:
Error Dynamics of xe
NH System x j , g =b j , L + rand j (0,1) × (b j ,U − b j , L )
(31)
for j 0,1,...,N p − 1
=
f1 and f2
TSMC where bj,L and bj,U are lower and upper limit vectors for
parameters unknown parameters, respectively. xj,g is the base vec-
tor, g denotes the generation and randj(0,1) produces
Intelligent
Algorithm
a random number with normal distribution in the in-
terval of (0,1].
ISE Mutation. Differential mutation is a difference vec-
tor that is being sampled randomly from the initial
Cost Function population. If vi is the target vector, the mutation vec-
tor is produced with the following equation:
v i , g= x r0 , g + F × (x r1 , g − x r2 , g ), (32)
In this paper, our goal is to design the parameters pi ,
qi ( i = 1,..., n ), β= where x r0 , g (base vector) and x r1 , g − x r2 , g (difference
i (i 1,..., n − 1 ) and k2 considering
the constraints mentioned above for each parameter vector) are selected randomly from initial population
by using evolutionary algorithms such that the tracking considering r0 ≠ r1 ≠ r2 ≠ i. Moreover, F is a scaling fac-
errors reach zero in finite time. In the following subsec- tor which is a strictly positive real number that varies
tions, it will be illustrated how such algorithms work. in the interval (0 , 1].
Crossover. Each current vector combines with muta-
3.3.1. Differential Evolution (DE) Algorithm tion vector and produces a new temporary response.
DE algorithm has been proposed by Storn and Price In this paper, we use uniform crossover with the fol-
in 1995 [31]. They have used this algorithm to solve lowing equation:
continuous optimization problems, and recently it
has been used for integer, discrete and other type of v j,g
if rand j (0,1) ≤ Cr or j =
jrand
engineering problems. This algorithm has many sim- u j,g = , (33)
x j , g
O.W.
ilarities with other intelligent algorithms, such as ge-
Information Technology and Control 2018/1/47 33
where Cr ∈ [0 , 1] is user defined crossover probabil- Step3. Update the frequency, velocity and position:
ity and uj,g is called trial vector. In fact, (33) implies
that trial vector is constructed at the randomly cho- fi =f min + β ( f max − f min )
sen parameter index, jrand, which implies that the tri- vi =v i + f i (xi − x best ) (36)
al vectors are inherited from the mutant vector until
x= xi + v i ,
randj (0,1) ≤ Cr. The first time that randj (0,1) ≤ Cr, all i
erned by four processes, namely, chemotaxis, swarm- If moving condition is satisfied, i.e.,
ing, reproduction and elimination-and-dispersal. All J(i, j+1, k, l) < Jlast, then move and let
the steps of this algorithm are explained as follows:
Jlast=J(i,j+1,k,l).
Initialization: In this step, the parameters of the al-
Otherwise and also if m=Ns, swimming loop is done.
gorithm, i.e., s, p, Ns, Nc, Nre, Ned, ped that are respectively
__ If i ≠ s, go to the next bacterium and repeat the
the number of bacteria, problem dimensions, number
above steps.
of swims, number of chemotactic steps, reproduction
steps, elimination-dispersal steps, elimination-dis- __ If j < Ns, go to step 3; otherwise moving loop ends.
persal probability are initialized. Moreover, C(i), for Step5. Reproduction:
i=1,2,…,s, is defined as chemotactic step size of each a For the given k, l and for each bacterium, sort bac-
bacteria and θi is considered as initial position of all teria based on the objective evaluation function.
bacteria.
b A half of the bacteria with worst value dies and
Main Loop: The main loop consists of the following the other half with better value is divided into two
steps: parts in the same position.
Step1. Elimination-dispersal loop: l = l + 1, c If k < Nre, go to step 2.
Step2. Reproduction loop: k = k + 1, d Otherwise, end the reproduction loop.
Step3. Chemotactic loop: j = j + 1, Step6. Elimination-dispersal loop: l = l + 1.
Step4. For each bacterium, i = 1, 2, … , s take the che- a If condition of elimination and dispersal is satis-
motactic step as follows: fied, the bacterium is eliminated and replaced with
_ _ Calculate the evaluation function and add the
another bacterium that is produced randomly.
attractant effect to the nutrient concentration:
b If l < Ned, go to step 1.
c Otherwise, end the elimination-dispersal loop.
J (i, j , k , l ) = J (i, j , k , l )
+ J cc (θi ( j , k , l ), P ( j , k , l ) ) (39) 3.3.4. Cuckoo Optimization (CO) Algorithm
J last = J (i, j , k , l ), In 2009, research about cuckoo search has been de-
veloped by Yang and Deb [40]. Since then, CO algo-
where J is the objective function that has to be rithm has been presented by Rajabioun in 2011 [30].
minimized, and Jcc is the cell-to-cell attractant ef- The main reason that motivates researches to devel-
fect to the nutrient concentration. In addition, P op this optimization algorithm is that cuckoo has a
represents the position of each member in the pop- different life style in comparison with other birds. CO
ulation of s bacteria at the j-th chemotactic step, algorithm starts with random initial population simi-
k-th reproduction step, and l-th elimination-dis- lar to the other optimization algorithms. Each cuckoo
persal event. lays a random number of eggs. Hence, the population
_ _ Tumble: Generate a random vector, D(i), on [-1, 1]. is divided into two types, cuckoos and eggs. Some kind
of effort occurs between this population and the bet-
_ _ Move: Obtain new position by: ter cuckoos survive in this competition. These better
cuckoos immigrate to a better environment, start re-
D(i ) production and laying eggs. This cycle continues until
θi ( j + 1, k , l=
) θi ( j , k , l ) + C (i ) . (40)
T
D (i )D(i ) the stopping criteria are satisfied. Steps of this algo-
rithm are as follows:
_ _ For this position, evaluate the objective function Step1. Generate the initial habitats for cuckoos ran-
J(i, j+1, k, l). domly in the search space.
_ _ Swim: Step2. Dedicate a random number of eggs to each of the
Let m = 0 (counter for swim length). initial cuckoo habits (in nature it is between 5 and 20).
While m < Ns: Step3. Define the egg laying radius (ELR) for each
Let m = m + 1. cuckoo with the following equation:
Information Technology and Control 2018/1/47 35
other host birds’ nest inside its corresponding ELR. ISE ( x ie2 ( ))d . (44)
t0 i 2
Step5. Those eggs that are verified because of lesser
similarity to the host birds’ eggs will be killed by the In order to choose the parameters that are deter-
host birds. mined by the intelligent optimization algorithm, we
Step6. Let eggs hatch and grow. The first-born cuck- make a change in this objective function. For tracking
oo ruins other eggs because of its three times larger purposes, occurring peaks at the beginning are preva-
body; it eats more food and other chicks die from star- lent, while peaks at the end are not desirable. In order
vation. Hence, we have only one cuckoo remaining in to avoid the integration of these two peaks in the ob-
each nest in the end. jective function, they should be separated. Therefore,
Step7. Evaluate the habitat of each newly grown we consider the integration of the square error from a
cuckoo. short time after the beginning. In fact, for simulation,
Step8. Determine the maximum number of cuckoos we take t0 = 0.4 and t = 10 in (44).
that can live in each environment and kill those that The value of ISE obtained from (44) is allocated to
live in the worst habitats. each parameter vector. It is obvious that the vector
Step9. Cluster cuckoos, find the best group and se- with less value of the allocated ISE is more successful
lect the best of it as a target habitat. in the algorithm. We continue the algorithm until the
algorithm gets us the parameter vector with the best
Step10. Immigrate toward the target habitat:
objective function value.
x next =x current + F (x goal − x current ), (42)
Figure 2
x1 = θ
A nonholonomic WMR with two driving wheels and a
passive wheel − x sin θ + y cos θ
x2 =
=x3 x cos θ + y sin θ (47)
u1 = ω
Y
u2 = v − x3ω ,
where ( x2 , x3 ) denotes the coordinates of the center
of mass, x2 is the axis aligned with the vehicle orien-
y tation. Moreover, u1 and u2 are the control inputs
which depend on the pushing force T1 , steering torque
T2 and linear and angular velocity as defined in (48):
u1 0 1 T 1 0 2
T + −x ω .
−1
x
O X =u 1 −x M (48)
2 3 2 2
Figure 3
a) Tracking error of x1e, input u1 of first subsystem (5), b) control law f1 defined in (7)
3 6
x1e
2.5 4
u1
2
2
1.5
0
1
f1
-2
0.5
-4
0
-6
-0.5
-1 -8
-1.5 -10
0 2 4 6 8 10 0 2 4 6 8 10
Time(second) Time(second)
f1
-2
0.5
-4
0
38 Information Technology and Control
-6
2018/1/47
-0.5
-1 -8
20 2
0 1
x 2e
x 3e
-20 0
-40 -1
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (second) Time (second)
5
a b
x 10
2000 2
1000
0
0
u2
v
-2
-1000
-2000 -4
0 0.5 1 1.5 2 0 0.5 1 1.5 2
Time (second) Time (second)
c d
Now, we apply each algorithm presented in Section (KNN) clustering method with three clusters. In ad-
3 to find the parameters of the terminal sliding mode dition, we take α = 5, F = 9.
surface structure proposed to control the second Considering disturbance for the second subsystem de-
subsystem of WMR. For each algorithm, in order to fined in (51) as Δf = sin(t), simulation results of con-
2
compare the performance of the algorithms, the same trolling second subsystem with TSMC tuned using
objective function is used which is integral square er- these four algorithms and by considering the input
ror, ISE, defined in (44). Each algorithm needs its own
disturbance have been depicted in Figures 5 and 6. In
preliminary parameters to be defined. These parame-
addition, Table 1 compares the results of applying the
ters are taken into consideration as discussed below:
intelligent algorithms to the system by considering the
DE: For simulation use, the population size is as- number of function evaluations (NFE), consumed time
sumed as 50, and the scaling factor is considered ran- for running the algorithms and the best cost of ISE.
domly in the range of 0.2 < F < 0.8 with uniform distri-
It can be seen that by applying the TSMC in which the
bution. We additionally take Cr = 0.2 as the crossover
probability. parameters of terminal sliding surface structure are
chosen using the intelligent algorithms (Figures 5 and
BFO: For this algorithm, we assume that s = 50, Nc =
6), the results are considerably improved compared
10, Ns = 4, Nre = 4, Ned = 2 and Ped = 0.25. 11
with the case in which the parameters have been se-
BO: We use 30 bats with fmin = 0, fmax = 2. Other parameters lected randomly (Figure 4). Simulation results show
are selected randomly in the range stated in Section 3. that the intelligent TSMC is robust against distur-
CO: For simulating this algorithm, a number of 20 bance, i.e., applying disturbance to the system does
cuckoos is used. There are between two to five eggs. not affect the tracking error dynamics of the system.
For clustering, we use k-nearest-neighborhood By using the intelligent algorithm, the tracking errors
Information Technology and Control 2018/1/47 39
Figure 5
a) Tracking error of x2e, b) tracking error of x3e, c) dynamic control law u2 and d) WMR linear velovity v when the parameters
of the proposed method are selected using intelligent algorithms H. Ghasemi, B. Rezaie and Z. Rahmani
20 2
DE
1.5 H. Ghasemi, B. Rezaie and Z. CO
Rahmani
0 BO
BFO
DE 1
3e
-20
x 2e
CO
x
20 BO
0.5
2
BFO
-40 DE
0 1.50 CO
BO
BFO
-60 DE -0.5
1
-20 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
3e
x 2e
CO
Time(second) Time (second)
x
BO
BFO
0.5
-40 a b
2000 06
DE
-60
1000 -0.5 CO
40
a) 0 0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
BO
0 62
2000
u2
v
DE
DE
CO
-1000
1000 BO
40
CO
BO
BFO
BFO
-2000
0
u2
2-2
v
3000
1000
DE
CO
20000
f2
BO
BFO
1000
-1000
0
-2000
f2
-1000
-3000
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (second)
-2000
-3000
0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time (second)
40 Information Technology and Control 2018/1/47
Table 1
The results of using intelligent algorithms with or without disturbance for WMR. Parameters are a vector with the form of
[β1, β2, k2, q1, p1, q2, p2, q3, p3] that shows the simulation outputs of the algorithms
CO 3516 52 min 0.01138 [21.7, 10.6, 1.6, 49, 41, 65, 97,9, 39]
DE 2050 40 min 0.01139 [11.7, 20.8, 1, 177, 217, 169, 197, 5, 271]
BO 1230 24 min 0.0119 [14.3, 15, 54.1, 275, 299, 235, 267, 113, 177 ]
BFO 8000 3 h 54 min 0.04 [22, 16, 148, 219, 261, 219, 261,5, 131]
converge to zero very quickly in less than a second. In In order to see how tuning parameters affects the tra-
addition, the control effort has been reduced by using jectory tracking, Figures 7 to 10 show the effectiveness
the intelligent control scheme. of each algorithm in tracking the desired trajectory. It
Figure 7
Tracking the desired trajectory of x by four algorithms
Desired Trajectory
23
BFO
DE
CODesired Trajectory
BOBFO
12 DE
CO
BO
01
x
-1 0
x
-2-1
-2
-3
0.5 1 1.5 2 2.5 3
Time (second)
-3
0.5
Figure 7. Tracking1 the desiredTime 1.5
trajectory of x by four2 algorithms.
(second)
2.5 3
Figure 8
Figure 7. Tracking the desired trajectory of x by four algorithms.
Tracking the desired
4 trajectory of y by four algorithms
34
23 Desired Trajectory
DE
CO
BO
12
Desired Trajectory
y
BFO
DE
CO
BO
01
y
BFO
-1 0
-2-1
0.5 1 1.5 2 2.5 3
Time (second)
-2
Figure
0.5 8. Tracking
1 the desired trajectory
1.5 of y by four
2 algorithms. 2.5 3
Time (second)
Figure 9
Tracking the desired trajectory of θ by four algorithms
4.5
5
4
Desired Trajectory
4.5
3.5 Four Algorithms
4
3 Desired Trajectory
3.5 Four Algorithms
2.5
3
2
2.5
1.5
2
1
1.5
0.5
1
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
0.5
Time (second)
0
0 0.5 Figure19. Tracking
1.5 the desired
2 trajectory
2.5 of 3 by four 3.5
algorithms.
4 4.5 5
Time (second)
Figure 10
15 tracking of WMR inFigure
Trajectory 9. Tracking
Cartesian the desired
coordinate using trajectory
four 15 of by algorithms
intelligent four algorithms.
a) BFO, b) BO, c) DE, d) CO
10 10
15 15
5 5
10 10
0 0
y
y
5 5
-5 -5
0 0
y
y
a)-15 b)-15
-15 -10 -5 0 5 10 15 -15 -10 -5 0 5 10 15
15 x x
15
a b
10 10
15 15
5 5
10 10
0 0
y
5 5
-5 -5
0 0
y
-10 Trajectory
-10
Trajectory
CO TSM
-5 DE TSM
-5
-15 -15
-10-15 -10 -5 0 5 10Trajectory 15
DE TSM
-10-15 -10 -5 0 5 10 Trajectory
CO TSM
15
x x
c) d)
-15 -15
-15 -10 -5 0 5 10 15 -15 -10 -5 0 5 10 15
x x
c d
42 Information Technology and Control 2018/1/47
can be seen from these figures that using the parame- WMR. Table 1 also reveals that BFO leads to a greater
ters of TSMC obtained from BFO and BO algorithms, objective function and it takes much time compared
the desired trajectory is tracked with a little fluctua- with the others. Therefore, it could not find the best
tion. However, using DE and CO algorithms, the pa- parameter values. However, CO algorithm tunes the
rameters are properly tuned and therefore the trajec- parameters of terminal sliding surface such that leads
tory is accurately tracked in a short period of time. to the least ISE as objective function. Thus, it can be
concluded that CO algorithm is the best algorithm for
Simulation results show that among these four al- tuning these parameters. The tracking error and the
gorithms, by using CO algorithm, the best parame- state variables, when CO algorithm is used for select-
ter values can be obtained for trajectory tracking of ing the parameters, have been shown in Figure 11.
Figure 11
a) Error of transformed system x1e, x2e, x3e, b) tracking of desired trajectory x, c) tracking of desired trajectory y, d) tracking
of desired trajectory θ when the parameters of terminal sliding surface are selected using CO algorithm
5 5
x
x
d
0 0
x
1e
x
2e
-5 x -5
3e
-10 -10
0.2 0.4 0.6 0.8 1 0 1 2 3 4 5
Time (second) Time (second)
a b
5 2.5
2
0
1.5
-5
y 1 d
y
d
-10
0.5
-15 0
0 1 2 3 4 5 0 0.5 1 1.5 2 2.5
Time (second) Time (second)
c d
6. Conclusion
In this paper, an intelligent TSMC has been proposed terminal sliding surface structure designed for error
for controlling the NH systems in extended chained dynamic tracking. Disturbance has also been consid-
form. In this method, intelligent algorithms have been ered in tracking dynamic error of the NH system. BFO,
used to determine the parameters of the recursive BO, DE and CO are the optimization algorithms, used
Information Technology and Control 2018/1/47 43
for designing the parameters. It has been shown that, show that CO algorithm has provided the most effec-
using these intelligent algorithms, appropriate param- tive tuning of parameters among other algorithms
eters can be found such that the system can track the such that the tracking error can reach zero in a very
desired trajectory very well. Simulation results per- short period of time with a small amount of chattering.
formed on WMR as a benchmark of the NH systems
have shown the effectiveness of the proposed design Acknowledgments
for error dynamic tracking even in the presence of We would like to present our thanks to anonymous re-
the input disturbance. In addition, simulation results viewers for their helpful suggestions.
References
1. Astolfi, A. Discontinuous Control of Nonholonomic 10. Jiang, Z. P. Robust Exponential Regulation of Non-
Systems. Systems and Control Letters, 1996, 27(1), 37- holonomic Systems with Uncertainties. Automatica,
45. https://doi.org/10.1016/0167-6911(95)00041-0 2000, 36(2), 189-209. https://doi.org/10.1016/S0005-
2. Binazadeh, T., Shafiei, M. H. Nonsingular Terminal 1098(99)00115-6
Sliding-Mode Control of a Tractor–Trailer System. Sys- 11. Jin, M., Lee, J., Chang, P. H., Choi, C. Practical Nonsin-
tems Science and Control Engineering, 2014, 2(1), 168- gular Terminal Sliding-Mode Control of Robot Ma-
174. https://doi.org/10.1080/21642583.2014.890552 nipulators for High-Accuracy Tracking Control. IEEE
3. Bloch, A. M. Nonholonomic Mechanics and Con- Transactions on Industrial Electronics, 2009, 56(9),
trol. Springer-Verlag, New York, 2003. https://doi. 3593-3601. https://doi.org/10.1109/TIE.2009.2024097
org/10.1007/b97376 12. Khooban, M. H. Design an Intelligent Proportional-De-
4. Brockett, R. W. Asymptotic Stability and Feedback Sta- rivative (PD) Feedback Linearization Control for Non-
bilization. In: Brockett R. W., Millman R. S., Sussmann holonomic-Wheeled Mobile Robot. Journal of Intelli-
H. J. (Eds.), Differential Geometric Control Theory, gent and Fuzzy Systems, 2014, 26(4), 1833-1843.
Birkhauser, Boston, 181-191, 1983.
13. Kocamaz, U. E., Göksu, A., Taşkın, H., Uyaroğlu, Y. Synchro-
5. Chen, C. Y., Li, T. H. S., Yeh, Y. C. EP-Based Kinematic nization of Chaos in Nonlinear Finance System by Means
Control and Adaptive Fuzzy Sliding-Mode Dynam- of Sliding Mode and Passive Control Methods: A Compar-
ic Control for Wheeled Mobile Robots. Informa- ative Study. Information Technology and Control, 2015,
tion Sciences, 2009, 179(1-2), 180-195. https://doi. 44(2), 172-181. https://doi.org/10.5755/j01.itc.44.2.7732
org/10.1016/j.ins.2008.09.012
14. Kolmanovsky, I., McClamroch, N. H. Develop-
6. Chen, N., Song, F., Li, G., Sun, X., Ai, C. S. An Adaptive ments in Nonholonomic Control Problems. IEEE
Sliding Mode Backstepping Control for the Mobile Ma- Control Systems, 1995, 15(6), 20-36. https://doi.
nipulator with Nonholonomic Constraints. Communi-
org/10.1109/37.476384
cations in Nonlinear Science and Numerical Simula-
tion, 2013, 18(10), 2885-2899. https://doi.org/10.1016/j. 15. Lee, T. H., Lam, H. K., Leung, F. H. F., Tam, P. K. S. A Prac-
cnsns.2013.02.002 tical Fuzzy Logic Controller for the Path Tracking of
Wheeled Mobile Robots. IEEE Control Systems, 2003,
7. Chen, S. Y., Lin, F. J. Robust Nonsingular Terminal
23(2), 60-65. https://doi.org/10.1109/MCS.2003.1188772
Sliding-Mode Control for Nonlinear Magnetic Bearing
System. IEEE Transactions on Control Systems Tech- 16. Liang, Z. Y., Wang, C. L. Robust Stabilization of Non-
nology, 2011, 19(3), 636-643. https://doi.org/10.1109/ holonomic Chained form Systems with Uncertainties.
TCST.2010.2050484 Acta Automatica Sinica, 2011, 37(2), 129-142.
8. Fang, Y., Yuan, Z., Fei, J. Adaptive Fuzzy Backstepping Con- 17. Liu, Y. L., Wu, Y. Q. Output Feedback Control for Sto-
trol of MEMS Gyroscope Using Dynamic Sliding Mode chastic Nonholonomic Systems with Growth Rate Re-
Approach. Information Technology and Control, 2015, striction. Asian Journal of Control, 2011, 13(1), 177-185.
44(4), 380-386. https://doi.org/10.5755/j01.itc.44.4.9110 https://doi.org/10.1002/asjc.230
9. Ferrara, A., Giacomini, L., Vecchio, C. Control of Non- 18. Man, Z., Paplinski, A. P., Wu, H. R. A Robust MIMO
holonomic Systems with Uncertainties via Second-Or- Terminal Sliding Mode Control Scheme for Rigid
der Sliding Modes. International Journal of Robust and Robotic Manipulators. IEEE Transactions on Au-
Nonlinear Control, 2008, 18(4-5), 515-528. https://doi. tomatic Control, 1994, 39(12), 2464-2469. https://doi.
org/10.1002/rnc.1202 org/10.1109/9.362847
44 Information Technology and Control 2018/1/47
19. Man, Z., Yu, X. H. Terminal Sliding Mode Control of MIMO Efficient Heuristic for Global Optimization over Contin-
Linear Systems. IEEE Transactions on Circuits and Sys- uous Spaces. Journal of Global Optimization, 1997, 11,
tems I: Fundamental, Theory and Applications, 1997, 341-359. https://doi.org/10.1023/A:1008202821328
44(11), 1065-1070. https://doi.org/10.1109/81.641769 32. Sun, T., Pei, H., Pan, Y., Zhou, H., Zhang, C. Neural Net-
20. Marchand, N., Alamir, M. Discontinuous Exponential work-Based Sliding Mode Adaptive Control for Robot
Stabilization of Chained form Systems. Automatica, Manipulators. Neurocomputing, 2011, 74(14-15), 2377-
2003, 39(2), 343-348. https://doi.org/10.1016/S0005- 2384. https://doi.org/10.1016/j.neucom.2011.03.015
1098(02)00229-7
33. Tian, Y. P., Li, H. Exponential Stabilization of Nonho-
21. Mobayen, S. Finite-Time Tracking Control of Chained- lonomic Dynamic Systems by Smooth Time-Varying
Form Nonholonomic Systems with External Distur- Control. Automatica, 2002, 38(7), 1139-1146. https://
bances Based on Recursive Terminal Sliding Mode doi.org/10.1016/S0005-1098(01)00303-X
Method. Nonlinear Dynamics, 2015, 80, 669-683.
34. Wu, Y., Wang, B., Zong, G. D. Finite-Time Tracking
https://doi.org/10.1007/s11071-015-1897-4
Controller Design for Nonholonomic Systems with Ex-
22. Mon, Y. J., Lin, C. M. ANFIS-Based Integral Terminal tended Chained Form. IEEE Transactions on Circuits
Sliding Mode Control for Disturbed Chaotic System. and Systems II: Express Briefs, 2005, 52(11), 798–802.
Journal of Intelligent and Fuzzy Systems, 2014, 27(1), https://doi.org/10.1109/TCSII.2005.852528
443-450.
35. Wu, Y., Yu, X., Man, Z. Terminal Sliding Mode Con-
23. Mon, Y. J., Lin, C. M. Double Inverted Pendulum Decou-
trol Design for Uncertain Dynamic Systems. Systems
pling Control by Adaptive Terminal Sliding-Mode Re-
and Control Letters, 1998, 34(5), 281-287. https://doi.
current Fuzzy Neural Network. Journal of Intelligent
org/10.1016/S0167-6911(98)00036-X
and Fuzzy Systems, 2014, 26(4), 1723-1729.
36. Wu, Y. Q., Zhu, C. L., Zhang, Z. C. Finite-Time Stabiliza-
24. Ou, M., Du, H., Li, S. Finite-Time Tracking Control of
tion of a General Class of Nonholonomic Dynamic Sys-
Multiple Nonholonomic Mobile Robots. Journal of
tems via Terminal Sliding Mode. International Journal
Franklin Institute, 2012, 349(9), 2834-2860. https://
of Automation and Computing, 2016, 13(6), 585-595.
doi.org/10.1016/j.jfranklin.2012.08.009
https://doi.org/10.1007/s11633-015-0931-9
25. Pan, Y., Liu, Y., Xu, B., Yu, H. Hybrid Feedback Feedfor-
ward: An Efficient Design of Adaptive Neural Network 37. Xi, Z. R., Feng, G., Jiang, Z. P., Cheng, D. Output Feed-
Control. Neural Networks, 2016, 76(6), 122-134. https:// back Exponential Stabilization of Uncertain Chained
doi.org/10.1016/j.neunet.2015.12.009 Systems. Journal of Franklin Institute, 2007, 344(1),
36-57. https://doi.org/10.1016/j.jfranklin.2005.10.002
26. Pan, Y., Yu, H. Biomimetic Hybrid Feedback Feedfor-
ward Neural-Network Learning Control. IEEE Trans- 38. Xu, W. L., Huo, W. Variable Structure Exponential Sta-
actions on Neural Networks and Learning Systems, bilization of Chained Systems Based on the Extended
2016, 28(6), 1481-1487. https://doi.org/10.1109/TNN- Nonholonomic Integrator. Systems and Control Let-
LS.2016.2527501 ters, 2000, 41(4), 225-235. https://doi.org/10.1016/
S0167-6911(00)00057-8
27. Pan, Y., Yu, H., Er, M. J. Adaptive Neural PD Control with
Semiglobal Asymptotic Stabilization Guarantee. IEEE 39. Yang, X. S. A New Metaheuristic Bat-Inspired Algorithm.
Transactions on Neural Networks and Learning Sys- In: Gonzalez, J. R., Pelta, D. A., Cruz C., Terrazas, G., Kras-
tems, 2014, 25(12), 2264-2274. https://doi.org/10.1109/ nogor, N. (Eds.) Nature Inspired Cooperative Strategies
TNNLS.2014.2308571 for Optimization (NICSO 2010), Springer, Berlin, 65-74,
28. Passino, K. M. Biomimicry of Bacterial Foraging for 2010. https://doi.org/10.1007/978-3-642-12538-6_6
Distributed Optimization and Control. IEEE Control 40. Yang, X. S., Deb, S. Engineering Optimization by Cuckoo
Systems, 2002, 22(3), 52-67. https://doi.org/10.1109/ Search. International Journal of Mathematical Model-
MCS.2002.1004010 ing and Numerical Optimization, 2010, 1(4), 330-343.
29. Ploeg, J., Schouten, H. E., Nijmeijer, H. Position Con- https://doi.org/10.1504/IJMMNO.2010.035430
trol of a Wheeled Mobile Robot Including Tire Behav- 41. Yu, X., Man, Z. Model Reference Adaptive Control
ior. IEEE Transactions on Intelligent Transportation Systems with Terminal Sliding Modes. International
Systems, 2009, 10(3), 523-533. https://doi.org/10.1109/ Journal of Control, 1996, 64(6), 1165-1176. https://doi.
TITS.2009.2026316 org/10.1080/00207179608921680
30. Rajabioun R. Cuckoo Optimization Algorithm. Applied 42. Zhang, C., Sun, T., Pan, Y. Neural Network Observ-
Soft Computing, 2011, 11(8), 5508-5518. https://doi. er-Based Finite-Time Formation Control of Mobile
org/10.1016/j.asoc.2011.05.008 Robots. Mathematical Problems in Engineering, 2014,
31. Storn, R., Price, K. Differential Evolution – A Simple and 267307. https://doi.org/10.1155/2014/267307