You are on page 1of 19

Received: 22 October 2019 Revised: 7 May 2020 Accepted: 11 July 2020

DOI: 10.1002/asjc.2418

REGULAR PAPER

Trajectory tracking control of a wheeled mobile robot in the


presence of matched uncertainties via a composite control
approach

Hamid Reza Shafei | Mohsen Bahrami

Department of Mechanical Engineering,


Amirkabir University of Technology,
Abstract
Tehran, Iran In the present work, a novel method is devised for controlling an uncertain
wheeled mobile robot (WMR) in a desired path. To achieve this objective, an
Correspondence
Mohsen Bahrami, Department of optimal and robust control system with adaptive gains is combined to benefit
Mechanical Engineering, Amirkabir from the advantages of both methods. In fact, applying this controller not
University of Technology, Tehran, Iran.
only controls a nonlinear system robustly against uncertainties and external
Email: mbahrami@aut.ac.ir
disturbances, but also optimizes a quadratic cost function. Also, since the
upper bound of uncertainty is determined by an adaptive law, it does not
need to be adjusted manually. To ensure the designed controller's finite time
stability, Lyapunov theory is used. Finally, to illustrate the effectiveness of the
proposed control method, a case study involving a WMR is presented. By
comparing the results of this approach with those of an adaptive sliding mode
control (ASMC), it is shown that the proposed controller uses less control
effort, relative to the ASMC controller, to stabilize the uncertain WMR in the
presence of external disturbances.

1 | INTRODUCTION managed by eliminating the dependent variables from


their motion equations. Therefore, the standard control
The wheeled mobile robot (WMR) is a mechanical system laws developed for holonomic robotic systems may not
that can alleviate the work pressure in labor-intensive be applicable to nonholonomic ones [2]. Stabilization
environments such as mines, factories, hospitals, etc. The problems are concerned with obtaining a feedback law
varied applications of WMRs and the difficulties involved that can guarantee the asymptotical stability of a closed-
in their navigation control have attracted the attention loop system about an equilibrium state. If all the unstable
of researchers in recent years. Designing trajectory eigenvalues for a linear time-invariant system are
tracking controllers for nonholonomic WMRs, which are controllable, then the system's origin can be asymptotically
underactuated nonlinear systems (i.e. systems with fewer stabilized by a linear time-invariant static state feedback.
actuators than degrees of freedom), is more difficult than However, the linearization of a nonholonomic system
designing controllers for fully actuated robotic systems [1]. about any equilibrium state cannot be asymptotically
Nonholonomic systems normally refer to finite stabilized. Consequently, linear stabilization tools cannot
dimensional mechanical systems where the constraints be used even locally. Moreover, there is a fundamental
imposed on system motions are not integrable, i.e. the obstacle to the existence of smooth (or even continuous)
constraints cannot be written as the time derivatives of time-invariant feedback laws (Brockett's condition) [3]. This
some function of the generalized coordinates. Contrary to implies that, for nonholonomic control systems, there is
holonomic systems, nonholonomic systems cannot be no smooth (or even continuous) time-invariant static state
© 2020 Chinese Automatic Control Society and John Wiley & Sons Australia, Ltd

Asian J Control. 2021;23:2805–2823. wileyonlinelibrary.com/journal/asjc 2805


2806 SHAFEI AND BAHRAMI

feedback that can locally and asymptotically stabilize a In practice, obtaining the exact model of a WMR which is
specific equilibrium state of a closed-loop system [2]. affected by undesired factors such as uncertainties and
Moreover, there is no dynamic continuous time-invariant external disturbances is a challenging task, and it shows
feedback controller that can locally and asymptotically the importance of designing a composite control approach
stabilize the closed-loop system [3]. Consequently, a that can systematically approximate the upper bounds
nonholonomic control system cannot be asymptotically of uncertainties. Chen et al. [31] applied an adaptive
stabilized to an equilibrium state by means of feedback backstepping sliding mode scheme to develop a controller
linearization or any other control approach that relies on for WMRs and used the SMC to secure the system against
smooth time-invariant feedback. uncertainties. An adaptive second-order SMC was also
Because of the uncertainties and nonlinearities designed to handle the trajectory tracking of a skid-steered
associated with WMRs, it is essential to design a control WMR in the face of external disturbances and parametric
scheme that can robustly and optimally control such uncertainties [32]. They designed an adaptive algorithm to
systems [4]. From a theoretical viewpoint, Brockett's calculate the uncertainty upper bound. Tang and Cai [33]
conditions [2] and the results in [3] have established the designed an adaptive high-order SMC for a car-like mobile
fact that nonholonomic systems cannot be asymptotically robot. They defined the sliding surface such that the
stabilized about a single equilibrium state by means of reaching phase of the SMC was eliminated. However, the
smooth, or even continuous, time-invariant feedback. To missing link in all these works is the lack of improvement
overcome this limitation, research has focused on two in performance criteria (e.g., control effort minimization).
general groups of remedies: methods that employ smooth This drawback has been remedied in the present paper
and time-varying feedback, but have a slow convergence by including an optimal controller which is superior to
[5], and methods that rely on non-smooth feedback, but previous control systems.
achieve faster convergence. However, the latter solutions Optimal control is a very popular and efficient control
are more preferred in practical applications [6]. The strategy, and it has been widely used in many mechanical
WMR trajectory tracking control design is a challenging and robotic systems. The linear quadratic regulator
issue, which has been investigated by various approaches (LQR) is a popular method of minimizing a specific cost
such as nonlinear control methods [7,8], optimal control function in linear systems [34]. In contrast to linear
[9–11], adaptive control [12–16] and robust control. systems, the design of optimal control for nonlinear
[17–21]. The designed controller for a WMR should systems is a difficult task which requires the numerical
enable the robot to follow a defined path, while tackling solution of a nonlinear two-point boundary value
the existing uncertainties and disturbances. The trajectory problem (TPBVP) [35]. Shafei and Korayem [36] applied
tracking control of WMRs has received renewed attention the Pontryagin Minimum Principle on a flexible robotic
in the last two decades. For example, Jiang and Nijmeijer arm to find its maximum load carrying capacity. Despite
[22] designed a backstepping strategy for both the local having smooth control inputs, that approach cannot
and the global tracking control of an uncertain WMR. In guarantee system stability in the face of external
order to deal with system uncertainties, and to enhance the disturbances and uncertainties. In another work, with
robustness of the system, they combined the sliding mode the aim of minimizing the position and velocity errors
control (SMC) with backstepping controller. Owing to its related to a WMR's actual and desired trajectories, Nayak
robustness against uncertainties and its simple design, et al. [11] developed an optimal control scheme based
SMC has become very popular as a robust controller for on Pontryagin Minimum Principle to properly manage
uncertain systems[23–28]. The main goal of using an SMC the navigation of a WMR. One way of dealing with
is to control a system which is subjected to uncertainties optimality issues in nonlinear systems is by solving the
and disturbances. Shojaei and Shahri [5] proposed a hybrid Hamilton–Jacobi–Bellman (HJB) equation; which usually
controller for an uncertain nonholonomic WMR. They cannot be solved analytically, but numerically [37]. Ma
handled parametric and non-parametric uncertainties by et al. [37] solved the HJB equation to obtain the optimal
applying adaptive and SMC control methods, respectively. control law for a nonlinear system affected by unknown
In another work, a robust and adaptive control technique disturbances. In that paper, they circumvented the
was developed to manage the nonholonomic constraints difficulties of solving the HJB equation by employing the
of a mobile robot [29]. However, the major disadvantage inverse optimal control approach. To tackle the unknown
of that controller was that the system's dynamic disturbances of the examined nonlinear system, they also
uncertainties were bounded by known values. Peng et al. adopted a robust disturbance-observer-based controller in
[30] devised an adaptive robust tracking controller based the optimal control law. Recently, researchers have also
on Lyapunov function that could handle the unknown considered the use of state dependent Riccati equation
upper bounds of external disturbances and uncertainties. (SDRE) in the optimal control of nonlinear systems
SHAFEI AND BAHRAMI 2807

[38, 39]. Korayem et al. [38] combined the SDRE control the presence of external disturbances. In that paper, the
law with an estimator in order to reduce the state error of sliding manifold's gain was calculated by the SDRE
an uncertain flexible joint manipulator subjected to noise approach, while the upper bound of uncertainties was
and external disturbance. In general, its major weakness defined manually. In fact, the main purpose of their
is that it can only be applied to a specific class of work was to design an optimal surface that can lead to
nonlinear systems [40]. Among the existing optimal reduced convergence time. In another work, Chen [49]
control approaches, the control Lyapunov function (CLF) designed an optimal sliding mode controller to manage
has been extensively used in many robotic systems an uncertain system in the presence of external
[41]. The CLF, which solves the HJB equation, optimizes disturbances. In that work, LQR was employed to obtain
certain performance criteria. Therefore, if a CLF exists the optimal control law, while a classic sliding mode
for a nonlinear system, then a feedback control law control was integrated with the previously designed
can be found that stabilizes the nonlinear system and optimal controller in order to make the control system
optimizes its performance index without solving the HJB robust against undesired factors. As we can see, the main
equation. Nonetheless, in all these works related to optimal objective of the cited works is to design an optimal
control, the robustness of the control system against the surface that can reduce the convergence time; while the
applied uncertainties and external disturbances cannot be main achievement of our proposed control strategy is
guaranteed. To tackle this drawback of optimal control the development of a composite control system that
methods, a combination of a robust control approach and optimizes the control effort. Compared to previous works,
optimal control method has been proposed as the main the most important contributions of our proposed control
achievement of the present paper. system are the following:
In composite control systems, several methodologies
are applied to stabilize uncertain linear systems • Developing a comprehensive control system based on
[42]. While an LQR controller is designed to optimally optimal and robust control for managing uncertain
control a linear system, the SMC scheme is employed to nonlinear systems subjected to external disturbances.
make the LQR controller robust against uncertainties. • Using the CLF control system, as an efficient method
Pang and Yang [42] designed an integral sliding mode for solving optimization problems, and the nonsingular
control system to make the LQR controller robust against terminal sliding mode control (NTSMC), as a robust
uncertainties. With the aim of reducing the convergence control approach.
time, robust control has been achieved via an optimal • Reducing the chattering of control signal with a
approach [43]. The proposed method transforms a robust high-order sliding mode control scheme.
control problem into an optimal control problem, where • Eliminating the calculation of the upper bound of
the uncertainties and disturbances are reflected by a uncertainties by applying an adaptive law.
performance index. These control systems modify only • Applying this novel control scheme to an uncertain
the time response of a system, including the convergence WMR in the presence of external disturbances.
time, settling time and the overshoot [44,45]. In another
work, Xu [45] developed an SMC with time-varying This paper has been structured as follows. In
surfaces to solve an optimal control problem. In that Section 2, the composite control law based on robust
work, the original optimal control problem was NTSMC and optimal CLF control with adaptive gains is
transformed into a problem of finding the optimal sliding described. In this section, by using Lyapunov theory, the
surfaces. In another research, Mokhtari et al. [46] stability of the composite controller is proven. In Section 3,
proposed a composite control system for dealing with a WMR is examined to evaluate the performance of
uncertainties and disturbances in nonlinear systems. this novel controller. In this regard, the dynamic motion
They developed a high-order super twisting SMC, whose equations of the WMR with nonholonomic constraints
gains are updated by an adaptive law. Furthermore, the are derived by employing the Gibbs-Appell (G-A)
controller parameters of that method were determined by method. Then, our proposed control law is applied on
applying a Harmony search algorithm. With the aim of the examined WMR. In addition to verifying the stability
designing an optimal sliding surface, Ferrara et al. [47] of this control law by means of Lyapunov principle,
proposed an adaptive optimization-based SMC algorithm the results of this method are compared with those of
for uncertain nonlinear systems. In that work, by the ASMC law. Two simulations are presented in this
optimally determining the reaching law of the SMC, they section to highlight the effectiveness of the proposed
reduced the control effort of the proposed controller. approach in controlling the trajectory tracking of an
Korayem et al. [48] developed an optimal-robust control uncertain WMR. Finally, in Section 4, the work is
system for managing an uncertain nonlinear system in concluded.
2808 SHAFEI AND BAHRAMI

2 | C OM P OSITE C ONTROL  ! !  ! !  !  !  ! ! ! ! 
h x dt x , u,t = Δf x + Δh x uðt Þ + d x , t ð3Þ
SYSTEM DESIGN
! !  
!
In this section, a composite control law is designed for where Δf x , Δh x are system uncertainties, and
!! 
nonlinear systems in general. The main goal of developing
this controller is to control, both robustly and optimally, an d x ,t is the external disturbance applied to the system.
uncertain system in the presence of external disturbances It should be noted that the considered uncertainties of
and, by doing so, to guarantee the stability of the system as the system and external disturbances satisfy the matching
well as to improve its performance characteristics. This condition. As Equation 3 clearly shows, external disturbances
novel controller will be designed in two steps. Before as well as the system uncertainties have been defined in
designing the composite controller, the WMR's dynamic the input vector space. In the following section, the
motion equations are divided into the nominal section and proposed controller (Equation 1) will be designed for
the uncertain section related to parameter uncertainties this nonlinear uncertain system (Equation 2).
and external disturbances. Then, an optimal controller
based on CLF approach is designed for the nominal
section, while a high-order SMC law based on the NTSMC
scheme is designed for the uncertain part of the system 2.1 | Optimal control strategy for the
with the aim of tackling the matched uncertainties and nominal part
external disturbances. The considered composite control
!
law can be represented as To design the control input u 1 ðtÞ , we only need to
consider the nominal part of the uncertain system.
! ! !
u ðt Þ = u 1 ðt Þ + u 2 ðt Þ ð1Þ Hence, Equation 2 is written as

!
where u 1 ðt Þ is the optimal control law that minimizes _
! !!  
! !
the defined cost function and stabilizes the nominal part x nom = f x + h x u 1 ðt Þ: ð4Þ
!
of the system, and u 2 ðt Þ is the sliding mode controller
!
that ensures the robustness of the system in the face of In order to optimize the control input u 1 ðt Þ, the cost
applied uncertainties and disturbances. Figure 1 shows function J is defined as follows:
the schematic of the planned composite control strategy.
The detailed procedures for designing the proposed ð∞   
! !T !
controller have been presented below. In this regard, a J= r x + u 1 ðt ÞRu 1 ðtÞ dt ð5Þ
0
nonlinear uncertain system can be expressed as
 
_
! !!  
! ! ! ! !  !
where r x is a positive semi-definite continuously-
x = f x + h x u ðt Þ + d t x , u ,t ð2Þ
differentiable function, while R 2 Rm × m is a positive
! ! definite matrix that should be properly chosen so that the
where x ðt Þ 2 Rn and u ðt Þ 2 Rm are the state vector and h! i
the control input signal of the system, respectively. desired performance is achieved and f , r becomes zero
!!  
!
Furthermore, f x , h x are the uncertain system's state detectable. Based on the CLF method, a candidate
 !   Lyapunov function should be considered and then the
! ! ! !
nominal sectors, while h x d t x , u ,t is defined as feedback control u 1 ðt Þ should be appropriately determined

F I G U R E 1 Schematic of the
proposed controller
SHAFEI AND BAHRAMI 2809

  !
!
so that V_ x becomes negative definite [50, 51]. parameter s (the deviation from the desired response)
!
is defined as
Therefore, the stabilizing control input u 1 ðt Þ can be  ðt 
found if the proper Lyapunov function is determined. ! ! _
!
s ðt Þ = a x ðt Þ− x nom ðt Þdt ð9Þ
0
Considering a radially-unbounded
 !and positive definite
candidate Lyapunov function V x , its derivative is where a 2 Rm × n is a parameter which should be defined
 
obtained as !
so that ah x is invertible. By taking the derivative of
 !  !  !  ! ! Equation 9, Equation 10 is obtained
V_ x = Lf V x + Lh V x  u1 ðt Þ ð6Þ
 
_
! _
! !_
s ðt Þ = a x ðt Þ− x nom ðtÞ : ð10Þ
where L is the Lie  ! derivative operator. It is worth
mentioning that V x is a CLF, which can be found ana- Now, by substituting Equation 2 into Equation 10,
lytically for two dimensional systems, if we get
   !
!_ ! !
s ðt Þ = a h x u 2 + Ψ ð11Þ
! ! !  !
Lh V x = 0 ) Lf V x < 0: ð7Þ
!  ! !  ! ! 
where Ψ = h x dt x , u, t . Now, to reduce the chattering
!
and converge parameter s ðt Þ to zero in finite time
Furthermore, if a CLF exists for Equation 4, then (converge the system to the integral sliding manifold), a
!
there will certainly be a control law u 1 ðt Þ that can nonsingular terminal sliding variable is defined as [56]
stabilize the Equation 4. Based on the CLF approach, α
! ! !_β
Sontag proposed the following formula [41, 52]: I ðt Þ = s ðt Þ + λ s ðt Þ where λ > 0 ð12Þ

8 2   rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
    T   ffi3
>
> ! ! ! ! ! − ! !
>
>
2
6m x + m x + r x n x R n x 7!!
1
  !
>
>
< −6     7 n x for !
n
!
x ≠0
! 4 !T ! ! ! 5
u 1 ðt Þ = n x n x ð8Þ
>
>
>
>   !
>
> ! ! !
: 0 for n x = 0

 !  ! ! T
! ! !
where m x = Lf V x and n x = Lh V x . The where the values of α, β are chosen in such a way that
proof can be found in [53–55]. satisfy the following conditions:
Up to now, an optimal control law has been obtained
for the nominal section of the uncertain nonlinear system α
1< < 1:5 where α, β ϵ 2n + 1: ð13Þ
to minimize the control effort. Now, a high-order β
SMC law should be combined with the designed optimal
!
control law to assure the robustness of the nonlinear When I ðt Þ converges to zero in finite time, then
! _
!
system against the matching uncertainties and external both parameters of s ðt Þ, s ðtÞ will also approach zero. By
disturbances. taking the time derivative of Equation 12, we will get

 
_
! _
!
α
α !_ β − 1 !
€ _ ϕ 1 !_ ε !
! €
2.2 | Nonsingular terminal SMC strategy I = s +λ s s =κs s +s ð14Þ
β κ
An optimal control law is highly sensitive to uncer-
tainties and disturbances. Therefore, to make the designed Where ϕ = αβ − 1 , κ = λ αβ , ε = 2 − αβ and α, β satisfy
optimal controller robust against external disturbances, it the conditions of Equation 13. Komurcugil [57]
is integrated with a high-order SMC law. For this purpose, proved that
2810 SHAFEI AND BAHRAMI

8 ϕ
<! _ !_ ! of an uncertain system, the sliding mode gain should be
s > 0 for s ≠ 0
ð15Þ determined in such a way that excludes the effects of
: !_ ϕ !_ !
s = 0 for s = 0 systems uncertainties as well as external disturbances;
In view of Equation 13 and Equation 15 and for this will be possible when the upper bound of the system
!_ ! !_ ϕ uncertainty is known. However, this upper bound is
s ≠ 0 , κ s can be replaced by μ1 > 0. So, Equation 14
can be written as usually not known or too complicated to calculate, and
this issue becomes even more challenging when the
  dimensions of a system increase. Therefore, to resolve this
_
! 1 !_ ε !

I = μ1 s +s ð16Þ problem, an adaptive law that systematically determines
κ
the sliding mode gain will be used. In this respect, we will
apply the adaptive law presented by Plestan et al. [58];
By considering the constant plus proportional which guarantees a real sliding mode. The sliding mode
_
!
reaching law [57], I can be defined as gain μ(t) is found by using the following law [58]:

_
! !  ! ( !  ! !
I = −μ2 sgn I −δ1 I , ð17Þ Ξ s ðt Þ sgn s ðt Þ − ς if μ > σ
μ_ ðt Þ = ð21Þ
σ if μ ≤ σ
where μ2 > 0 and δ1 > 0. Now, by equating Equation 16
to Equation 17, €s can be determined as
!
Where Ξ > 0, ς > 0, σ > 0 have small values. It is
worth mentioning that gain μ(t) has an upper bound that
 ε  ! 
1 !_ €
! ! there exists a constant μ* > 0. Therefore, μ(t) ≤ μ*,
μ1 s + s = −μ2 sgn I −δ1 I
κ 8 t > 0. By applying Equation 21, the sliding mode gain
! ! 1 ε
!€ _
! can be determined without knowing the upper bound of
) s = −μsgn I − δI − s ð18Þ
κ system uncertainty.

where μ = μ2/μ1 > 0 and δ = δ1/μ1 > 0. By taking the first Theorem 1. The uncertain nonlinear system presented
derivative of Equation 11, Equation 19 is obtained: in Equation 2 will be asymptotically stable if the
sliding variable is defined as Equation 12 and the
     

! ! !_ ! ! _
! composite controller presented in Equation 1,
_
s ðt Þ = a h x u 2 + h x u 2 + Ψ : ð19Þ ! !
where u 1 ðtÞ and u 2 ðt Þ are respectively expressed as
Equation 8 and Equation 20, is applied.
!
Thus, the NTSMC law u 2 ðt Þ is established as
Proof 1. To prove this theorem via the Lyapunov approach,
ðt    − 1 the following Lyapunov candidate is considered:
! !
u 2 ðt Þ = − ah x ð20Þ
0
 ε  1 !T ! 1
1 !_  
! !
! ! V 1 = I  I + ðμ −μ Þ2 where ℷ > 0: ð22Þ
_
s + ah x u 2 + μsgn I + δI dτ: 2 2ℷ
κ

By taking time derivative of Equation 22 and


In general, this robust control law should be solved employing Equation 14, Equation 23 is obtained.
numerically. Note that the NTSMC gain should be

!_  
chosen so that aΨ < μ (the reason for this has been !T !_ 1 !T _
! !_ ϕ!

V_ 1 = I  I + ðμ−μ Þμ_ ðt Þ = I  s + κ s s ð23Þ

expounded in the proof of Theorem 1).  
1 ! ! !
+ ðμ−μ ÞΞ s ðt Þ sgn s ðt Þ − ς

       
!T _
! _ϕ
! ! ! _ ! ! !_
2.3 | Adaptive gain = I  s +κs a h x u 2 + h_ x u 2 + Ψ
 
1 ! ! !
Up to this point, an optimal and robust control law has + ðμ −μ ÞΞ s ðt Þ sgn s ðt Þ − ς

been applied to stabilize an uncertain nonlinear system
exposed to external disturbances. To secure the stability The time derivative of Equation 20 yields
SHAFEI AND BAHRAMI 2811

   −1 1 ε   ! 
_ _ ! ! !
! !
u 2 = − ah x
! ! !
s + ah_ x u 2 + μsgn I + δI ð24Þ s ðt Þ > ς is guaranteed by considering any initial condition
κ
! !
s ð0 Þ > ς .
As μ* is the upper bound of μ, and by substituting
! ! !
Equation 24 into Equation 23, we achieve Case 2. Suppose sðt Þ < ς (one recalls that ς is a small
   parameter). From Equation 28, we can conclude
! ! _
!T _
! _ϕ
! 1 !_ ε !
that Γ < 0 and that a time exists at which V_ 1 would
V_ 1 = I  s +κs − s −μsgn I −δI + aΨ ð25Þ
κ
be sign indefinite, and it would be impossible to
 
1 ! ! ! verify the closed-loop stability of the system. In this
+ ðμ −μ ÞΞ s ðt Þ sgn s ðt Þ − ς
ℷ domain, the gain μ(t) decreases according to the

! adaptation law presented in Equation 21. According
As was mentioned above, κ s can be replaced by to this law, with the reduction of μ(t) and by
μ1 > 0. Therefore, Equation 25 can be rewritten as entering into μ ≤ σ domain, the second law in
Equation 21 becomes valid for finite time. As soon
  ! !  as μ becomes less or equal to σ, its value starts to
!T _
!
V_ 1 = I  μ1 −μsgn I − δI + aΨ ð26Þ increase immediately, so that μ(t) = σ+σ.t (inside

  !
1 ! ! ! μ ≤ σ domain[58]). Therefore, s ðt Þ may become
+ ðμ− μ ÞΞ s ðt Þ sgn s ðt Þ − ς
ℷ ! ! !
 ! !T ! !T !_  larger than ς (comes out of s ðt Þ < ς domain). As
!T
= μ1 −μI  sgn I −δI  I + I  aΨ
! !
soon as s ðt Þ becomes greater than ς (but is kept
 
1 ! ! ! !T !
+ ðμ −μ ÞΞ s ðt Þ sgn s ðt Þ − ς bounded[58]), V_ 1 ≤ −μ1 δI  I and, consequently,

 ! !T ! ! !_  V1 starts to decrease.
!T
≤ μ1 − μI  sgn I − δI  I + I aΨ
  As is observed in Figure 2, the sliding mode variable
1 ! ! !
− jμ− μ jΞ s ðt Þ sgn s ðt Þ − ς : !
ℷ I will approach zero in finite time (black line in
Figure 2).
Thus, the black line can be reached from anywhere
!_ ! !
Note that as aΨ ≤ μ so we get on the phase plane of the system (I ðt Þ = 0 ). Now, to
show that point A will be finally reached, the following
  two cases are discussed (red and blue lines in Figure 2).
!T ! 1 ! ! !
V_ 1 ≤ −μ1 δI  I − jμ −μ jΞ s ðt Þ sgn s ðt Þ − ς : ð27Þ

|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Γ

Concerning the stability of the proposed controller,


two different cases may occur, that should be considered
separately.
! !
Case 1. Suppose sðtÞ > ς. Then it can be found that
Γ > 0, and Equation 27 can be written that

_ !T !
V ≤ −μ1 δI  I
1
  ð28Þ
1 ! ! ! !T !
− jμ −μ jΞ s ðt Þ sgn s ðt Þ − ς ≤ − μ1 δI  I

|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Γ

Thus, based on the Lyapunov stability theory, the


stability of the proposed control system in the domain FIGURE 2 The phase plot of the system
2812 SHAFEI AND BAHRAMI

! ! !
Case 1. Let us assume that I = 0 and s > 0 (Red line). TABLE 1 Parameters of the considered WMR
!_ !
Based on Figure 2, s < 0 ; so, s will decrease (the
Parameter Description
flow is to the left side) and finally reach point A,
! !_ l Distance between points O and G
where s and s become zero.
d Distance between the center of driving wheels
! ! ! and point O
Case 2. Let us assume that I = 0 and s < 0 (blue line).
_
! ! r Driving wheel radius
As is clearly shown in Figure 2, s > 0 ; so, s
will increase (the flow is to the right side) and θr , θl Angular positions of the right and left wheels
! _
!
finally reach point A, where both s and s become MTot Total mass of WMR
zero [59]. ■ ITot Total moment of inertia of WMR about the y0
axis
Thus, an optimal-robust controller with adaptive gain Iw Each driving wheel's moment of inertia about its
has been designed. To show the performance of this novel rotation axis
controller, it is applied on a WMR in the next section.

3 | CASE STUDY: WHEELED The parameters of the examined WMR are described
MOB ILE ROB OT in Table 1.
The dynamic modeling of the examined WMR by G-A
In this section, the designed optimal adaptive sliding formulation is presented in the next section.
mode control (OASMC) law is applied on a WMR. Due to
the existence of nonholonomic constraints, Lagrange
multipliers have to be calculated in order to obtain the 3.1 | Dynamic modeling of WMR
motion equations of the considered WMR via the
Lagrangian approach; which is very time-consuming and With regards to the WMR model (Figure 3 and Table 1),
cumbersome [60-63]. Therefore, to avoid the calculation the absolute velocity of Point O, in the x0y0z0 frame
0! !
of Lagrange multipliers, the WMR equations of motion attached to Point O, is expressed as v 0 = vO i . Also, the
are derived by G-A formulation [64-66]. The schematic of angular velocity of the platform can be written as
the considered WMR is depicted in Figure 3. !
0
_
!
In modeling the WMR, the following assumptions ψ = ψ_ k . Now, the absolute velocity of the center of mass
have been considered: is represented as
Assumptions:
0! ! !
vG = vO i + lψ_ j ð29Þ
1. Each wheel is modeled as a solid disk with a single
point contact with the ground. !
! !
2. The path of the WMR is obstacle-free. where i = f1 0 0gT , j = f 0 1 0 gT , and k = f 0 0 1 gT .
3. The wheels of WMR do not slip in the lateral To derive the motion equations by the G-A approach, the
direction. center of mass's absolute acceleration should be determined.
4. WMR wheels roll only in the forward direction. By taking time derivative of Equation 29, we get

F I G U R E 3 The schematic
of the considered wheeled
mobile robot
SHAFEI AND BAHRAMI 2813

o
_
!  ! ! To complete the governing equations of motion, the
v G = v_ O −lψ_ 2 i + ðvO ψ_ + l€
ψÞ j ð30Þ
external torques applied to each driving wheel (τR and τL)
should be considered. Therefore, the generalized work is
Based on the third assumption, the velocity of driving determined as
wheels can be represented as
1 1
0! !
U = ðvO + dψ_ ÞτR + ðvO −dψ_ ÞτL : ð38Þ
v R=L = r θ_ R=L i ð31Þ r r

Now, Equations 39 and 40 are obtained by differentiating


The velocities of driving wheels can also be _
Equation 38 with respect to vO and ψ:
developed as
∂U 1
! = ðτ R + τ L Þ ð39Þ
0!
v R=L = ðvO  dψ_ Þ i ð32Þ ∂vO r

∂U d
= ðτR −τL Þ ð40Þ
In view of Equation 31 and Equation 32, the angular ∂ ψ_ r
velocity of each driving wheel is expressed as
Thus, by using Equations 36 and 39, the translational
1 motion equations, and by using Equations 37 and 40, the
θ_ R=L = ðvO  dψ_ Þ ð33Þ
r rotational motion equations of WMR can be achieved.

As was mentioned above, the angular acceleration of • Translational motion of WMR


each driving wheel is required to extract the motion
equations of WMR via the G-A method. So, by taking ∂S ∂U
derivative of Equation 33 with respect to time, the = ð41Þ
∂ v_ O ∂vO
angular accelerations of the wheels will be obtained as

• Rotational motion of WMR


€θR=L = 1 ðv_ O  d€
ψ Þ: ð34Þ
r
∂S ∂U
= ð42Þ
∂ ψ€ ∂ ψ_
To derive the WMR equations of motion by the G-A
formulation, we need the Gibbs function in terms of
accelerations, as follows: The inverse dynamics of the equations can be
expressed as
h 2 i  ! !  
1
S = M Tot v_ O −lψ_ 2 + ðvO ψ_ + l€ ψ Þ2 +
1
I Tot ψ€ 2 ð35Þ € ! ! !_
2 2|fflfflfflffl{zfflfflfflffl} I Θ Θ = T Θ, Θ ð43Þ
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
Tra:motion of the WMR Rot:Motion of the WMR

1 €2 €2 
+ I w θ R + θL where each term of Equation 43 is represented as
2
|fflfflfflfflfflfflfflfflffl
ffl{zfflfflfflfflfflfflfflfflfflffl}
Rot:motion of the driving wheels
2 3
2I w
! 6 M Tot + 2 0 7
r
Now, by taking derivative of the Gibbs function with I Θ =6
4
7
25 ð44Þ
d
respect to quasi-accelerations (_vO and ψ€ ), we will have 0 2
M Tot l + I Tot + 2I w 2
r
 
∂S 2I w €
= −M Tot lψ_ 2 + M Tot + 2 v_ O ð36Þ !
Θ = f v_ O ψ€ gT ð45Þ
∂ v_ O r
 
! ! !_ T
 2  T Θ, Θ = 1r ðτR + τL Þ + MTot lψ_ 2 dr ðτR −τL Þ−MTot lvO ψ_
∂S d
= M Tot lðvO ψ_ + l€
ψ Þ + 2 2 I w + I Tot ψ€ : ð37Þ
∂ ψ€ r ð46Þ
2814 SHAFEI AND BAHRAMI

F I G U R E 4 Actual and reference


trajectories of the examined WMR

! ! ! ! !
The time responses of quasi-velocities will be v e = v Or − v O = ðvOr cosðψ e Þ−vO Þ i + ðvOr sinðψ e Þ −0Þ j ð51Þ
obtained by solving Equation 43. Since the primary
purpose of this paper is the trajectory tracking control Now, by using Equation 49 and Equation 51, we get
of the WMR, the tracking error, which is defined as
the difference between the actual and the desired x_ e = vOr cosðψ e Þ −vO + ye ψ_ ð52Þ
trajectories, should be determined. As shown in
Figure 4, the position and the orientation errors of
y_ e = vOr sinðψ e Þ −x e ψ:
_ ð53Þ
point O can be expressed as

0! ! ! Thus, Equation 50 along with Equation 52 and


Re = x e i + ye j ð47Þ
Equation 53 represent the tracking error dynamics of the
considered WMR. In the next section, these equations
ψ e = ψ r −ψ: ð48Þ
will be employed to design the kinematic and dynamic
G! G!
controllers.
Based on Figure 4, R o r and R o a are the absolute
positions of point O of the examined WMR with regard to
reference and actual trajectories, respectively. As can be 3.2 | Applying the composite control law
0! G! G! on WMR
seen, Re = Ror − Roa
denotes the error between the ref-
erence and the actual trajectories of the WMR. So, by tak-
0!
The proposed controller will be applied on the considered
! !
ing the time derivative of R e = x e i + ye j , we get WMR to direct it onto its desired path. The main purpose
of the control system is to determine the proper values of
  
_ ! _ _
0
! ! ! ! ! right and left wheel torques in such a way that the WMR
R e = ve = x_ e i + y_ e j + x e i + ye j ð49Þ
can track its predefined trajectory. As it is demonstrated
|fflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl}
Due to value changes Due to frame rotation
in Figure 5, the trajectory tracking control of the WMR
! ! ! can be achieved in two steps.
ve = ðx_ e −ye ψ_ Þ i + ðy_ e + x e ψ_ Þ j First, a kinematic control system is designed based on
the ASMC scheme. In this step, the input control
and the time differentiation of Equation 48 can be velocities are obtained so that the robot can track its
achieved as follows predefined path. Second, by developing a dynamic
controller based on the OASMC approach, the wheel
ψ_ e = ψ_ r − ψ:
_ ð50Þ torques are determined so that the errors between WMR
velocities and the velocity inputs achieved from the
!
Furthermore, in view of Figure 4, v e is written as previous stage converge to zero.
SHAFEI AND BAHRAMI 2815

F I G U R E 5 The overall closed-loop


control system of the WMR

3.2.1 | Kinematic control law the state of the system, q2, also converges to zero [67].
By taking the derivative of Equation 57 with respect to
With the proposed kinematic controller, the velocities of time and substituting Equations (54–56), Equation 58
WMR (vOd , ψ_ d ) that enable the robot to track its desired will be obtained.
time-varying trajectory will be obtained. To design the
proposed control law, the state space form of kinematic s_ 1 = vOr cosðq3 Þ −u1 + q2 u2 ð58Þ
equations are rewritten as follows:
vOr ðvOr sinðq3 Þ−q1 u2 Þ + q2 v_ Or
s_ 2 = ψ_ r − u2 +
1 + ðq2 vOr Þ2
q_ 1 = vOr cosq3 −u1 + q2 u2 ð54Þ

q_ 2 = vOr sinq3 − q1 u2 ð55Þ Now, by defining the following control inputs with
constant reaching law [68], the asymptotic stability of the
kinematic system of WMR will be achieved:
q_ 3 = ψ_ r − u2 : ð56Þ

!
u1 = q2 u2 + vOr cosðq3 Þ + K 1 sgnðs1 Þ ð59Þ
In the above equations, q = f q1 q2 q3 gT = f x e ye ψ e gT
!
is the state error vector and u = f u1 u2 gT = f vOd ψ_ d gT
is the desired velocity vector (control inputs). Since the    
number of state variables is more than the number of ψ_ r + q2 =1 + ðq2 vOr Þ2 v_ Or + v2Or =1 + ðq2 vOr Þ2 sinðq3 Þ
u2 =   ð60Þ
control inputs, it would be challenging to design a control 1 + vOr =1 + ðq2 vOr Þ2 q1
! !
law that stabilizes the considered system (i.e., q ! 0
+ K 2 sgnðs2 Þ:
as t ! ∞). After defining the tracking error, a control
law that makes this error converge to zero should be
determined. To design an adaptive SMC law, the follow- The stability proof of these control laws is presented
ing sliding surface is considered in Theorem 2.

Theorem 2. The nonlinear system presented in


s1 = q1
ð57Þ Equations (54–56) will be asymptotically stable if
s2 = q3 + arctanðq2 vOr Þ : the control laws expressed in Equations 59 and 60
are applied.

The sliding mode control law will make s1,2 ! 0 to Proof 2. To prove the stability of the aforementioned
obtain q1 = 0 and q3 = −arctanðq2 vOr Þ. Notice that when system, a positive definite Lyapunov function is
q1 converges to 0 and q3 converges to −arctanðvOr q2 Þ , considered as follows:
2816 SHAFEI AND BAHRAMI

 
1 1 !
V = s21 + s22 : ð61Þ x_ 2 = f 2 x + b2 U 2 : ð65Þ
2 2
   
! !
By taking the derivative of Equation 61 with respect Functions f 1 x and f 2 x and parameters b1 and
to time, V_ will be achieved as follows: b2 can be written as

   
V_ = s1 s_ 1 + s2 s_ 2 = s1 ðvOr cosðq3 Þ −u1 + q2 u2 Þ ð62Þ !
f1 x =
M Tot lx 22
,
!
f2 x = −
M Tot lx 1 x 2
ð66Þ
M Tot + 2Ir 2w
2
! M Tot l2 + I Tot + 2I w dr2
vOr ðvOr sinðq3 Þ −q1 u2 Þ + q2 v_ Or
+ s2 ψ_ r −u2 + :
1 + ðq2 vOr Þ2 1 d
b1 =  , b2 =   ð67Þ
r M Tot + 2Ir2w 2 2
r M Tot l + I Tot + 2I w dr 2
Now, by substituting Equations 59 and 60 into Equa-
tion 62, we get As mentioned above, the optimal control law (presented

V_ = s1 ðvOr cosðq3 Þ − ðq2 u2 + vOr cosðq3 Þ + K 1 sgnðs1 ÞÞ + q2 u2 Þ ð63Þ


0 0    1
  ψ_ r + q2 v_ Or =1 + ðq2 vOr Þ2 + v2Or sinðq3 Þ=1 + ðq2 vOr Þ2
+ s2 @ψ_ r − 1 + q1 vOr =1 + ðq2 vOr Þ2 @   A
1 + q1 vOr =1 + ðq2 vOr Þ2
   
+ K 2 sgnðs2 Þ + v2Or sinðq3 Þ=1 + ðq2 vOr Þ2 + q2 v_ Or =1 + ðq2 vOr Þ2 = −s1 ðK 1 sgnðs1 ÞÞ −s2 ðK 2 sgnðs2 ÞÞ

= −K 1 js1 j−K 2 js2 j < 0:

As can be seen from Equation 63, V_ is negative as Equation 8) optimizes the nominal part of the WMR
definite; which confirms the asymptotic stability of the system.
proposed system. ■ For designing a robust controller for the uncertain
!
section of the WMR model, parameter s and the sliding
Remark 1. To reduce the amount of chattering in the variables are defined as Equations 9 and 12, respectively.
control signals, function sgn(.) will be replaced with With respect to Equation 12, the NTSMC law is
function sat(.). derived as Equation 20. Now, the optimal-robust control
law (the combination of Equation 8 and Equation (20))
Remark 2. As mentioned before, the SMC gain can be is applied to the uncertain WMR exposed to external
achieved by an adaptive law. Here, the SMC gains K1 disturbances. It should be mentioned that the gain of
and K2 in Equations 59 and 60 have been determined NTSMC is obtained by the adaptive law presented in
by the adaptive law presented in Equation 21. Equation 21.

3.2.2 | Dynamic control law TABLE 2 Values of WMR parameters [69]

Parameter Value
Here, the proposed control law, which was thoroughly
l 0.065 m
explained in Section 2, will be applied to an uncertain
WMR. For this purpose, the dynamic model of the WMR d 0.145 m
(Equation 43) can be expressed in the form of Equation 2, r 0.08 m
as follows: MTot 6.64 kg
ITot 0.06363 kg m2
 
!
x_ 1 = f 1 x + b1 U 1 ð64Þ Iw 0.0008 kg m2
SHAFEI AND BAHRAMI 2817

To illustrate the performance of this proposed compare the abilities of OASMC and ASMC controllers
controller, the trajectory tracking control of an uncertain in handling the trajectory tracking of an uncertain
WMR will be simulated in the next section. WMR. Several undesired conditions such as external
disturbances, parameter uncertainty, and actuator signal
saturation have been considered in the simulated model.
3.3 | Simulation results For this purpose, 25% uncertainty in system's mass prop-
erties, including MTot, ITot and Iw has been applied after
In the following, based on our proposed control an elapse of 5 sec. The nominal WMR parameters are
approach, two different trajectories have been defined to presented in Table 2.

F I G U R E 6 (a) Trajectories of
the uncertain WMR achieved by
applying the OASMC and ASMC
algorithms, (b) posture tracking
errors for the OASMC and ASMC
algorithms

F I G U R E 7 (A) Tracking errors


related to linear velocity of WMR,
(B) tracking errors related to angular
velocity of WMR
2818 SHAFEI AND BAHRAMI

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Furthermore, to make the simulation results more
vOr ðtÞ = x_ Or 2 ðt Þ + y_ Or 2 ðtÞ ð71Þ
realistic, external disturbances have been added to the
WMR model after an elapse of 8 s. In order to take
Coulomb and viscous frictions into consideration, the €yOr ðt Þx_ Or ðt Þ−€x Or ðt Þy_ Or ðt Þ
following disturbance has been applied to the simulated ψ_ r ðt Þ =
x_ Or 2 ðt Þ + y_ Or 2 ðtÞ
WMR [70]

d1 ðt Þ = F C signðvO Þ + BvO : ð68Þ From Equation 69, the initial postures  (position
 and
orientation) of the reference robot are x Or , yOr = ð0, 0Þ

and ψ Or = 63:43 . The values of these variables
 for the
In the above equation, FC stands for the Coulomb actual robot are considered as x OA ,yOA = ð −0:3, −0:1Þ

friction, B denotes the viscous friction coefficient, and vO and ψ OA = 30 . Also, from Equation 71, the initial linear
is the linear velocity of the mobile platform. The values and angular velocities of the reference robot are
of the parameters are chosen as FC = 1.5 N and B = 0.8. vOr ð0Þ = 1:34 m=s and ψ_ Or ð0Þ = 0 rad=s . For the actual
To consider a real-world WMR, the control signals are robot, these parameters are set to vOA ð0Þ = 1m=s and
limited to |τR,L| ≤ 15 N. m. The following trajectory is ψ_ OA ð0Þ = 0 rad=s . To evaluate the great performance of this
considered for the first simulation: novel controller, its results are compared with those of an
ASMC. The reference trajectory of the WMR in the x-y
plane and its actual trajectories (obtained by the ASMC
x Or ðtÞ = sinð0:6t Þ and OASMC schemes) are illustrated in Figure 6a. In
ð69Þ
yOr ðtÞ = sinð1:2t Þ addition, the desired and actual position and orientation
of the WMR are displayed in Figure 6b. As can be seen,
despite the existence of parametric uncertainties, external
In optimal controller design, by choosing high values disturbances, and actuator signal saturation, the kinematic
for matrix R (or low values for matrix Q), control effort is control laws proposed through Equation 59 and Equation 60
reduced and trajectory tracking error is increased; and have been able to accomplish the set trajectory tracking
vice versa. In the simulation, the penalty matrix are objectives with acceptable precision.
considered to be Q = diag(1) and R = diag(10). Design The reference and the actual linear and angular
parameters of OASMC are also elected as follows: velocities of the WMR are depicted in Figure 7. As it was
previously mentioned, the desired values of these
velocities are determined by applying the control laws
a = 9, β = 7, λ = 0:5, δ = 2
presented in Equation 59 and Equation 60, respectively.
ð70Þ
According to the results, the proposed controller
Ξ = 0:5, σ = 0:8, ς = 0:08:
converges the trajectory tracking errors of the uncertain
system to zero faster than the ASMC control law;
The reference velocities of the WMR can be obtained which confirms the superiority of the proposed controller
as follows: over the ASMC.

F I G U R E 8 (a) Right wheel control


torques, (b) left wheel control torques
SHAFEI AND BAHRAMI 2819

F I G U R E 9 (a) Trajectories of the


uncertain WMR achieved by applying
the OASMC and ASMC algorithms,
(b) posture tracking errors for OASMC
and ASMC algorithms


Figure 8 shows the torques exerted on each wheel and ψ Or = 90 . The values ofthese variables
 for the actual
of the WMR. The obtained control laws are smooth robot are considered as x OA , yOA = ð0:2, −0:1Þ and

and can be appropriately applied to a real-world WMR. ψ OA = 45 . Also, from Equation 71, the initial linear
As it can be seen, the OASMC controller can manage and angular velocities of the reference robot are
the trajectory tracking control of the uncertain WMR vOr ð0Þ = 0:07 m=s and ψ_ Or ð0Þ = 0:033 rad=s . For the actual
with less effort than the ASMC controller; which robot, these parameters are set to vOA ð0Þ = 0:1 m=s and
shows the superiority of the former control scheme over ψ_ OA ð0Þ = 0 rad=s.
the latter. The reference trajectory for the WMR in the x-y plane
In the second simulation, the following reference and its real trajectories are shown in Figure 9a. Also,
trajectory is considered: the desired and the actual position and orientation of
the WMR are depicted in Figure 9b. As this figure
    
36t 6t shows, similar to the previous simulation, the proposed
x Or ðt Þ = 0:02 20 + cos cos ð72Þ
35 35 controller accomplishes the trajectory tracking objective
     with excellent precision.
36t 6t The desired and actual linear/angular velocities of the
yOr ðt Þ = 0:02 20 + cos sin :
35 35 WMR are also illustrated in Figure 10. Note that the
desired velocities are determined by Equation 71. As can be
Again, all the aforementioned undesired conditions, clearly seen, the proposed controller accurately converges
including parameter uncertainties, external disturbances the velocities of the system to their desired values.
and actuator saturation are considered in the WMR The torques applied to the left and right wheels are
model. Similar to the previous computer simulations, shown in Figure 11. As it is observed, the OASMC
parameter uncertainties and external disturbances are controller handles the trajectory tracking control of
applied after time elapses of 5 and 8 s, respectively. the uncertain WMR with less effort than the ASMC
From Equation 72, the initial postures
 (position
 and controller. Compared to ASMC signal, the control inputs
orientation) of the reference robot are x Or ,yOr = ð0:42, 0Þ of the proposed controller are smoother and seem to be
2820 SHAFEI AND BAHRAMI

F I G U R E 1 0 (a) Tracking errors


related to linear velocity of WMR,
(b) tracking errors related to angular
velocity of WMR

F I G U R E 1 1 (a) Right wheel


control torques, (b) left wheel control
torques

TABLE 3 Comparing the performances of the proposed controller and ASMC controllers
Simulation I Simulation II
Proposed controller ASMC Proposed controller ASMC
Ðtf 0.0019 0.0220 0.0026 0.0025
q2e1 dt ðm2 secÞ
0
Ðtf 0.131 0.014 0.1620 0.1885
q2e2 dt ðm2 secÞ
0
Ðtf   0.0254 0.0272 0.0186 0.0242
q2e3 dt rad2 sec
0
Ðtf   48.975 798.215 25.201 361.8294
τ2R dt ðN:mÞ2 sec
0
Ðtf   50.498 768.637 28.975 339.6624
τ2L dt ðN:mÞ2 sec
0

more suitable for practical purposes; which shows the than that of the ASMC system. Also, in order to better
superiority of this method over ASMC. demonstrate the higher efficiency of the proposed
As can be seen from simulation results, the proposed method relative to the ASMC scheme, the time integrals
control system is able to manage an uncertain WMR of the obtained results are presented in Table 3. This
subjected to parameter uncertainties and external table clearly shows that the OASMC control law needs
disturbances robustly and with less control effort. The less control effort and enjoys smaller tracking errors
control signal of our proposed control system is smaller compared to the ASMC controller.
SHAFEI AND BAHRAMI 2821

4 | C ON C L U S I ON 2. A. M. Bloch, M. Reyhanoglu, and N. H. McClamroch, Control


and stabilization of nonholonomic dynamic systems, IEEE Trans
The main novelty of this paper is the development of a Auto Cont, 37 (11) (1992), 1746–1757.
3. R. W. Brockett, Asymptotic stability and feedback stabilization,
composite control method that can successfully control
Diff Geom Cont Theory, 27 (1) (1983), 181–191.
the trajectory tracking of nonlinear uncertain WMR 4. F. Alonge et al., A hybrid observer for localization of mobile
systems subjected to externals disturbances. To achieve vehicles with asynchronous measurements, Asian J Cont, 21 (4)
this objective, an optimal SMC controller, whose gain is (2019), 1506–1521.
determined by an adaptive law, has been presented. The 5. K. Shojaei and A. Shahri, Adaptive robust time-varying control
CLF method has been employed to formulate an optimal of uncertain non-holonomic robotic systems, IET ContTheory
control law for the nominal section of the proposed system, App, 6 (1) (2012), 90–102.
6. C. C. Tsai, M. B. Cheng, and S. C. Lin, Robust tracking control
while a NTSMC law has been developed to stabilize the
for a wheeled mobile manipulator with dual arms using hybrid
uncertain part of the system. The stability of these control
sliding-mode neural network, Asian J Cont, 9 (4) (2007), 377–389.
laws has been verified by the Lyapunov approach. Then, to 7. J. Keighobadi and M. B. Menhaj, From nonlinear to fuzzy
show the effectiveness and the superiority of the proposed approaches in trajectory tracking control of wheeled mobile
strategy, it has been used to control the trajectory tracking robots, Asian J Cont, 14 (4) (2012), 960–973.
of an uncertain WMR. 8. K. L. Fetzer, S. Nersesov, and H. Ashrafiuon, Full-state
The Gibbs-Appell approach, which, in contrast to the nonlinear trajectory tracking control of underactuated surface
Lagrangian and the Newton-Euler methods, is an efficient vessels, J Vib Control, (2020), 1286–1296. https://doi.org/10.7754/
6319895658
technique for modeling systems with nonholonomic
9. K. J. Kali nski and M. Mazur, Optimal control of 2-wheeled
constraints, has been employed to model the proposed sys- mobile robot at energy performance index, Mecha Syst Signal
tem. Then, two control systems have been designed to Proc, 70 (2016), 373–386.
kinematically and dynamically control the considered 10. R. Havangi, Mobile robot localization based on PSO estimator,
uncertain WMR. The results clearly show that the proposed Asian J Cont, 21 (4) (2019), 2167–2178.
control system successfully controls the WMR over its 11. A. Nayak et al., Optimal Trajectory Tracking of Nonholonomic
desired trajectories. Mechanical Systems: a geometric approach, In 2019 American
The main emphasis of this paper is on theoretical Control Conference (ACC), 2019 IEEE.
frameworks. As a future work, the approach developed 12. N. T. Binh et al., An adaptive backstepping trajectory tracking
control of a tractor trailer wheeled mobile robot, Int J Control
in this work can be modified via a disturbance
Autom Syst, 17 (2) (2019), 465–473.
observer to manage uncertain nonlinear systems in the
13. H. Sun et al., Controlling tractor-semitrailer vehicles in
presence of mismatched uncertainties. Moreover, by automated highway systems: Adaptive robust and Lyapunov
manufacturing and testing the mentioned robotic minimax approach, Asian J Cont. 1–15. https://doi.org/
system in the presence of wheel slippage, the 10.1002/asjc.2372
computational results could be compared with 14. P. Shu, M. Oya, and J. Zhao, A new adaptive tracking control
experimental data. Also, for confirming the reliability scheme of wheeled mobile robot without longitudinal velocity
of the proposed controller, its stochastic robustness can measurement, Int J Robust Nonlinear Cont, 28 (5) (2018),
1789–1807.
be analyzed by applying the Monte Carlo technique as
15. Z. Jin et al., Adaptive backstepping tracking control of a car with
a robustness metrics.
n trailers based on RBF neural network, Asian J Cont, (2019),
1–11. https://doi.org/10.1002/asjc.2255
AUTHORS' CONTRIBUTIONS 16. Z. Miao and Y. Wang, Adaptive control for simultaneous
Hamid Reza Shafei: Conceptualization; formal analysis; stabilization and tracking of unicycle mobile robots, Asian J
investigation; methodology; software; validation; visuali- Cont, 17 (6) (2015), 2277–2288.
zation. Mohsen Bahrami: Conceptualization; formal 17. M. Thomas, B. Bandyopadhyay, and L. Vachhani, Finite-time
analysis; project administration; resources; supervision. posture stabilization of the unicycle mobile robot using only
position information: A discrete-time sliding mode approach, Int
J Robust Nonlinear Cont, 29 (6) (2019), 1990–2006.
ORCID
18. Y. Cheng et al., Robust finite-time consensus formation control
Mohsen Bahrami https://orcid.org/0000-0001-5022- for multiple nonholonomic wheeled mobile robots via output
2929 feedback, Int J Robust Nonlinear Cont, 28 (6) (2018), 2082–2096.
19. J.-C. Ryu and S. K. Agrawal, Differential flatness-based robust
R EF E RE N C E S control of mobile robots in the presence of slip, Int J Rob Res,
1. C. Ben Jabeur and H. Seddik, Design of a PID optimized neural 30 (4) (2011), 463–475.
networks and PD fuzzy logic controllers for a two-wheeled mobile 20. R. S. Inoue et al., Robust recursive linear quadratic regulator for
robot, Asian J Cont, (2020), 1–19. https://doi.org/10.1002/asjc. wheeled mobile robots based on optical motion capture cameras,
2356 Asian J Cont, 21 (4) (2019), 1605–1618.
2822 SHAFEI AND BAHRAMI

21. H. Sun et al., Controlling the differential mobile robot with 39. S. Rafee Nekoo, A PDE breach to the SDRE, Asian J Cont,
system uncertainty: Constraint-following and the adaptive robust 22 (2) (2020), 667–676.
method, J Vib Control, 25 (6) (2019), 1294–1305. 40. S. R. Nekoo, Nonlinear closed loop optimal control: A modified
22. Z.-P. JIANGdagger and H. Nijmeijer, Tracking control of mobile state-dependent Riccati equation, ISA Trans, 52 (2) (2013),
robots: A case study in backstepping, Automatica, 33 (7) (1997), 285–290.
1393–1399. 41. J. Zhang, Z. Han, and J. Huang, Homogeneous feedback control
23. W. Qin, W.-B. Shangguan, and K. Zhao, A research of sliding of nonlinear systems based on control L yapunov functions,
mode control method with disturbance observer combining Asian J Cont, 16 (4) (2014), 1082–1090.
skyhook model for active suspension systems, J Vib Control, 26 42. Pang, H.-P. and Q. Yang. Optimal sliding mode output tracking
(11–12) (2020), 952–964. control for linear systems with uncertainties. In 2010 International
24. A. R. Yasin et al., Fixed frequency sliding mode control of Conference on Machine Learning and Cybernetics. 2010. IEEE.
renewable energy resources in DC micro grid, Asian J Cont, https://doi.org/10.1109/ICMLC.2010.5580607
21 (4) (2019), 2074–2086. 43. M. Khoshhal Rudposhti, M. A. Nekoui, and M. Teshnehlab,
25. Y. Feng et al., Full-order sliding-mode control of rigid robotic Robust optimal control for a class of nonlinear systems with
manipulators, Asian J Cont, 21 (3) (2019), 1228–1236. uncertainties and external disturbances based on SDRE, Cogent
26. M. Boukattaya, N. Mezghani, and T. Damak, Adaptive Eng, 5 (1) (2018), 1451014.1–16.
nonsingular fast terminal sliding-mode control for the tracking 44. F. Dinuzzo and A. Ferrara, Higher order sliding mode controllers
problem of uncertain dynamical systems, ISA Trans, 77 with optimal reaching, IEEE Trans Auto Cont, 54 (9) (2009),
(2018), 1–19. 2126–2136.
27. M. Hou and Y. Wang, Data-driven adaptive terminal sliding 45. R. Xu, Optimal sliding mode control and stabilization of
mode control with prescribed performance, Asian J Cont, (2019), underactuated systems, Doctoral dissertation, The Ohio State
1–12. https://doi.org/10.1002/asjc.2245 University, 2007.
28. Chen, S.-H., C.-T. Lin, and L.-C. Fu. Second order sliding mode 46. M. Mokhtari, M. Taghizadeh, and M. Mazare, Optimal adaptive
control on task-space of a 6-DOF Stewart platform. in IECON high-order super twisting sliding mode control of a lower limb
2012-38th Annual Conference on IEEE Industrial Electronics exoskeleton robot, Modares Mech Eng, 19 (3) (2019), 777–787.
Society. 2012. IEEE. 47. A. Ferrara, G. P. Incremona, and E. Regolin, Optimization-
29. Z. Li et al., Robust adaptive control of uncertain force/motion based adaptive sliding mode control with application to vehicle
constrained nonholonomic mobile manipulators, Automatica, dynamics control, Int J Robust Nonlinear Cont, 29 (3) (2019),
44 (3) (2008), 776–784. 550–564.
30. J. Peng, J. Yu, and J. Wang, Robust adaptive tracking control for 48. A. Korayem, S. Nekoo, and M. Korayem, Sliding mode control
nonholonomic mobile manipulator with uncertainties, ISA design based on the state-dependent Riccati equation: Theoretical
Trans, 53 (4) (2014), 1035–1043. and experimental implementation, Int J Cont, 92 (9) (2019),
31. N. Chen et al., An adaptive sliding mode backstepping control 2136–2149.
for the mobile manipulator with nonholonomic constraints, 49. K.-Y. Chen, Robust optimal adaptive sliding mode control with
Commu Nonlinear Sci Nume Simu, 18 (10) (2013), 2885–2899. the disturbance observer for a manipulator robot system, Int J
32. I. Matraji et al., Trajectory tracking control of skid-steered mobile Control Autom Syst, 16 (4) (2018), 1701–1715.
robot based on adaptive second order sliding mode control, Cont 50. J. A. Primbs, V. Nevistic, and J. C. Doyle, Nonlinear optimal
Eng Pract, 72 (2018), 167–176. control: A control Lyapunov function and receding horizon
33. W. Tang and Y. Cai, High-order sliding mode control design perspective, Asian J Cont, 1 (1) (1999), 14–24.
based on adaptive terminal sliding mode, Int J Robust 51. M. Pahlevaninezhad et al., A nonlinear optimal control
Nonlinear Cont, 23 (2) (2013), 149–166. approach based on the control-Lyapunov function for an AC/DC
34. A. Erreygers et al., Optimal control of a linear system subject to converter used in electric vehicles, IEEE Trans Indust Inform,
partially specified input noise, Int J Robust Nonlinear Cont, 8 (3) (2012), 596–614.
29 (12) (2019), 3892–3914. 52. E. D. Sontag, A ‘universal' construction of Artstein's theorem on
35. M. Korayem, R. A. Esfeden, and S. Nekoo, Path planning nonlinear stabilization, Sys Cont Lett, 13 (2) (1989), 117–123.
algorithm in wheeled mobile manipulators based on motion of 53. N. Marchand, S. Durand, and J. F. G. Castellanos, A general
arms, J Mech Sci Technol, 29 (4) (2015), 1753–1763. formula for event-based stabilization of nonlinear systems, IEEE
36. A. Shafei and M. Korayem, Theoretical and experimental study Trans Auto Cont, 58 (5) (2012), 1332–1337.
of dynamic load-carrying capacity for flexible robotic arms in 54. H. Wang et al., Finite-time synchronization of uncertain unified
point-to-point motion, Opt Cont App Method, 38 (6) (2017), chaotic systems based on CLF, Nonlinear Anal Real World App,
963–972. 10 (5) (2009), 2842–2849.
37. H. Ma, M. Chen, and Q. Wu, Inverse optimal control for 55. W. Yu, Finite-time stabilization of three-dimensional chaotic
unmanned aerial helicopters with disturbances, Opt Cont App systems based on CLF, Phys Lett A, 374 (30) (2010), 3021–3024.
Method, 40 (1) (2019), 152–171. 56. Y. Yang, A time-specified nonsingular terminal sliding mode
38. M. Korayem, N. Lademakhi, and S. Nekoo, Application of control approach for trajectory tracking of robotic airships,
the state-dependent Riccati equation for flexible-joint arms: Nonlinear Dynamics, 92 (3) (2018), 1359–1367.
Controller and estimator design, Opt Cont App Method, 39 (2) 57. H. Komurcugil, Adaptive terminal sliding-mode control strategy
(2018), 792–808. for DC–DC buck converters, ISA Trans, 51 (6) (2012), 673–681.
SHAFEI AND BAHRAMI 2823

58. F. Plestan et al., New methodologies for adaptive sliding mode AUTHOR BIOGRAPHIES
control, Int J Cont, 83 (9) (2010), 1907–1919.
59. J. M. T. Thompson and H. B. Stewart, Nonlinear dynamics and
chaos. John Wiley & Sons, 2002.
Hamid Reza Shafei was born in
60. H. Mirzaeinejad and A. M. Shafei, Modeling and trajectory Kerman, Iran, in 1988. He received
tracking control of a two-wheeled mobile robot: Gibbs–Appell his B.Sc. degree (with first class
and prediction-based approaches, Robotica, 36 (10) (2018), honors) from Shahid Bahonar Uni-
1551–1570. versity of Kerman in 2012 and then
61. M. Korayem, A. Shafei, and H. Shafei, Dynamic modeling of obtained his M.Sc. degree from
nonholonomic wheeled mobile manipulators with elastic joints Amirkabir University of Technology
using recursive Gibbs–Appell formulation, Scientia Iranica,
in 2014 in Mechanical Engineering. His research
19 (4) (2012), 1092–1104.
62. M. Korayem et al., Theoretical and experimental investigation of
interests mostly include dynamic and control robotic
viscoelastic serial robotic manipulators with motors at the joints systems, dynamic of elastic mechanical manipulators,
using Timoshenko beam theory and Gibbs–Appell formulation, nonlinear control, and adaptive control.
Proc Inst Mech Eng, Part K: J Multi-Body Dynamics, 230 (1)
Mohsen Bahrami received his B.-
(2016), 37–51.
63. A. M. Shafei and H. R. Shafei, Considering Link Flexibility in Sc. degree from Amirkabir University
the Dynamic Synthesis of Closed-Loop Mechanisms: A General of Technology (Tehran Polytechnic),
Approach, J Vibration Acoustics, 142 (2) (2020). 021004. Iran, in 1975. He obtained his M.Sc.
64. A. Shafei and H. Shafei, A systematic method for the hybrid and Ph.D. degrees in Mechanical
dynamic modeling of open kinematic chains confined in a closed Engineering from Oregon State Uni-
environment, Multibody Syst Dyna, 38 (1) (2016), 21–42. versity, USA, in 1977 and 1982,
65. A. Shafei and H. Shafei, Dynamic modeling of planar
respectively. He is now Professor of Mechanical Engi-
closed-chain robotic manipulators in flight and impact phases,
Mech Mach Theory, 126 (2018), 141–154.
neering at Amirkabir University of Technology. He
66. A. Shafei and H. Shafei, Planar multibranch open-loop robotic has been member of numerous national committees,
manipulators subjected to ground collision, J Comp Nonlinear and has authored and co-authored many articles and
Dynamics, 12 (6) (2017). 061003. books. His research interests are dynamic and control,
67. J. H. Lee et al., Sliding mode control for trajectory tracking of robotics, medical robotics, future studies, and technol-
mobile robot in the RFID sensor space, Int J Control Autom Syst, ogy forecasting.
7 (3) (2009), 429–435.
68. H. Wang, X. Zhao, and Y. Tian, Trajectory tracking control of XY
table using sliding mode adaptive control based on fast double
power reaching law, Asian J Cont, 18 (6) (2016), 2263–2271.
69. M. Korayem, M. Nazemizadeh, and V. Azimirad, Optimal How to cite this article: Shafei HR, Bahrami M.
trajectory planning of wheeled mobile manipulators in cluttered Trajectory tracking control of a wheeled mobile
environments using potential functions, Scientia Iranica, 18 (5) robot in the presence of matched uncertainties via
(2011), 1138–1147. a composite control approach. Asian J Control.
70. Luo, S., et al., Wheeled mobile robot RBFNN dynamic 2021;23:2805–2823. https://doi.org/10.1002/asjc.
surface control based on disturbance observer. ISRN Applied
2418
Mathematics, 2014. https://doi.org/10.1155/2014/634936

You might also like