Hindawi

Mathematical Problems in Engineering

Volume 2021, Article ID 8853441, 15 pages
https://doi.org/10.1155/2021/8853441

Research Article
Adaptive Observer-Based Sliding Mode Control for a
Two-Wheeled Self-Balancing Robot under Terrain
Inclination and Disturbances

Ines Jmel,1 Habib Dimassi ,1,2 Salim Hadj-Said,1 and Faouzi M’Sahli1
1
University of Monastir, Ecole Nationale d’Ingénieurs de Monastir, LAS2E, Monastir 5019, Tunisia
2
University of Sousse, Institut Supérieur des Sciences Appliquées et de Technologie de Sousse, Sousse, Tunisia

Correspondence should be addressed to Habib Dimassi; [email protected]

Received 3 September 2020; Revised 28 November 2020; Accepted 24 December 2020; Published 7 January 2021

Academic Editor: Raúl Villafuerte-Segura

Copyright © 2021 Ines Jmel et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
This paper investigates an output feedback sliding mode control scheme for a two-wheeled self-balancing robot under terrain
inclination and disturbances. First of all, an adaptive high-gain observer is designed for the robot to estimate, simultaneously, the
unmeasured states and the unknown terrain inclination angle which appears nonlinearly in the dynamics of the wheeled robot,
using the only measured linear and angular positions. Then, the estimated states and the reconstructed unknown inclination angle
are used by an appropriate continuously implemented sliding mode controller whose the design is based on the boundary layer
approximation approach to reduce the chattering phenomenon. The objective of the proposed robust controller is to ensure the
tracking control of the two-wheeled robot despite the unknown terrain inclination and the presence of friction disturbances. The
stability of the adaptive observer-based output feedback system is established through a Lyapunov analysis, and it is inspired from
sliding modes theory. Numerical simulations results highlight the effectiveness of the proposed tracking control scheme applied
on two-wheeled self-balancing robot subject to terrain inclination even in the presence of unavailable disturbances.

1. Introduction effectiveness as a rigorous method in robotics and control

applications. In [10], a fuzzy control scheme was proposed for a
The two-wheeled self-balancing robot is an exceptional type of two-wheeled inverted pendulum. Also, in [11], a nonsingleton
wheeled mobile robots which has been broadly applied in many general type-2 fuzzy logic controller was designed for an
areas such as hostile terrain, agriculture, and industry. The hard underactuated mobile two-wheeled self-balancing robot. In
nonlinearity and the underactuation of these systems make the [12, 13], an adaptive fuzzy control design was performed for
trajectory tracking control a challenging research problem dynamic balance and stable tracking of desired trajectories for
[1, 2]. Some classical control strategies such as backstepping wheeled inverted pendulum in the presence of uncertainties.
[3, 4] and adaptive controllers [5] are treated by several re- In addition, a PD-PI controller based on Kalman filter
searchers to control the movement of the wheeled inverted algorithm was designed in [14] to stabilize a two-wheeled
pendulum. In [6, 7], the backstepping technique is combined self-balancing robot while avoiding acute and dynamic
with an adaptive controller to ensure the stabilization and the obstacles in the sensed environment. Furthermore, common
convergence of the tracking error. The authors in [8] repre- networked control strategies have been implemented in [15]
sented a combination between H∞ and backstepping tech- for stabilizing a two-wheeled inverted pendulum robot over
niques to stabilize tracking errors under bounded disturbances. a wireless channel despite time-varying delays and paket
In [9], a Kalman filter-based optimal (H-infinity) controller loss. Also, a new feedback reinforcement learning method
was proposed for the linearized model of the wheeled pen- was proposed in [16] to solve the LQR control problem for
dulum. Moreover, Fuzzy systems had attested their the two-wheeled self-balancing robot. The suggested method
2 Mathematical Problems in Engineering

scheme was completely online and did not require any stability of the two-wheeled self-balancing robot moving on an
knowledge of the system parameters. inclined terrain [30–33]. In [34], the authors had shown the
In any formulation of a control problem, the mathematical effect of terrain inclination on the performance and stability
model developed to establish the control law does not reflect region of two-wheeled mobile robots. Moreover, they estab-
the actual process exactly. These differences may be due to, for lished and analyzed the dynamic model on slope of the treated
example, the direct approximation of complex process be- robot. The linear quadratic regulation (LQR) method was used
haviors and the variations of system parameters or unmodeled to design a linear controller whose efficiency was presented
dynamics, especially for systems which are characterized by an through simulation results. In [35], a disturbance observer was
inherent instability, nonlinearity, and underactuation like the used to estimate the handling force and the slope angle when
two-wheeled self-balancing robots. Nevertheless, it must be getting on and off a human-riding wheeled inverted
ensured that, despite all these uncertainties, the resulting pendulum vehicle. A sliding mode control was used in [33] for a
control should reach the predefined objectives. For this reason, two-wheeled inverted pendulum mobile robot driving on
it is necessary to synthesize a robust controller able to overcome uniform slopes. Also, to eliminate the influence of the inclined
this problem. Sliding mode control is an efficient robust control plane considered as external disturbances, the authors had
method which has the advantage to be insensitive to distur- developed, in [36], an active disturbance rejection control
bances and characterized by the finite-time convergence scheme for two-wheeled self-balancing robots which achieved
property and the simplicity of its implementation [17]. In this control aims even in the presence of a slope. However, most of
context, a sliding mode controller was developed, in [18, 19], mentioned works had resorted to linearization in order to design
for the control of underactuated systems. In [20], the authors controller. To deal with problems related to rough terrain,
discussed the control problem and gave design technology of wheeled robots should consistently recognize the current situ-
the sliding mode controller for the two-wheeled mobile robot ation of the surface in order to preserve wheel traction and
with lower center of gravity. The validity of the proposed battery energy. The solution that we adopt is to develop an
controller was analyzed through numerical simulations. adaptive observer to estimate unmeasured states and the un-
Likewise, the authors in [21, 22] developed a sliding mode known terrain inclination angle.
controller for nonholonomic mobile robot in order to realize a In fact, specific attention has been booked to the design of
tracking trajectory in the presence of model uncertainties, adaptive observers in order to ensure, under specific condi-
frictional disturbances, and measurement noise. Moreover, a tion, a joint estimation of unmeasured states and system
novel LMI-based sliding mode controller has been proposed in parameters with exponential convergence. Various works had
[23] for the control of a class of underactuated systems which focused to design an observer for a linear system as it was
are featured as in cascade form with external disturbances. In improved in [37]. Others were interested in nonlinear systems
[24], an optimization-based nonlinear controller was designed with linear parameterization as in [38, 39] and nonlinear
for trajectory tracking for nonholonomic wheeled mobile parameterization as in [40], where the authors had developed
robots. In the same context, a fast terminal sliding mode an adaptive observer for the nonlinearly parameterized class
strategy has been designed in [25] for finite-time tracking of nonlinear systems. The exponential convergence is
control of nonholonomic systems. The efficiency of this achieved under certain persistent condition by the adjustment
method was illustrated to apply on a wheeled mobile robot as a of the gain observer. Simulation results highlighted the ef-
benchmark of a nonholonomic system. Besides, two sliding fectiveness of the developed observer. More recently, the
mode controllers were designed in [26] to control the balancing robustness of the latter adaptive observer has been improved
and the steering movement of a two-wheeled inverted pen- in [41] by adding a sliding mode term. The proposed adaptive
dulum robot with friction compensation. In addition a sliding observer in [41] was also combined with an auxiliary high gain
mode control was designed in [27] to track the smooth curved observer to satisfy the so-called observer matching condition
welding path. In [28], a sliding mode velocity control was and applied for the inverted pendulum system to solve the
proposed for mobile wheeled inverted pendulum systems. problem of simultaneous estimation of states, unknown pa-
More recently, in [29], the authors proposed a sliding mode rameter (mass variation parameter), and friction disturbances
controller (SMC) associated to a high-order disturbance ob- with experimental validation.
server (HODO) for mobile wheeled inverted pendulum In this context, since the inclination parameter appears
(MWIP) systems. The stability of the closed loop system and nonlinearly in the dynamics of the self-balancing robot, we
the convergence of the estimation errors have been proved with adopt the adaptive estimation approach developed in [40] to
a Lyapunov analysis, and theoretical results have been also deal with terrain inclination.
validated experimentally. Motivated by the incontestable ad- Then, the estimated states and parameters are combined
vantages of the latter robust control method, we propose, in this with the continuously implemented sliding mode control
paper, a continuously implemented sliding mode controller for law in order to provide a robust adaptive observer-based
the two-wheeled self-balancing robot where the boundary layer sliding mode controller. The convergence of the estimated
approximation method is adopted in the designed controller to tracking error is established through the Lyapunov analysis.
reduce the well-known chattering phenomenon often linked to In summary, the two-wheeled self-balancing robot is a
the use of discontinuous controllers. benchmark of underactuated system. The hard nonlinearity
For outdoor applications, mobile robots have to face more of this system makes its control an interesting issue.
challenges such as uneven, sloped, or rough terrain surface. This For outdoor application, the two-wheeled self-balancing
invites researchers to design controllers in order to control the robot is always faced to several phenomenons such as
Mathematical Problems in Engineering 3

sloping and skidding that can cause a loss of system stability. n × n dimension. 0n×m is a null matrix with n rows and m
Hence, motivated by the advantages of the sliding mode columns. λmax (M) and λmin (M) denote, respectively, the
control as a robust controller and to deal with the variation maximum and the minimum eigenvalue of M.
of terrain inclination, we suggest, in this paper, an adaptive diag(M1 , . . . , Mn ) represents the block-diagonal
observer-based sliding mode controller for a two-wheeled M1 0 · · · 0
⎛ 0 ⋱ ⋱ ⋮⎟
⎜
⎜ ⎞
⎟
matrix � ⎜
⎜
⎝⋮ ⋱ ⋱ ⋮⎟
⎜ ⎟
⎟
self-balancing robot subject to terrain inclination. From a
theoretical viewpoint, the demonstration of the convergence ⎜ ⎟
⎠, where M1 , . . ., Mn are square
of the adaptive observer with nonlinear parameterization 0 · · · 0 Mn
and in the presence of bounded disturbance will be firstly matrices. R+ ∈ Rn×m denotes the pseudoinverse (generalized
established based on a Lyapunov analysis; then, the analysis inverse) of a matrix R ∈ Rm×n . That is, R+ is a matrix such
of the stability of the output feedback tracking control that RR+ R � R and R+ RR+ � R+ .
system composed of the adaptive observer and the con-
tinuously implemented sliding mode controller will be
carried out based on sliding modes theory. In summary, the 2. Context and Problem Statement
main contributions of this paper from both theoretical and 2.1. Model of a Two-Wheeled Self-Balancing Robot. The
application viewpoints are given as follows: two-wheeled self-balancing robot is an underactuated sys-
(i) Adaptive estimation of the unmeasured states tem, and it has fewer actuators (2 actuators) than degrees of
and the unknown terrain inclination parameter freedom (3 degree of freedom). The dynamics of this robot is
which appears nonlinearly in the dynamics of the similar to that of a pendulum cart. In fact, the two-wheeled
two-wheeled self-balancing robot in the presence torques CL and CR are transformed, using a decoupling unit,
of disturbances into two torques Cδ and Cψ . The torque Cψ supervises both
(ii) Design of the robust adaptive output feedback the robot position and the pitch angle. It ensures the
controller (adaptive observer + sliding mode con- translation along the x-axis while preserving the upright
troller) for the tracking control of the two-wheeled position, whereas the torque Cδ controls the right and left
robot despite the unknown terrain inclination and rotation of the robot which ensures the desired yaw
friction disturbances trajectory.
Referring to [42–44], the model of the two-wheeled
(iii) Proof of the convergence of the designed adaptive self-balancing robot is as follows:
observer and stability analysis of the closed-loop
system based on the Lyapunov analysis x_ 1 � x2 ,
The remainder of the paper is organized as follows. The (1)
x_ 2 � F(x, ρ) + G(x, ρ)u + d(t),
next section presents the problem statement and prelimi-
naries. A robust adaptive observer-based sliding mode .
where x1 � [x, ψ, δ]T , x2 � [x, tψn,
_ qδ] _ T , and x � [x , x ]T .
1 2
control, as well as the convergence of the tracking control u � [Cψ , Cδ ]T is the control input vector. x, ψ, and δ rep-
scheme, is analyzed in Section 3. Simulation results through resent, respectively, the linear displacement of the chassis,
Matlab/Simulink assert the performance of the proposed the pitch angle, and the yaw angle. d(t) � [d1 (t),
method in Section 4. Finally, we conclude the paper with d2 (t), d3 (t)]T represents the disturbance vector. d(t) is
conclusive remarks in Section 5. assumed to be bounded and differentiable with respect to
time. For each t ≥ 0, |d(t)| ≤ ζ, where ζ is a positive number.
1.1. Notations. | · | represents the euclidean norm for vectors G(x, ρ) � [g11 (x, ρ), 0; g21 (x, ρ), 0; 0, g32 (x, ρ)] and
and induced norm for matrices. In is an identity matrix with F(x, ρ) � [f1 (x, ρ), f2 (x, ρ), 0]T , where

􏽨Mp L R g sin(ψ)cos(ψ + α) + 􏼐Mp L + Jp 􏼑􏼐Mp + 2Mr 􏼑gR sin(α) − 􏼐Mp L + Jp 􏼑Mp LR ψ_ sin(ψ + α)􏽩f2 (x, ρ)
1 2 2 2 2 2 2 2 2
f1 (x, ρ) �
B

� 􏽨M2p L2 R2 ψ_ 2 sin(ψ + α)cos(ψ + α) − 􏼐2Mr R2 + Mp R2 + 2Jr 􏼑Mp gL sin(ψ) + Mp 􏼐Mp + 2Mr􏼑LR2 g sin(α)cos(ψ + α)􏽩,
1
B

􏽨􏼐Mp L + Jp 􏼑R + Mp LR cos(ψ + α)􏽩,

−2 2 2
g11 (x, ρ) �
B

􏽨Mp LR cos(ψ + α) + 􏼐2Mr R2 + Mp R2 + 2Jr 􏼑􏽩,

−2
g21 (x, ρ) �
RB

with B � M2p L2 R2 cos2 (ψ + α) − 􏼐2Mr R2 + Mp R2 + 2Jr 􏼑􏼐MpL2 + Jp􏼑.

2D
g32 (x, ρ) �
􏽨 Mr R + Jr /R􏼁􏼁D2 + 2Jδ R􏽩
(2)
4 Mathematical Problems in Engineering

􏼌􏼌 􏼌􏼌
Jp is the moment of inertia of the chassis with respect to lim 􏼌􏼌x(t) − xr (t)􏼌􏼌 � 0, (4)
the z-axis, Jδ is the moment of inertia of the chassis with t⟶∞

respect to the y-axis, and Jr is the moment of inertia of the where xr (t) is the reference trajectory.
wheel. Mp is the total mass of the robot. D is the lateral
distance between the contact patches of the wheels, R is the
radius of the wheels, and Mr is its mass. 3. Adaptive Observer-Based Sliding Mode
ρ � α is the angle of terrain inclination, which is assumed Control for the Two-Wheeled Self-
unknown and which will be estimated later through an Balancing Robot
adaptive observer.
In this section, we present first the adaptive observer design
method for the wheeled robot, and then, we combine it with
2.2. Problem formulation. The two-wheeled self-balancing the sliding mode controller whose objective is to ensure
robot is a nonlinear MIMO underactuated system; thus, it is tracking control despite the presence of unknown terrain
very challenging to keep balance when it climbs or descends inclination and disturbances. Before describing the adaptive
on a slope and, especially, in the presence of nonmeasurable observer-based sliding mode control system, the following
disturbances. In this work, the considered problem consists assumption must be taken into account.
in designing an output feedback sliding mode controller for
the self-balanced robot model subject to unknown distur- Assumption 1. For any bounded control input u, the state x
bances, to generate a robust command for going up and is assumed bounded. Moreover, the unknown parameter ρ is
down the slope. In case of going up the slope, the controller also assumed to be bounded and slowly varying, that is,
generates an acceleration of the gear-motor driving the ρ(t)
_ � 0, almost everywhere.
wheels, whereas in the going down case, a deceleration is It is to be noticed that some physical autonomous
performed for the safety of the vehicle. Such behavior is systems such as chaotic systems verify the boundedness
required for ensuring the global engine stability. Despite the assumption of the trajectories naturally (Assumption 1). In
importance of the studied issue, according to our knowledge, the case of controlled systems (nonautonomous systems)
theoretical developments dealing with the terrain inclination which include the two-wheeled self-balancing robot con-
and the disturbances simultaneously for the nonlinear model sidered in this paper, the boundedness of the states may be
of the two-wheeled self-balancing robot are infrequent. guaranteed by synthesizing an appropriate bounded control
In addition, it is well known that the dynamics of input. We notice also that assuming the unknown angle
underactuated systems such as the two-wheeled self-bal- inclination parameter ρ(t) � α(t) is usually satisfied in
ancing robot may contain hard nonlinearities and non- practice in the self-balancing robot system under terrain
holonomic constraints which make the control of these inclination, which is trivially reasoning from a physical
systems an open and interesting issue. These features should viewpoint.
be considered during control design in order to establish a By considering Assumption 1, let (u, x, ρ) ∈ (U, X, Θ),
robust controller which guarantees the system performance where U ⊂ R2 , X ⊂ R6 , and Θ ⊂ R are three compact sets.
even under different operating conditions. Sliding mode For the two-wheeled self-balancing robot system, the
control is an efficient robust controller, thanks to its ef- nonlinearities F(x, ρ) and G(x, ρ) are just once continuously
fectiveness for dealing with uncertain and disturbed systems. differentiable but not globally Lipschitz. In such a case, we
Persuaded by its advantages, we adopt this popular tech- use, under Assumption 1, the Lipschitz prolongation ap-
nique to ensure the tracking objectives. proach in order to build prolongations F 􏽥 of the
􏽥 and G
Furthermore, when a mobile robot moves on a trajec- nonlinearities F and G using saturation functions [40, 45],
tory, several phenomenons such as sloping and skidding can where F 􏽥 and G 􏽥 are globally Lipschitz and defined as
cause a loss of wheel traction and battery energy, as well as 􏽥
F(x, ρ) � F(σ(x), σ ρ (ρ)) and G(x, 􏽥 ρ) � G(σ(x), σ ρ (ρ)),
system stability. Thus, having an instantaneous knowledge of where σ: Rn ⟶ X, x↦σ(x) and σ ρ : R ⟶ Θ, ρ↦σ ρ (ρ)
the terrain inclination angle could be lucrative to keep the are smooth bounded saturation functions such that σ(x) � x
performance of the adopted control law. and σ ρ (ρ) � ρ for all x ∈ X and ρ ∈ Θ.
The solution to deal with this problem is to design an In this way, the trajectories of the wheeled robot system
adaptive observer to estimate conjointly unmeasured states (46) coincide with those of the following system:
_ ψ,
(x, _ and the angle of terrain inclination which represents
_ δ)
x_ 1 � x2 ,
the unknown parameter ρ � α. That is,
􏽥 􏽥
(5)
x_ 2 � F(x, ρ) + G(x, ρ)u + d(t),
􏽢 (t)| � 0,
􏽥
lim |x(t) − x
t⟶∞
where G(x, ρ) � [􏽥 􏽥 21 (x, ρ), 0; 0, g
g11 (x, ρ), 0; g 􏽥 32 (x, ρ)] and
􏽥 (x, ρ), f􏽥 (x, ρ), 0]T are the Lipschitz extension
(3)
lim |ρ(t) − ρ􏽢(t)| � 0. 􏽥
F(x, ρ) � [f 1 2
t⟶∞
functions of G(x, ρ) and F(x, ρ), respectively.
Then, the estimated states x 􏽢 (t) and the reconstructed In the rest of the paper, we will rather focus on system (5)
unknown parameter 􏽢ρ(t) generated by the adaptive observer for the design of our proposed adaptive observer-based
are to be injected into the sliding mode control law in order tracking control approach for the considered two-wheeled
to perform the following trajectory tracking aims: self-balancing robot.
Mathematical Problems in Engineering 5

3.1. Adaptive Observer for States and Terrain Inclination 􏽥

⎨ x_ � Ax + f(x,
⎧ 􏽥 (x, ρ)u + B d(t),
ρ) + g
Angle Estimation. Adaptive observer design represents a ⎩ (6)
popular method for states and unknown parameter esti- y � Cx � x1 ,
mation. Since the unknown terrain inclination angle appears
A�􏼢 􏼣, B�􏼢 􏼣, and C � 􏼂 I3 , 03 􏼃.
03 I3 03
nonlinearly in the dynamic of the two-wheeled self-bal- where
ancing robot and given the triangular structure of our 03 03 I3
system, we adopt the adaptive observer treated in [40] and g
0 􏽥
􏽥 (x, ρ) � 􏼢 􏽥 3×2 􏼣, f(x, ρ) � 􏼢 􏽥 3×1 􏼣, and x � 􏼢 1 􏼣.
0 x
we recall, in this section, the main features of this estimation G(x, ρ) F(x, ρ) x2
approach. Conforming to [40], the adaptive observer developed for
Actually, the two-wheeled self-balancing robot system system (6) is given by
(5) is included in the class of systems considered in [40] and
may be written in the following form:

⎪
⎧
⎪ 􏽢_ � A􏽢 􏽥 (􏽢 􏽥 x, 􏽢ρ) − θΔ− 1 􏼒S− 1 + Y(t)P(t)YT (t)􏼓CT K(C􏽥
x, 􏽢ρ)u + f(􏽢
⎪
⎪
⎪
x x+g x),
⎪
θ
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ ρ􏽢_ (t) � − θP(t)YT (t)CT K(C􏽥
⎪
⎪
x),
⎨
⎪
⎪
(7)
⎪
⎪ 􏽥
⎪
⎪ Y(t) � θ􏼒A − S CT C􏼓Y(t) + Δθ (􏽢
zf
x, ρ􏽢) + Δθ (􏽢
z􏽥
g
x, ρ􏽢)u, with Y(0) � 0,
⎪
− 1
⎪
_
⎪
⎪
⎪
zρ zρ
⎪
⎪
⎪
⎪
⎪
⎩ _
P(t) � − θP(t)YT (t)CT CY(t)P(t) + θP(t), with P(0) � PT (0) > 0,

􏽢1
􏽢�􏼢 􏼣; Δθ � diag[I3 , 1/θI3 ], where θ > 0 is a real
x presence of the disturbance d(t). We consider the Lyapunov
􏽢2
where x function
x
􏽥�x 􏽢 − x, and S is the unique solution of the
W(t) � Ω(t)T SΩ(t) + ρ􏽥T (t)P− 1 􏽥ρ(t),
number, x
(9)
following algebraic Lyapunov equation:
S + AT S + SA − CT C � 0, (8) where
Ω(t) � x(t) − Y(t)􏽥ρ(t),
where S is the symmetric positive definite (SPD) and the 􏼨 (10)
− 1
matrix (A − S CT C) is Hurwitz. K is a design matrix of x(t) � Δθ (􏽢
x − x).
appropriate dimensions to be determined later.
Proceeding as in Theorem 3.1 in [40] and taking into
For the convergence of the unknown parameters esti-
account the presence of the perturbation term d(t), one may
mation errors, the following assumption is required.
obtain
Assumption 2. For any trajectory (􏽢 x, t􏽢ρ) ∈ X × Ω, the ma- _ ≤ − 2μW − 2ΩT SΔθ B d(t)
W
trix CΥ(t) must satisfy the condition of persistent excitation, 􏽱������� 􏽰������
and it means: ∃δ1 , δ2 > 0; ∃T > 0; ∀t ≥ 0: δ1 Im ≤ 􏽒t
t+T
≤ − 2μW + 2|B|ζ λmax (S)θ− 1 W(Ω)
YT (t)CT CY(τ)dτ ≤ δ2 Im . 􏽰������ 􏽰������ 􏽱�������
The persistency of excitation is a classical condition usually ≤ − μW − μ W(Ω)􏼔 W(Ω) − 2|B|ζμ− 1 λmax (S) θ− 1 􏼕,
adopted in the literature of adaptive estimation. It is behind the
(11)
asymptotic stability and parametric convergence, and it sig-
nifies that the studied system is sufficiently rich in frequencies. where we recall that ζ is a positive constant such that, for all
t ≥ 0, |d(t)| ≤ ζ and
Proposition 1. Consider the system (6) subject to the un-
known parameter ρ(t) and the disturbance d(t), together μ�
1 ⎜
θ − c1 􏼁⎛
⎜
⎝1 − 􏽱�������
c2
�⎞ ⎟
⎟
⎠, (12)
with the adaptive observer (7), under Assumptions 1 and 2. 2 2 θ θ − c1 􏼁
Then, the state estimation error x 􏽥�x 􏽢 − x and the adaption
error ρ􏽥 � 􏽢ρ − ρ converge to a small compact set whose radius where c1 and c2 are positive constants which depend on the
may be reduced by choosing sufficiently high values of the upper bounds of the different bounded signals and on the
design parameter θ. minimal and maximal eigenvalues of the matrices S and P
(see the proof of Theorem 3.1 in [40]).
Proof. The proof of Proposition 1 is mainly based on the 􏽰Next, ������� (11), we deduce that as long
�����from the last􏽱inequality
proof of Theorem 3.1 in [40] while taking into account the as W(t) ≥ 2|B|ζμ− 1 λmax (S)θ− 1 , one has
6 Mathematical Problems in Engineering

_
W(t) ≤ − μW(t). (13) where λ is a positive constant.
􏽰����� 􏽱������� Inspired from sliding modes theory, we propose the
If W(0) ≥ 2|B|ζμ− 1 λmax (S)θ− 1 , Ω(t) and ρ􏽥(t) will following sliding mode controller incorporating the esti-
decrease exponentially such that mated states generated by the adaptive observer (7):
W(t) ≤ W(0)e− μt . (14) 􏽥 x, t􏽢ρ)]+ 􏼢x€r − F(􏽢
u � [G(􏽢 􏽥 x, t􏽢ρ) + λ x_ r − x
􏽢2􏼁 − k 􏼣 ,
S
|S|
Consequently, there exists a finite time Ta such that for
all t ≥ Ta , (20)
􏽰����� 􏽱�������
where k is a positive constant.
W(t) ≤ 2|B|ζμ− 1 λmax (S)θ− 1 . (15)
However, the presence of the discontinuous function, in
the sliding mode control, gives rise to the well-known
We have
􏽱������
1 􏽱��������� 􏽰�����
chattering phenomenon which represents the main disad-
√� λmin (S)|Ω(t)| + √� λmin 􏼐P− 1 􏼑|􏽥ρ(t)| ≤ W(t).
1 vantage of sliding modes controllers. To overcome this
2 2 problem, it is lucrative to transform the discontinuous
(16) function by a continuous one using the boundary layer
approximation approach to eliminate the chattering phe-
Next, combining inequalities (15) and (16) and using nomenon [46]. Then, we rather consider the following
􏽥 (t) and
(10), it may be deduced that the estimation error x “continuously implemented sliding mode controller” in-
the adaptation error ρ􏽥(t) converge to a compact set whose stead of (20):
radius may be reduced by choosing a sufficiently large value
of the design parameter θ. 􏽥 x, t􏽢ρ)]+ 􏼢x€r − F(􏽢
u � [G(􏽢 􏽥 x, t􏽢ρ) + λ x_ r − x
􏽢2􏼁 +
kS
+ βS􏼣,
Once the unmeasured states and the unknown param- |S| + ε
eter are reconstructed by the adaptive observer (6), the (21)
􏽥 x, t􏽢ρ)]+ represents the pseudoinverse matrix of
estimated signals are employed, in the next section, by the
where [G(􏽢
􏽥
sliding mode control law in order to build an adaptive
sliding mode controller. □ G(􏽢 x, t􏽢ρ). β and ε are positive constants.
Before proving the convergence of the tracking error of
the closed loop system, we show first, in the proof of the
3.2. Adaptive Observer-Based Sliding Mode Control for the following theorem, that S(t) converges to a compact set
Two-Wheeled Self-Balancing Robot. In this section, we apply whose radius may be made as small as possible by adjusting
the sliding mode control on the two-wheeled self-balancing the design parameters ε and β.
robot and we replace the unknown parameter ρ � α and the
_ ϕ,
speed states (x, _ which are not accessible to mea-
_ δ),
surement by their estimates obtained from the proposed Theorem 1. We consider the system (6) under the continu-
adaptive observer as it is denoted in Figure 1. ously implemented sliding mode control input (21) and in-
The problem of trajectory tracking consists to determine corporate the adaptive observer (7) such that Assumptions 1
the control law u which insures the convergence of the state and 2 are satisfied. Then, S(t) is uniformly bounded with an
vector x1 � 􏼂 x ψ δ 􏼃T to the reference vector xr � upper bound which may be made arbitrarily small by reducing
􏼂 xr ψ r δr 􏼃 with a tracking error on the first component of the design parameter ε and increasing the design parameter β.
T

the state vector:

e � xr − x1 . (17) Proof. Let us start with an appropriate Lyapunov function:
1
Affected by the influence of terrain inclination α, the pitch V � ST S. (22)
2
angle ψ r will not be identical to zero when the two-wheeled
self-balancing robot becomes stable on the slope. In fact, ψ r Differentiating this function and referring to (19), one
will become a constant ψ eq which depends on α, and it is obtains
defined as [31] 􏽥_ 2 􏽩
V_ � SS_ � S􏽨λ x_ r − x2 􏼁 + x€r − x_ 2 􏼁 + x
⎜ ⎟ 􏽥
� S􏽨λ x_ r − x2 􏼁 + x€r − F(x, 􏽥_ 2 􏽩
ρ) − G(x, ρ)u − d(t)􏼁 + x
ψ eq � arccos􏼠 􏼡 − arccos⎛
⎝􏽱������������ ⎞
⎜ ⎟
⎠.
− Mr R sin(α) R sin(α)
Mp 2 2
R sin (α) + L 2
􏽥 x, t􏽢ρ) − G(􏽢
� S[λ x_ r − x2 􏼁 + x€r − F(􏽢 􏽥_ 2
x, t􏽢ρ)u − d(t)􏼁 + x
(18) 􏽥
− F(x, 􏽥
ρ) − G(x, ρ)u + F(􏽢 􏽥 x, 􏽢ρ)u].
􏽥 x, 􏽢ρ) + G(􏽢
We select the sliding variables vector as follows: (23)
􏽢2􏼁
S � λ xr − x1 􏼁 + x_ r − x Let
􏽢2
� λ xr − x1 􏼁 + x_ r − x2 􏼁 + x2 − x (19) 􏽥
􏽢 , ρ, ρ􏽢) � − F(x,
η(x, x 􏽥
ρ) − G(x, 􏽥 x, ρ􏽢)u.
􏽥 x, ρ􏽢) + G(􏽢
ρ)u + F(􏽢
􏽥2,
� λ xr − x1 􏼁 + x_ r − x2 􏼁 + x (24)
Mathematical Problems in Engineering 7

Two-wheeled
Sliding mode control Adaptive observer
self-balancing robot
·
  + [x··τ –· F(x,p)
Ueq = [G(x,p]   – D(t) + · x = Ax + f(x,p)
  + g(x,p)U

x = Ax + f(x,p) + g(x,p)U + B D(t) –1 – –1
–θ∆θ (s + γ(t)P(t)γ (t))
T
λ(x· τ – x)]
 T
C 
K(Cx)
  + [kS/|S| + ε + βS]
U = Ueq – [g(x,p)] y = Cx =x1 ·
p = –θP(t)γT (t)CT K(Cx) 

x

p

Figure 1: Synoptic diagram of the proposed control method.

Now, replacing the control input vector u by its ex- From the convergence of the adaptive observer [40], it
pression described in (21), one has follows that there exists a positive constant kη > 0 such that
􏽢 , ρ, ρ􏽢)| ≤ kη .
􏽢 , ρ, 􏽢ρ) − d(t) + x
V_ � S􏼢η(x, x 􏽥_ 2 − λ􏽥 + βS􏼣
kS |η(x, x (29)
􏽥_ 2 . To that
x2 −
|S| + ε Next, we need to show the boundedness of x
end, conforming to [40], taking into account the presence of
k|S|2
􏽥_ 2 − λ􏽥 􏽢 , ρ, 􏽢ρ) − d(t)􏽩.
the disturbance term d(t) and using (7), we have
� − βS2 − + 􏽨x
􏽥 x, ρ􏽢) + g 􏽥
x2 + η(x, x
|S| + ε 􏽥_ � A􏽥
x x + f(􏽢 x, ρ􏽢)u − (f(x,
􏽥 (􏽢 􏽥 (x, ρ)u)
ρ) + g
(25) (30)
x) + Δ−θ 1 Y􏽥ρ_ − Bd(t).
− 1
− θΔ−θ 1 S CT K(C􏽥
At this stage, it is needful to prove the boundedness of
􏽢 , ρ, ρ􏽢) and x
􏽥_ 2 in order to admit the convergence S(t). 􏽥 and Δc � 􏼢 3 3 􏼣. Referring to (30)
θI 0
η(x, x We set xc � Δc x
􏽢 , ρ, ρ􏽢). We
03 I3
Let us first prove the boundedness of η(x, x
and tacking into account that Δc AΔ−c 1 � θA and CΔc � C, we
have
get
􏽥
􏽢 , ρ, 􏽢ρ) � − F(x, 􏽥 􏽥 x, 􏽢ρ)u.
􏽥 x, 􏽢ρ) + G(􏽢
η(x, x ρ) − G(x, ρ)u + F(􏽢 􏽥 x, 􏽢ρ) + g
x_ c � θ􏼐A − θKo S− 1 CT C􏼑xc + θYρ􏽥_ + Δc (f(􏽢 􏽥 (􏽢
x, ρ􏽢)u
􏽥
(26)
− (f(x, 􏽥 (x, ρ)u) − B d(t)).
ρ) + g
This can be written as
(31)
􏽥
􏽢 , ρ, 􏽢ρ) � − F(x,
η(x, x 􏽥 x, ρ) − F(􏽢
ρ) + F(􏽢 􏽥
􏽥 x, ρ) − G(x, ρ)u
􏽥 x, ρ)u − G(􏽢
􏽥 x, ρ)u + F(􏽢 􏽥 x, ρ􏽢)u
􏽥 x, ρ􏽢) + G(􏽢
Hence,
􏽢 , ρ􏽢),
+ G(􏽢
x_ c � θAxc + W(u, x
􏽥 􏽥
􏽥 x, ρ)) − (G(x, 􏽥 x, ρ))u
(32)
􏽥 x, 􏽢ρ) + g 􏽥
� − (F(x, ρ) − F(􏽢 ρ) − G(􏽢
􏽢 , ρ􏽢) � f(􏽢 􏽥 (􏽢
x, ρ􏽢)u − (f(x, 􏽥 (x, ρ)
􏽥 x, ρ) − F(􏽢 􏽥 x, ρ) − G(􏽢
􏽥 x, 􏽢ρ)) − (G(􏽢 􏽥 x, 􏽢ρ))u. where W(u, x ρ) + g
− (F(􏽢 u) − B d(t) + θYρ􏽥_ and A � A − θKo S− 1 CT C. A is Hurwitz,
(27) and W _ is bounded. In fact, we recall that x(t), u(t), d(t), and
􏽥 􏽥 ρ(t) are assumed to be bounded. Furthermore, z􏽥
􏽥 􏽥 and g
g/zx and
Tacking into account that F(x, ρ) and G(x, ρ) are Lip- 􏽥 are assumed contin-
schitz with respect to x uniformly in ρ and F(􏽢􏽥 x, tρ) and zf/zx are bounded because f
􏽥 x, tρ) are Lipschitz with respect to ρ uniformly in x
G(􏽢 􏽢 , we
uously differentiable.
Therefore, according to Theorem 1 of [47], we can
can write that conclude that limθ⟶+∞ x_ c � 0. As a result, limθ⟶+∞ x 􏽥_ 1 � 0
􏽥_ 2 � 0.
􏽢 , ρ, ρ􏽢)| ≤ kF |x − x
􏽢 | + kG |u||x − x
􏽢|
and limθ⟶+∞ x
|η(x, x Now, let k2 > 0 and k􏽥2 > 0 such that |􏽥 x2 (t)| ≤ k2 and
+ kρ1 |ρ − ρ􏽢| + kρ2 |u||ρ − 􏽢ρ| (28) 􏽥 2 (t)| ≤ k􏽥2 .
_
|x
Returning to (25) and using (29), we deduce that
x| + 􏼐kρ1 + kρ2 ku 􏼑|􏽥ρ|,
≤ kF + ku kG 􏼁|􏽥
+ 􏼐k􏽥2 + λk2 + kη + ζ 􏼑|S|.
k|S|2
where kF , kG , ku , kρ1 , and kρ2 are a positive constants. V_ ≤ − βS2 − (33)
|S| + ε
8 Mathematical Problems in Engineering

Table 1: Two-wheeled self-balancing robot parameters.

Symbol Name Value (unit)
Mp Body weight including the loads 20.82 (kg)
Mr Mass of the wheel 0.420 (kg)
D Distance between the contact patches of the wheels 0.438 (m)
L Distance between the chassis and the center 0.4 (m)
R Radius of the wheel 0.106 (m)
g Gravity constant 9.8 (ms− 2 )
Jp The moment of inertia of the chassis with respect to the z-axis 0.28 (kg·m2 )
Jδ The moment of inertia of the chassis with respect to the y-axis 1.12 (kg·m2 )
Jr Moment of inertia of the wheel 0.1 (kg·m2 )

35 0.1

0.08
30
0.06
25 0.04

Pitch angle (rad)

20 0.02
X (m)

0
15
–0.02
10 0.2 –0.04 0.4
0 0.2
5 –0.06
–0.2 0
–0.08 –0.2
–0.4
0 0 0.2 0.4 0.6 0.8
–0.6 –0.1
0 0.5 1 1.5 2 0 10 20 30 40 50 60 70 80 90 100
0 10 20 30 40 50 60 70 80 90 100 Time (s)
Time (s)
Desired angle
Desired trajectory Estimated angle
Estimated trajectory Real angle
Real trajectory
Figure 3: Response of the pitch angle ψ according to terrain
Figure 2: Linear displacement x. inclination.

which means that S(t) is uniformly bounded and its upper

Supposing ks � k􏽥2 + λk2 + kη + ζ, one obtains
bound (2ks ε/β)1/2 may be made small by reducing the design
parameter ε and increasing the design parameter β.
k|S|2 ks (|S| + ε)|S| Now, it remains to show the convergence of the tracking
V_ ≤ − βS2 −
|S| + ε
+
|S| + ε error of the closed-loop system. □
k|S|2 ks |S|2 ks ε|S|
≤ − βS2 − + + (34) Corollary 1. We consider the system (6) with control input
|S| + ε |S| + ε |S| + ε
(21) and incorporate the adaptive observer (7) such that
|S|2
≤ − βS2 − k − ks 􏼁
|S| Assumptions 1 and 2 are satisfied. Then, the tracking errors of
+ ks ε, since < 1.
|S| + ε |S| + ε the closed-loop system converge to a compact set whose radius
Choosing k sufficiently large such that k > ks , we have may be reduced by choosing small values of the design pa-
rameter ε and by increasing the design parameter β.
_
V(S(t)) ≤ − βS2 (t) + ks ε
(35)
≤ − βV(S(t)) − βV(S(t)) − ks ε􏼁.
Proof. Referring to (17), the derivative of the tracking error
As long as βV(S(t)) ≥ ks ε, i.e, V(S(t)) ≥ ks ε/β, we have becomes
_
V(S(t)) ≤ − βV(S(t)). Then, if V(S(t)) ≥ ks ε/β, S(t) will be 􏽥2.
e_ � S − λe − x (37)
exponentially decreasing, and there exists a finite time Ts
such that, for all t ≥ Ts , (V(S(t)) ≤ )ks ε/β, which means that, Let us consider the Lyapunov function V1 � eT P1 e; its
for all t ≥ Ts , we have derivative is given by
1/2
V_ 1 � 2eT P1 e_ . (38)
|S(t)| ≤ 􏼠 s 􏼡 ,
2k ε
(36)
β
Using (37), this allows to obtain
Mathematical Problems in Engineering 9

0.45 0.5

0.4 0.4

0.35 0.3

0.3 0.2

Angular speed (m/s)

Yaw angle (rad)

0.25 0.1

0.2 0

0.15 0.4 –0.1

0.2 –0.2
0.1 0
0 –0.3
0.05 –2

0 2 4 6 8 10 –4
0 –0.4
0 0.1 0.2 0.3 0.4 0.5
–0.5
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
Time (s) Time (s)

Desired angle Estimated

Estimated angle Real
Real angle
Figure 6: State estimation of pitch angular speed.
Figure 4: Response of the yaw angle δ.

0.8 and (40) can be written as

􏽰��� 􏽰���
V_ 1 ≤ − V1 λ V1 − c3 􏼁 − λV1 . (42)
0.6 􏽰���
Hence, as long as λ V1 − c3 ≥ 0, i.e, V1 ≥ (c3 /λ)2 , we
have V_ 1 (e(t)) ≤ − λV1 (e(t)), which means that V1 (e(t))
Linear speed (m/s)

0.4
will be decreasing exponentially fast until V1 (e(t)) < (c3 /λ)2
after a finite time Tf2 .
0.2 Since λmin (P1 )|e(t)|2 ≤ V1 (e(t)), we have

|e(t)| ≤ 􏽱��������,
λc3
∀t ≥ Tf2 .
λmin P1 􏼁
0 0 (43)
–1
–2
–0.2 By considering the expression (41) of the constant c3 , we
0 0.2 0.4 0.6
deduce, from inequality (43), that the tracking errors of the
0 10 20 30 40 50 60 70 80 90 100 closed-loop system converge to a compact set whose radius
Time (s) may be made as small as possible by reducing the design
Estimated parameter ε and increasing the design parameter β. □
Real

Figure 5: State estimation of linear speed. Remark 1. Based on the different abovementioned dem-
onstrations and using the different assumptions considered
in this paper, we summarize the following algorithm to
􏽥2􏼁
V_ 1 � 2eT P1 S − λe − x describe how the different design parameters are selected
(39)
􏽥2.
� − 2λeT P1 e + 2eT P1 S − 2eT P1 x
and the steps of the implementation process:
Step 1: we compute the matrix S and the solution of the
x2 (t)| ≤ k2 and using (36), one obtains
Reminding that |􏽥 algebraic Lyapunov equation (8). The latter equation
(8) is solvable for S as long as the pair (A, C) is ob-
⎝􏼠2ks ε􏼡
V_ 1 ≤ − 2λV1 + 2|e|λmax P1 􏼁⎛ ⎠
+ k2 ⎞
1/2
servable, which is the case for the two-wheeled self-
β (40) balancing robot considered in this paper. For the choice
􏽰��� of S, we solve the equation S + AT S + SA − CT C � 0
≤ − 2λV1 + c3 V1 , using the pole placement method and the Matlab
function place.
where
Step 2: the choice of θ should ensure a compromise
􏽱��������
c3 � 2 λmin P1 􏼁 λmax P1 􏼁􏼁⎝
⎛􏼠2ks ε􏼡 + k2 ⎠
⎞,
1/2
between fast convergence of the state estimation and
β
(41) the satisfactory dealing with noise rejection (with θ > 1).
10 Mathematical Problems in Engineering

0.5 5

0.4
4
0.3
3
0.2
Angular speed (m/s)

Torque (103·Nm)
0.1 2

0 1
–0.1
0
–0.2 0

–1
–0.3 –1

–0.4 –2
–2
0 1 2 3
–0.5
0 10 20 30 40 50 60 70 80 90 100 –3
0 10 20 30 40 50 60 70 80 90
Time (s)
Time (s)
Estimated
Cdelta
Real
Cpsi
Figure 7: State estimation of yaw angular speed.
Figure 9: The control input u under the adaptive sliding mode
controller in the presence of a slope.
0.35

0.3 1.2

0.25
Angle of terrain inclination (rad)

1
0.2

0.15 0.8
Yaw angle (rad)

0.1
0.6
0.05

0 0.4

0.08
–0.05 0.06
0.04 0.2
0.02
–0.1 0
0 0.5 1 0
–0.15
0 10 20 30 40 50 60 70 80 90 100
Time (s) –0.2
0 10 20 30 40 50 60 70 80 90 100
Estimation of terrain inclination Time (s)
Terrain inclination
Desired trajectory
Figure 8: Estimation of terrain inclination α under the proposed Estimated trajectory
observer. Real trajectory

Figure 10: Response of the yaw angle δ.

Step 3: we choose a large value of the design parameter
􏽥 x, 􏽢ρ)]+ of the
β and a small value of the parameter ε such that
Step 6: we compute the pseudoinverse [G(􏽢
β≫1 􏽥
matrix function G(􏽢 x, 􏽢ρ) (see the definition of the
(44)
0 < ε ≪ 1. pseudoinverse of a matrix at the end of Section 1).

Step 4: we select a positive parameter λ according to the Remark 2. Terminal sliding mode control is an efficient
reference tracking priority that we should to give to the robust control approach that has proved interesting advan-
state x1 or x2 . tages compared to conventional sliding mode control: we
Step 5: we choose a sufficiently large parameter k such
that k > ks with ks � k􏽥2 + λk2 + kη + ζ, where ζ, k2 , and
report, for instance, its fast convergence rate and its high

k􏽥2 are, respectively, the upper bounds of the distur-

tracking accuracy [25]. In this context, an adaptive non-
singular integral terminal sliding mode control approach has
􏽥 2 (t), and its derivative x
bance d(t), the state x 􏽥_ 2 (t); kη is been proposed in [17]. The convergence rate of the latter
the upper-bound of the signal η(x, x 􏽢 , ρ, ρ􏽢) given by control approach was recently improved in [17] with appli-
equation (29). cation to the trajectory tracking control of autonomous
Mathematical Problems in Engineering 11

35

1
30

25 0.5

Linear speed (m/s)

20
0
X (m)

15

10 –0.5 1
0
0 –1
5 –0.2
–2
–1 –3
–0.4 0 0.2 0.4 0.6
0
0 1 2
–5 0 10 20 30 40 50 60 70 80 90
0 10 20 30 40 50 60 70 80 90 100
Time (s)
Time (s)
Estimated trajectory
Desired trajectory
Real trajectory
Estimated trajectory
Real trajectory Figure 13: State estimation of linear speed.
Figure 11: Linear displacement x.

0.4

0.1 0.2
Angular speed (rad/s)

0.05
0
Pitch angle (rad)

0
–0.2

–0.05 2
–0.4 0
0.4
0.2 –2
–0.1 0 –4
–0.6
–0.2 0 0.2 0.4 0.6 0.8 1
–0.15 0 0.5 1 0 10 20 30 40 50 60 70 80 90
0 10 20 30 40 50 60 70 80 90 Time (s)
Time (s)
Estimated trajectory
Desired trajectory Real trajectory
Estimated trajectory
Real trajectory Figure 14: State estimation of pitch angular speed.

Figure 12: Response of the pitch angle ψ according to terrain

inclination.
The robot parameters are summarized in Table 1.
underwater vehicles. Terminal sliding mode control may be The initial conditions for the position states are fixed as
investigated in future works for the two-wheeled self-bal- x � 0 m; ψ � 0.5 rad; and δ � 0 rad, and all the speed states
ancing robot to design a more efficient tracking control are fixed to zero (x_ � 0 m/s, ψ_ � 0 rad/s, and δ_ � 0 rad/s).
scheme with improved convergence properties. The disturbance vector d(t) is chosen such that
d1 (t) � − 0.3 sin(3πt)Nm, d2 (t) � − 0.02 sin(2πt + π)Nm,
and d3 (t) � − 0.2 sin(5πt)Nm.
4. Numerical Simulations Firstly, we apply the adaptive observer (7) to estimate
unmeasured states and the terrain inclination with the
􏽢 (0) � 0.1 m, ψ􏽢 (0) � 0.24 rad,
􏽢_ 􏽢_
In this section, the numerical simulations carried out using following initial condition: x
Matlab/Simulink software, applied on the two-wheeled 􏽢
δ(0) 􏽢_
� 0.5 rad, x(0) � 0.5 m/s, ψ(0) � 0.5 rad/s, δ(0) �
self-balancing robot, are performed to highlight the 0 rad/s, and α􏽢 (0) � 0 rad.
effectiveness of the proposed adaptive sliding mode The design parameters of the adaptive observer are se-
controller. lected as θ � 10, P(0) � 1 and Y(0) � [0 0 0 0 0 0]T .
12 Mathematical Problems in Engineering

0.5 10

0.4 8

0.3 6

0.2 4
Angular speed (rad/s)

Cpsi (103·Nm)
0.1 2

0 0

–0.1 –2

–0.2 –4
0
–6
–0.3
–1
–8
–0.4 –2
0 0.5 1 1.5 2
–10
10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70
Time (s) Time (s)

Estimated trajectory Cpsi

Real trajectory
Figure 17: The torque Cψ26 under the adaptive sliding mode
Figure 15: State estimation of yaw angular speed. controller in the presence of a slope.

1 1
⎡⎢⎢⎢ 2􏽥 x, 􏽢ρ) 2􏽥 x, ρ􏽢) ⎤⎥⎥⎥
⎢⎢⎢ g11 (􏽢 ⎥⎥⎥
0
+ ⎢ ⎥⎥⎥
0.35 g21 (􏽢
􏽥 x, t􏽢ρ)] � ⎢⎢ ⎢
⎢ ⎥⎥⎥,
⎢⎢⎢ ⎥⎥⎥
0.3 [G(􏽢 (45)
⎢⎣ 1 ⎥⎦
x, ρ􏽢)
􏽥 32 (􏽢
0.25 0 0
Angle of terrain inclination (rad)

g
0.2

0.15 where xr (t), ψ eq (t), and δr (t) are the desired reference
trajectories suitably planned as depicted in Figures 2–4.
0.1
The design parameters are chosen as follows: ε � 0.1,
0.05 k � 16, λ � 10, and β � 10.
0 The simulation results are illustrated through
0.1 Figures 2–9.
–0.05
0.05 The tracking problem studied in this paper consists to
–0.1 0 maintain the pitch angle equilibrium even in the presence of
–0.15
0 0.5 1 a slope and disturbances and to impose the linear dis-
0 10 20 30 40 50 60 70 80 90 100 placement to track a reference trajectory. Figures 2–4 attest
Time (s) that the tracking objective of displacement is achieved.
Estimation of terrain inclination Indeed, the two-wheeled self-balancing robot tracks well
Terrain inclination the desired trajectories (Figures 2 and 4) although keeping
the equilibrium of the pitch angle despite the presence of a
Figure 16: Estimation of terrain inclination α under the proposed
observer. slope (Figure 3).
Figure 5 shows that the linear speed traces a trapezoidal
profile. Figures 6 and 7 reveal a right harmony between
− 1 angular velocities and their estimates.
The design matrix S CT is chosen such that The performance of the proposed adaptive observer to

⎢ 0 2 0 ⎥⎥⎤⎥⎥
⎢
⎡
2 0 0
⎢
estimate the angle of terrain inclination is outstanding from
⎢
⎢ ⎥
⎢
⎢ 0 0 2 ⎥⎥⎥⎥⎥
Figure 8. The control input vector u is on view in Figure 9
⎢
⎢
S CT � ⎢ ⎢ ⎥.
⎢
⎢ 1 0 0 ⎥⎥⎥⎥⎥
− 1 where small oscillations are recorded. It may be observed in
⎢
⎢
⎢
⎢ 0 1 0 ⎥⎥⎦⎥
⎢ Figure 9 that the couple Cδ is increasing after the instant 20 s
⎢
⎣ which corresponds to the beginning of terrain inclination.
0 0 1 Thus, the wheeled robot continues to follow its reference
Once the terrain inclination and the unmeasured states trajectory despite the presence of the terrain inclination.
are available, we apply the continuously implemented In a more sophisticated and realistic scenario, we have
sliding mode controller given by (21) to the two-wheeled added a variation of the yaw angle as it is represented in
self-balancing robot, where S(t) is defined by equation (19) Figure 10. We have also added a white noise ω(t) of power
􏽥 x, t􏽢ρ)]+ is computed as follows:
such that pseudoinverse [G(􏽢 equal to 10− 9 . In addition, we have modified the external
Mathematical Problems in Engineering 13

0.6 sliding mode in order to improve the convergence prop-

0.5 erties of the controller. Also, we will focus on the problem
of observer based-fault tolerant control for the considered
0.4
system.
0.3

Data Availability
Cdelta (103·Nm)

0.2

0.1
The data used to support the findings of this study are in-
0 cluded within the article.
–0.1

–0.2 Conflicts of Interest

–0.3 The authors declare that they have no conflicts of interest.
–0.4
0 10 20 30 40 50 60 70
Time (s)
Acknowledgments
Cdelta This work was supported by the University of Monastir, the
National Engineering School of Monastir, and the Uni-
Figure 18: The torque Cδ26 under the adaptive sliding mode
versity of Sousse.
controller in the presence of a slope.

References
disturbances as d1 (t) � − 0.2 sin(3t)Nm, d2 (t) � − 0.2 cos
(2t)Nm, and d3 (t) � − 0.3 sin(5t)Nm. [1] Y. Ning, M. Yue, and Z. Lin, “Time-optimal control of
underactuated wheeled inverted pendulum vehicles along
The simulation results are illustrated through
specified paths,” in Proceedings of the 2020 5th International
Figures 11–18. Conference on Advanced Robotics and Mechatronics (ICARM),
Figures 10–12 attest that the tracking objective of dis- Shenzhen, China, 2020.
placement is obtained. Indeed, the two-wheeled self-bal- [2] S. Kim and S. Kwon, “Robust transition control of under-
ancing robot tracks well the desired trajectories (Figures 10 actuated two-wheeled self-balancing vehicle with semi-online
and 11) although keeping the equilibrium of the pitch angle dynamic trajectory planning,” Mechatronics, vol. 68, Article
despite the presence of a slope (Figure 12). Figure 13 shows ID 102366, 2020.
that the linear speed traces a trapezoidal profile. Figures 14 [3] C. Acar and T. Murakami, “Underactuated two-wheeled
and 15 reveal a right harmony between angular velocities mobile manipulator control using nonlinear backstepping
and their estimates. method,” in Proceedings of the 34th Annual Conference of
The performance of the proposed adaptive observer to IEEE Industrial Electronics, Orlando, FL, USA, November
2008.
estimate the angle of terrain inclination is shown from [4] C.-H. Chiu, Y.-F. Peng, and Y.-W. Lin, “Intelligent back-
Figure 16. Figure 17 represents the torque Cψ26 which en- stepping control for wheeled inverted pendulum,” Expert
sures the translation along the x-axis while preserving the Systems with Applications, vol. 38, no. 4, pp. 3364–3371, 2011.
upright position. It seems clear from Figure 18 that the [5] A. Benaskeur and A. Desbiens, “Application of adaptive
couple Cδ26 increases at the instant 45 s which corresponds backstepping to the stabilization of the inverted pendulum,”
to the beginning of the yaw angle variation. Thus, the in Proceedings of the IEEE Canadian Conference on Electrical
wheeled robot continues to follow its reference trajectory and Computer Engineering, Waterloo, Canada, 1998.
despite the presence of the terrain inclination and the [6] T. Nomura, Y. Kitsuka, H. Suemitsu, and T. Matsuo,
presence of disturbances and measurement noise. “Adaptive backstepping control for a two-wheeled autono-
mous robot,” in Proceedings of the 2009 ICCAS-SICE,
Fukuoka, Japan, 2009.
5. Conclusions [7] R. Cui, J. Guo, and Z. Mao, “Adaptive backstepping control of
wheeled inverted pendulums models,” Nonlinear Dynamics,
In this paper, a robust adaptive observer-based sliding vol. 79, no. 1, pp. 501–511, 2015.
mode control has been proposed for the two-wheeled self- [8] N. T. Binh, N. M. Hung, N. A. Tung, D. P. Nam, and
balancing robot subject to terrain inclination and distur- N. T. Long, “Robust H-infinity backstepping control design of
bances. The application of the adaptive observer guarantees a wheeled inverted pendulum system,” in Proceedings of the
the simultaneous estimation of unmeasured states and the 2017 International Conference on System Science and
terrain inclination angle which is assumed unknown. The Engineering, Ho Chi Minh City, Vietnam, 2017.
[9] G. Rigatos, K. Busawon, J. Pomares, and M. Abbaszadeh,
convergence of the proposed controller was illustrated
“Nonlinear optimal control for the wheeled inverted pen-
through a Lyapunov analysis and inspired from sliding dulum system,” Robotica, vol. 38, no. 1, pp. 29–47, 2019.
modes theory. Numerical simulations emphasize the per- [10] C. H. Huang, W. J. Wang, and C. H. Chiu, “Design and
formance of the designed control method applied to the implementation of fuzzy control on a two-wheel inverted
two-wheeled self-balancing robot. In future work, we will pendulum,” IEEE Transaction on Industrial Electronics,
enhance our control algorithm by using the terminal vol. 58, no. 7, pp. 2988–3001, 2010.
14 Mathematical Problems in Engineering

[11] T. Zhao, Q. Yu, S. Dian, R. Guo, and S. Li, “Non-singleton tracking smooth curved welding path,” KSME International
general type-2 fuzzy control for a two-wheeled self-balancing Journal, vol. 18, no. 7, pp. 1094–1106, 2004.
robot,” International Journal of Fuzzy Systems, vol. 21, no. 6, [28] J. Huang, Z. H. Guan, T. Matsuno, T. Fukuda, and K. Sekiyama,
pp. 1724–1737, 2019. “Sliding-mode velocity control of mobile-wheeled inverted-
[12] Z. Li, “Adaptive fuzzy output feedback motion/force control pendulum systems,” IEEE Transactions on Robotics, vol. 26,
for wheeled inverted pendulums,” IET Control Theory & no. 4, pp. 750–758, 2010.
Applications, vol. 5, no. 10, pp. 1176–1188, 2011. [29] J. Huang, Z. Guan, T. Matsuno, T. Fukuda, and K. Sekiyama,
[13] Y. J. Liu, S. Tong, and C. P. Chen, “Adaptive fuzzy control via “High-order disturbance observer based sliding mode control
observer design for uncertain nonlinear systems with for mobile wheeled inverted pendulum systems,” IEEE
unmodeled dynamics,” IEEE Transactions on Fuzzy Systems, Transactions on Industrial Electronics, vol. 67, no. 3,
vol. 21, no. 2, pp. 275–288, 2012. pp. 2030–2041, 2019.
[14] C. Iwendi, M. A. Alqarni, J. H. Anajemba, A. S. Alfakeeh, [30] D. S. Nasrallah, H. Michalska, and J. Angeles, “Controllability
Z. Zhang, and A. K. Bashir, “Robust navigational control of a and posture control of a wheeled pendulum moving on an
two-wheeled self-balancing robot in a sensed environment,” inclined plane,” IEEE Transactions on Robotics, vol. 23, no. 3,
IEEE Access, vol. 7, pp. 82337–82348, 2019. pp. 564–577, 2007.
[15] Z. Music, F. Molinari, S. Gallenmüller et al., “Design of a [31] F. Dai F, X. Gao, S. Jiang, Y. Liu, and J. Li, “A multi-DOF two
networked controller for a two-wheeled inverted pendulum wheeled inverted pendulum robot climbing on a slope,” in
robot,” IFAC-PapersOnLine, vol. 52, no. 20, pp. 169–174, 2019. Proceedings of the IEEE International Conference on Robotics
[16] L. Guo, S. A. A. Rizvi, and Z. Lin, “Optimal control of a and Biomimetics, vol. 201, Bali, Indonesia, December 2014.
two-wheeled self-balancing robot by reinforcement [32] K. Hirata, M. Kamatani, and T. Murakami, “Advanced mo-
learning,” International Journal of Robust and Nonlinear tion control of two-wheel wheelchair for slope environment,”
Control, 2020. in Proceedings of the IECON 2013-39th Annual Conference of
[17] L. Qiao and W. Zhang, “Trajectory tracking control of AUVs the IEEE Industrial Electronics Society, Vienna, Austria, 2013.
via adaptive fast nonsingular integral terminal sliding mode [33] M. S. Key, C. G. Jeon, and D. S. Yoo, “Sliding mode control for
control,” IEEE Transactions on Industrial Informatics, vol. 16, a two-wheeled inverted pendulum mobile robot driving on
no. 2, pp. 1248–1258, 2019. uniform slopes,” in Proceedings of the 12th International
[18] W. Wang, X. D. Liu, and J. Q. Yi, “Structure design of two Conference on Control, Automation and Systems, Jeju Island,
types of sliding-mode controllers for a class of under-actuated Republic of Korea, 2012.
[34] Z. Kausar, K. Stol, and N. Patel, “The effect of terrain
mechanical systems,” IET Control Theory & Applications,
inclination on performance and the stability region of
vol. 1, no. 1, pp. 163–172, 2007.
two-wheeled mobile robots,” International Journal of
[19] S. Mahjoub, F. Mnif, and N. Derbel, “Second-order sliding
Advanced Robotic Systems, vol. 9, no. 5, p. 218, 2012.
mode approaches for the control of a class of underactuated
[35] T. Takei, O. Matsumoto, and K. Komoriya, “Simultaneous
systems,” International Journal of Automation and Comput-
estimation of slope angle and handling force when getting on
ing, vol. 12, no. 2, pp. 134–141, 2015.
and off a human-riding wheeled inverted pendulum vehicle,”
[20] M. Yue, W. Sun, and P. Hu, “Sliding mode robust control for
in Proceedings of the IEEE/RSJ International Conference on
two-wheeled mobile robot with lower center of gravity,”
Intelligent Robots and Systems, St. Louis, MO, USA, 2009.
International Journal of Innovative Computing, Information [36] D. Liang, N. Sun, Y. Wu, and Y. Fang, “Differential flatness-based
and Control, vol. 7, no. 2, pp. 637–646, 2011. robust control of self-balanced robots,” IFAC-PapersOnLine,
[21] H. T. Yau, C. C. Wang, N. S. Pai, and M. J. Jang, “Robust vol. 51, no. 31, pp. 949–954, 2018.
control method applied in self-balancing two-wheeled robot,” [37] Q. Zhang, “Adaptive observer for multiple-input-multiple-output
in Proceedings of the 2nd International Symposium on (MIMO) linear time-varying systems,” IEEE Transactions on
Knowledge Acquisition and Modeling, Wuhan, China, 2009. Automatic Control, vol. 47, no. 3, pp. 525–529, 2002.
[22] N. K. Goswami and P. K. Padhy, “Sliding mode controller [38] A. Xu and Q. Zhang, “State and parameter estimation for
design for trajectory tracking of a non-holonomic mobile nonlinear systems,” in Proceedings of the 2002 IFAC World
robot with disturbance,” Computers & Electrical Engineering, Congress, vol. 35, no. 1, Barcelona, Spain, 2002.
vol. 72, pp. 307–323, 2018. [39] G. Besançon, “Remarks on nonlinear adaptive observer de-
[23] S. Mobayen, “Design of LMI-based sliding mode controller sign,” Systems & Control Letters, vol. 41, no. 4, pp. 271–280,
with an exponential policy for a class of underactuated sys- 2000.
tems,” Complexity, vol. 21, no. 5, pp. 117–124, 2019. [40] M. Farza, M. M’Saad, T. Maatoug, and M. Kamoun, “Adaptive
[24] H. Mirzaeinejad, “Optimization-based nonlinear control observers for nonlinearly parameterized class of nonlinear
laws with increased robustness for trajectory tracking of systems,” Automatica, vol. 45, no. 10, pp. 2292–2299, 2009.
non-holonomic wheeled mobile robots,” Transportation [41] I. Jmel, H. Dimassi, S. Hadj-Said, and F. M’Sahli, “An adaptive
Research Part C: Emerging Technologies, vol. 101, pp. 1–17, sliding mode observer for inverted pendulum under mass
2019. variation and disturbances with experimental validation,” ISA
[25] S. Mobayen, M. J. Yazdanpanah, and V. J. Majd, “A finite-time Transactions, vol. 102, pp. 264–279, 2020.
tracker for nonholonomic systems using recursive singularity-free [42] F. Grasser, A. D’Arrigo, S. Colombi, and A. C. Rufer, “JOE: a
FTSM,” in Proceedings of the 2011 IEEE American Control mobile, inverted pendulum,” IEEE Transactions on Industrial
Conference, San Francisco, CA, USA, 2011. Electronics, vol. 49, no. 1, pp. 107–114, 2002.
[26] F. Dai, X. Gao, S. Jiang, W. Guo, and Y. Liu, “A two-wheeled [43] M. Chen, “Robust tracking control for self-balancing mobile
inverted pendulum robot with friction compensation,” robots using disturbance observer,” IEEE/CAA Journal of
Mechatronics, vol. 30, pp. 116–125, 2015. Automatica Sinica, vol. 4, no. 3, pp. 458–465, 2017.
[27] T. L. Chung, T. H. Bui, T. T. Nguyen, and S. B. Kim, “Sliding [44] D. Liang, N. Sun, Y. Wu, and Y. Fang, “Modeling and motion
mode control of two-wheeled welding mobile robot for control of self-balance robots on the slope,” in Proceedings of
Mathematical Problems in Engineering 15

the 31st Youth Academic Annual Conference of Chinese As-

sociation of Automation, Wuhan, China, 2016.
[45] H. Dimassi, S. Hadj-Said, A. Loria, and F. M’Sahli, “An
adaptive observer for a class of nonlinear systems with a high
gain approach. Application to the twin-rotor system,” In-
ternational Journal of Control, 2019.
[46] J. J. E. Slotine and W. Li, Applied Nonlinear Control, Prentice
Hall, Englewood Cliffs, NJ, USA, 1991.
[47] K. Kalsi, J. Lian, S. Hui, and S. H. Żak, “Sliding-mode ob-
servers for systems with unknown inputs: a high-gain ap-
proach,” Automatica, vol. 46, no. 2, pp. 347–353, 2010.