sensors

Article
Robust Tracking Control of Wheeled Mobile Robot Based on
Differential Flatness and Sliding Active Disturbance
Rejection Control: Simulations and Experiments
Amine Abadi 1, * , Amani Ayeb 2 , Moussa Labbadi 3 , David Fofi 1 , Toufik Bakir 1 and Hassen Mekki 4

1 Laboratory ImViA EA 7535, University of Bourgogne, 21000 Dijon, France; [email protected] (D.F.);
[email protected] (T.B.)
2 National Institute of Applied Science and Technology, Physics and Instrumentation Department,
Tunis 1080, Tunisia; [email protected]
3 LIS UMR CNRS 7020, Aix-Marseille University, 13013 Marseille, France; [email protected]
4 NOCCS Laboratory, National School of Engineering of Sousse, University of Sousse, Sousse 4054, Tunisia;
[email protected]
* Correspondence: [email protected]

Abstract: This paper proposes a robust tracking control method for wheeled mobile robot (WMR)
against uncertainties, including wind disturbances and slipping. Through the application of the
differential flatness methodology, the under-actuated WMR model is transformed into a linear
canonical form, simplifying the design of a stabilizing feedback controller. To handle uncertainties
from wheel slip and wind disturbances, the proposed feedback controller uses sliding mode control
(SMC). However, increased uncertainties lead to chattering in the SMC approach due to higher
control inputs. To mitigate this, a boundary layer around the switching surface is introduced,
implementing a continuous control law to reduce chattering. Although increasing the boundary
layer thickness reduces chattering, it may compromise the robustness achieved by SMC. To address
this challenge, an active disturbance rejection control (ADRC) is integrated with boundary layer
Citation: Abadi, A.; Ayeb, A.; sliding mode control. ADRC estimates lumped uncertainties via an extended state observer and
Labbadi, M.; Fofi, D.; Bakir, T.; eliminates them within the feedback loop. This combined feedback control method aims to achieve
Mekki, H. Robust Tracking Control practical control and robust tracking performance. Stability properties of the closed-loop system are
of Wheeled Mobile Robot Based on established using the Lyapunov theory. Finally, simulations and experimental results are conducted to
Differential Flatness and Sliding compare and evaluate the efficiency of the proposed robust tracking controller against other existing
Active Disturbance Rejection Control: control methods.
Simulations and Experiments. Sensors
2024, 24, 2849. https://doi.org/
Keywords: differential flatness; sliding mode control; active disturbance rejection control; extended
10.3390/s24092849
state observer; wheeled mobile robot
Academic Editors: David Cheneler and
Stephen Monk

Received: 22 March 2024

1. Introduction
Revised: 17 April 2024
Accepted: 25 April 2024 The domain of robotics finds mobile robots to be particularly intriguing, attracting
Published: 29 April 2024 considerable fascination and study. Designed to operate in dynamic settings, be it indoors
or outdoors, these robots demonstrate the capacity to navigate autonomously or with
minimal human input. Central to their functionality is their mobility, achieved through
diverse locomotion methods, such as wheels, tracks, or legs. This mobility empowers
Copyright: © 2024 by the authors. them to traverse diverse terrains, overcoming obstacles encountered during their journeys.
Licensee MDPI, Basel, Switzerland. Recently, mobile robots have been used in various domains, including civilian, industrial,
This article is an open access article and military, to carry out diverse tasks such as surveillance [1], transportation [2], agri-
distributed under the terms and cultural operations [3], and exploration [4]. Given the broad application spectrum and
conditions of the Creative Commons
critical nature of tasks involving mobile robots, there exists an urgent need to develop
Attribution (CC BY) license (https://
performance tracking controllers to execute proposed missions with exceptional accuracy.
creativecommons.org/licenses/by/
However, achieving this objective remains a significant challenge due to the inherent
4.0/).

Sensors 2024, 24, 2849. https://doi.org/10.3390/s24092849 https://www.mdpi.com/journal/sensors

Sensors 2024, 24, 2849 2 of 24

under-actuation and nonlinearity in WMRs, constrained by nonholonomic limitations.

Consequently, researchers have directed their efforts towards investigating the control of
mobile robotic systems.
In the past few decades, substantial progress has been made in the field of tracking
control for wheeled mobile robots (WMR) through the application of nonlinear control
theory [5–7]. Among these control methodologies, linearization controllers, such as the
flatness controller [8], have risen as a popular approach that can significantly simplify
the controller design process. The flatness property is a technique used to define the
dynamic behavior of nonlinear underactuated models by identifying a set of core system
variables known as flat outputs. This perspective has significant implications for control
systems, as will be demonstrated. The first step in flatness control involves generating
a desired realizable trajectory that implicitly incorporates the system model. Following
that, the nonlinear WMR model can be linearized, resulting in the canonical Brunovsky
form [9,10]. This special form simplifies the concept of a feedback controller capable
of achieving exact trajectory tracking. In fact, controlling a linear system is easier than
controlling an underactuated nonlinear system, and this feature has encouraged researchers
to use the properties of flatness in several application domains, such as the control of
hydraulic systems [11], exoskeleton robots [12], microgrid [13], underwater robot [14],
and quadrotor [15,16].
Numerous research studies on WMR have utilized the concept of flatness control.
Abadi [17] introduced an approach for optimal path planning for WMR using the collo-
cation method, flatness control, and spline curves. This method effectively reduces the
time needed to compute optimal robot trajectories during navigation, which is crucial
for real-world applications. Kaniche [18] proposed a flatness visual servoing control for
WMR subjected to disturbances. Salah [19] developed an approach to generate the upper
coverage trajectory of a mobile robot by leveraging flatness. Yakovlev [20] combined flat-
ness control with predictive control to enable safe navigation of a WMR among static and
dynamic obstacles.
There is always is a difference between the mathematical model describing the move-
ment of WMR and reality. This difference is due to environmental phenomena neglected
during modeling, such as wind, slipping, etc. The question that arises is how flatness
control applied to WMR can ensure the accurate tracking of a desired trajectory despite the
presence of uncertainties. To resolve this problem, a robust feedback controller must be
combined with flatness, taking into account the impact of uncertainties to the model. Up to
the present, there have been limited methods in the literature concerning the robustness
issues of flatness systems. Among these approaches, the sliding mode control (SMC) has
been successfully utilized in a variety of systems [21–23].
SMC is a robust control technique used to manage dynamic systems in the presence of
uncertainties and disturbances. At its core, SMC aims to drive the system state onto a des-
ignated sliding surface within the state space. Once on this surface, the system’s behavior
is constrained, allowing for effective regulation. SMC achieves this through discontinuous
control actions, known as switching control, which dynamically alternate between different
control laws. This switching mechanism ensures that the system remains on the sliding
surface, enhancing robustness against external influences. Despite its effectiveness, SMC
is associated with a phenomenon called chattering [24], characterized by rapid switching
between control actions near the sliding surface. While chattering can theoretically improve
tracking accuracy, it can lead to practical issues such as mechanical wear and high-frequency
oscillations. To resolve this problem, numerous approaches have been suggested in the
existing literature, such as high-order SMC [25], boundary layer [26], and active adaptive
continuous nonsingular terminal sliding mode algorithm [27]. A frequently utilized ap-
proach for mitigating the chattering phenomenon involves incorporating the boundary
layer technique within SMC. This entails replacing the sign function with a smooth function.
However, this strategy presents its own set of challenges. Firstly, there exists a trade-off
between the size of the boundary layer and the performance of SMC, which impacts the
Sensors 2024, 24, 2849 3 of 24

effectiveness of chattering reduction. Secondly, the robustness and accuracy of the system
may not always be guaranteed within the boundary layer. Additionally, beyond addressing
the chattering issue, achieving precise control of a robotic system necessitates knowledge
about its state, typically obtained through real instruments, which can incur high costs
and complicate the system’s structure. Moreover, in many instances, directly measuring
certain system parameters may be impractical. To overcome these limitations, one potential
solution involves implementing software sensors or observers, commonly referred to as vir-
tual sensors. Therefore, to tackle both the reduced robustness resulting from the boundary
layer approach and the challenge of state estimation, we propose a novel robust feedback
controller that integrates the boundary layer sliding method with a disturbance observer.
In recent times, disturbance observers have emerged as potent tools for handling
consolidated uncertainties, closely tied to disturbance-observer-based control. Within the
domain of nonlinear disturbance observers, two notable approaches stand out: the uncer-
tainty and disturbance estimator (UDE) [28] and the active disturbance rejection control [29]
based on the extended state observer (ESO). In the UDE, only the disturbance is estimated,
though in general, the observer equations depend on system states and inputs. Thus, a state
observer is necessary unless all states are measurable. The idea of the ESO is to extend the
original state vector by the disturbance vector and possibly, some of its time-derivatives,
and then design a state observer for the extended system. ESO distinguishes itself by incor-
porating a dynamic model of disturbances or uncertainties into its estimation methodology,
enabling it to identify and mitigate uncertainties not explicitly accounted for in the system
model. The design of ESO is distinguished by its minimal dependence on system data
and its freedom from the traditional system model, which simplifies its implementation
process. Furthermore, several other types of disturbance observers are available, such as
the nonlinear extended state observer (NLESO) [30], the adaptive extended state observer
(AESO) [31], and the extended high-gain observer (EHGO) [32]. EHGO, part of the ESO
family, stands out in two key aspects: it does not necessitate slow variations in disturbances,
and it estimates a matched disturbance term originating from model uncertainty and exter-
nal disturbances. Given the advantages offered by ESO, considerable research effort has
been devoted to developing advanced controls for robotic systems.
In Ref. [33], Xie introduced a controller that integrates the backstepping technique
with ESO to improve tracking performance for underwater robots. Additionally, Qi [34]
improved the bandwidth of ESO to achieve more accurate disturbance estimation. Subse-
quently, they utilized a simple feedback controller to ensure attitude stabilization over a 3D
hovering quadrotor system. In the work by Aole [35], an improved ADRC methodology
for controlling lower limb exoskeletons is presented. The proposed approach integrates
Linear ESO with a tracking differentiator, nonlinear state error feedback, and a proportional
controller. Simulation results demonstrate the effectiveness of the suggested ADRC in effi-
ciently regulating the hip and knee movements of the robot in the presence of disturbances.
Hu [36] integrated a predictive control technique with ESO for unmanned underwater
vehicles, offering a solution to concurrently handle external disturbances and system mea-
surement noises. Based on this observation, the main contributions of our research can be
summarized as follows:
1. The kinematic model for WMR is structured in a standard format that systematically
tackles underactuation and transforms nonmatching disturbances into matching ones
through a flatness-based approach;
2. The designated trajectory is feasible in practice because of the concept of differential
flatness, which equates differential flatness with controllability, ensuring its physi-
cal achievability;
3. Continuous sliding mode control (SMC) is employed to eliminate chattering, an es-
sential necessity for the efficient application of control in real-world scenarios;
4. SMC is integrated with ESO for the uncertain kinematic WMR model. This strategy
seeks to improve the practicality and resilience of the tracking controller by reducing
Sensors 2024, 24, 2849 4 of 24

chattering through boundary layer SMC and estimating the lumped disturbance
affecting the WMRs via ESO, which is then employed as a feedforward compensation;
5. The proposed control method was compared with several other control methods,
including traditional flatness control, backstepping tracking control flatness-based
sliding control, and flatness active disturbance rejection control and backstepping slid-
ing active disturbance rejection control. These comparisons were validated through
simulations conducted in Matlab/Simulink and experiments carried out on the Turtle-
Bot WMR.
The structure of the remaining sections of this article is outlined as follows. Section 2
provides a thorough overview of the flatness control technique for WMR. Section 3 elabo-
rates on the concept of flatness-based sliding mode tracking control of WMRs. The proposed
robust tracking controller is delineated in Section 4. Sections 5 and 6 present and discuss
the results of simulations and experiments. Finally, Section 7 concludes the paper by
summarizing the key findings and suggesting potential future directions.

2. Flatness-Based Tracking Control

In our study, we analyzed a differential two-wheeled mobile robot (see Figure 1) that
consists of two independent active wheels and a third passive wheel (a standard freewheel).
This robotic system is widely regarded as an effective trade-off between control ease and the
degrees of freedom that enable the robot to meet mobility requirements. The configuration
of the mobile robot with wheels can be described by the vector qr = [ x, y, θ ]. In this notation,
x and y represent the coordinates of the robot’s center position in the stationary frame
(O, X, Y ), while θ represents the orientation angle of the robot. The state equation of the
WMR kinematic model, neglecting uncertainties, is represented as follows:

ẋ = cos(θ )v
ẏ = sin(θ )v (1)
θ̇ = w

Figure 1. Two-wheeled mobile robot.

The robot’s translational and rotational velocities are denoted by v and w, respectively.
The angular velocities of the right and left wheels (wr and wl ) can be defined as functions
of the robot’s translational and rotational velocities as follows:
wr + w l
v=( )r (2)
2
Sensors 2024, 24, 2849 5 of 24

wr − w l
w=( )r (3)
2b
where the variables r and 2b represent the radius and distance between the wheels, respec-
tively. The nonholonomic limitation is defined as follows, based on the nonslip requirement:

ẋsinθ − ẏcosθ = 0 (4)

The accuracy of the tracking will be guaranteed through the flatness property, which
involves describing all system states and inputs, as well as their finite time derivatives,
within the framework of a flat output. Considering the following nonlinear system:

ẋ = f ( x, u) (5)

where x ∈ Rn and u ∈ Rm represent the state and the input vector.

The nonlinear system (5) is differentially flat if there exists an output λ in the
following form:
λ = ξ ( x, u, u̇, . . . , u(c) ) ∈ Rm (6)
such that the state and the input can be expressed as follows:

x = κ1 (λ, λ̇, λ̈ . . . , λ(a) ) (7)

( a +1)
u = κ2 (λ, λ̇, λ̈ . . . , λ ) (8)

where a and c are finite multi-indices, and ξ, κ1 , and κ2 are smooth vector functions of
the output vector λ and its derivatives. By introducing the functions κ1 and κ2 , this flat
output is composed of a set of variables that enable the parameterization of all other system
variables: the state, the command, and also the output λ. Indeed, if the output of the system
is defined by a relation of the form λ = Ξ( x, u, u̇, . . . , u( p) ), then necessarily, the quantities
described in Equations (7) and (8) make it possible to affirm that there exists an integer c
such that:
f = Ξ(λ, λ̇, λ̈ . . . , λ(c) ) (9)
The flat output combines all unconstrained variables of the system since the compo-
nents of λ are differentially independent. Alternatively, based on Equation (9), we can argue
that the flat output λ solely relies on the state and the command. This would make it an
endogenous variable of the system, in contrast to the state of an observer, which would be
an example of an exogenous variable of the observed system. In addition, Lie–Bäcklund’s
notion of differential equivalence [8] shows that the number of components of λ is the same
as the number of components of the control:

dimλ = dimu (10)

This fundamental characteristic allows us to determine the requisite number of inde-

pendent variables needed in a model to establish its flatness. A key benefit of the flatness
property lies in its facilitation of various transformations, such as diffeomorphism and
feedback linearization. These transformations enable the conversion of a nonlinear system
into a controllable linear system, where the flat outputs represent the state vector.
Several studies in the literature, including Ref. [37], have shown that the WMR kine-
matic modeling can be defined as a differentially flat model, where the positional coordi-
nates denoted as λ = [λ11 , λ21 ] T = [ x, y] T serve as the flat outputs. Therefore, the entire set
of state and control components pertaining to the WMR system are expressed using the flat
variable λ and its derivatives, as demonstrated below:

λ̇21
θ = arctan (11)
λ̇11
Sensors 2024, 24, 2849 6 of 24

q
v= λ̇211 + λ̇221 (12)

λ̇11 λ̈21 − λ̈11 λ̇21

w= (13)
λ̇211 + λ̇221
The differentially flat nature of the WMR’s kinematic model has been demonstrated in
the literature by various researchers [37]. This implies that all the states and controls of the
kinematic WMR model can be expressed as functions of λ and its derivatives. However,
the noninvertible relationship between the control input vectors w and v and the highest
derivatives of the flat output limits the development of static feedback linearization for
the nonlinear WMR. To address this constraint, we incorporate the control input v into the
kinematic model defined by Equation (1) by treating it as an additional state. As a result,
we obtain a revised system that can be defined as follows:
ẋ = cos(θ )v
ẏ = sin(θ )v
(14)
v̇ = ur1
θ̇ = ur2

The state and control inputs of the modified system defined by Equation (14) are
represented by Xr = [ x, y, v, θ ] T and ur1 = v̇ and ur2 = w. In order to establish a bijective
relationship between the inputs ur1 , ur2 , and higher-order derivatives of λ11 = x, λ21 = y,
we apply successive differentiations to the flat outputs until at least one of the input
variables appears in the resulting expressions, as illustrated below:
   
λ̈11 u
= Brob r1 (15)
λ̈21 ur2

where Brob is described as follows:

 
cos(θ ) −vsin(θ )
Brob = (16)
sin(θ ) vcos(θ )

The matrix Brob is not singular if v ̸= 0. In this case, we can define the control
as follows:    
ur1 −1 λ̈11
= Brob (17)
ur2 λ̈21
To arrive at the linearized system, referred to as the Burnovsky Form (BF), we can
substitute the control input (17) into Equation (15). This substitution yields the following
modified expression:
 
λ̇11 = λ12
 λ̇21 = λ22

( BF1 ) λ̇12 = v1 ( BF2 ) λ̇22 = v2 (18)
 
Y1 = λ11 = x Y2 = λ21 = y
 

where v1 and v2 represent a suitable feedback controller defined as follows:

v1 = λ̈ xd − σx2 (λ12 − λ̇ xd ) − σx1 (λ11 − λ xd ) (19)

v2 = λ̈yd − σy2 (λ22 − λ̇yd ) − σy1 (λ21 − λyd ) (20)

where λ xd and λyd denote the desired trajectories for the flat output λ11 and λ21 , respectively.
Meanwhile, the controller gains are represented by σx1 , σx2 , σy1 , and σy2 . The polynomial
of the Burnovsky system (18) can be defined as follows:

s2 + σx2 s + σx1 = s2 + 2m x ϵxc + ϵ2xc (21)

Sensors 2024, 24, 2849 7 of 24

2
s2 + σy2 s + σy1 = s2 + 2my ϵyc + ϵyc (22)
where the parameters m x and my are the damping coefficients, and ϵxc and ϵyc are the
frequencies in Equations (21) and (22). We can calculate the controller gain as follows:

σx1 = ϵ2xc , σx2 = 2m x ϵxc , σy1 = ϵyc

2
, σy2 = 2my ϵyc (23)

By integrating the feedback law, as described in Equations (19) and (20), into the
system (17), we can express the flatness-based tracking control (FBTC) utilized for the
mobile robot in the following manner:
   
u FBTCx −1 λ̈ xd − σx2 ė1 − σx1 e1
= Brob (24)
u FBTCy λ̈yd − σy2 ė2 − σy1 e2

where e1 = λ11 − λ xd and e2 = λ21 − λyd .

In ideal conditions where uncertainties such as wind and wheel slip are negligible in
the kinematic model of the WMR, the control input defined by Equation (24) can achieve
satisfactory tracking performance for the desired trajectory. However, it is practically
impossible to have a model that accurately represents the real-world movement of the robot
in all environmental conditions. As a result, the following section will focus on developing
a robust tracking control for a WMR kinematic model that is subject to uncertainties.

3. Flatness-Based Sliding Tracking Control

In order to account for real-world conditions, we consider uncertainties such as
slippage and external environmental disturbances when describing the kinematic model of
WMR (Figure 2). As a result, the model is defined differently, as shown below.

Figure 2. Two-wheeled mobile robot subject to uncertainties.


 ẋ = cos(θ )v + vt cos(θ ) + vs sin(θ ) + p x

(Uncertain Kinematic Model ) ẏ = sin(θ )v + vt sin(θ ) − vs cos(θ ) + py (25)

θ̇ = w + ws


where the variables p x and py represent the external environmental disturbances, indicating
the potential influences from the surrounding conditions. On the other hand, vt and vs
represent the slip velocities, where vt denotes the slip velocity along the forward direction
and vs represents the slip velocity normal to it. Additionally, ws denotes the angular slip
velocity. According to [37], it is assumed that the slippage phenomenon can be defined and
bounded as follows:
vt ( t ) = vs ( t ) = ws ( t ) = κ1 v ( t ) (26)
Sensors 2024, 24, 2849 8 of 24

||vt || ≤ ε 1 ||v||, ||vs || ≤ ε 2 ||v||, ||ws || ≤ ε 3 (27)

where κ1 , ε 1 , ε 2 and ε 3 are positive constants.
Assuming that λ xd and λyd are the reference trajectories for λ11 and λ21 , respectively,
we can define the error dynamics as ei = λi1 − λid for i = 1, 2. To achieve convergence of
the tracking error ei to zero in the presence of uncertainties, we employ a sliding mode
control approach that relies on the principles of the flatness law. By incorporating this
control strategy, we aim to ensure robust and accurate tracking performance even in the
face of system uncertainties. The design of the sliding mode control involves two essential
stages: the choice of the sliding surface and the development of the control law. These
steps play a crucial role in establishing an effective and stable sliding mode control strategy.
The selection of the sliding surface determines the desired system behavior and convergence
properties, while the design of the control law focuses on generating control signals that
guide the system towards the desired sliding surface and ensure its maintenance on that
surface. In the context of the tracking example for the WMR, we make use of the sliding
variable σr = [s x , sy ] T to represent the tracking error. To define the sliding surface, we
consider the desired tracking behavior and express it as follows, taking into account the
specific requirements of the system:
s x = ė1 + β 1 e1 (28)

sy = ė2 + β 2 e2 (29)
where the gains β 1 and β 2 can be selected using pole-placement techniques to ensure the
asymptotic convergence of the tracking errors e1 = λ11 − λ xd and e2 = λ21 − λyd to zero.
In this tracking example, the sliding variable σr = [s x , sy ] T is chosen as the tracking error.
Therefore, the sliding surface for the WMR can be defined as follows:

ė1 + β 1 e1 = 0 (30)

ė2 + β 2 e2 = 0 (31)
As suggested by Mauledoux [38], to guarantee that the sliding surface σr = 0 is
attractive, we can enforce the dynamics of σr as follows:

σ̇r = −k i sgn(σr ) (32)

where the standard signum function is denoted by sgn, and ki (i = 1, 2) is a constant. One ap-
proach to proving the error dynamics stability is to analyze the following Lyapunov function:

1 2
Vs = σ (33)
2 r
The derivative of Vs is defined as follows:

V̇s = σr σ̇r (34)

We can conclude that Vs is a positive function and its derivative V̇s is negative or
zero. Hence, the system exhibits asymptotic Lyapunov stability. Using Equations (28), (29)
and (32) we obtain:
−k1 sign(s x ) = ë1 + β 1 ė1 (35)
−k2 sign(sy ) = ë2 + β 2 ė2 (36)
As a result, by using Equations (35) and (36), we can obtain:

λ̈11 = λ̈ xd − β 1 ė1 − k1 sgn(s x ) (37)

λ̈21 = λ̈yd − β 2 ė2 − k2 sgn(sy ) (38)

Sensors 2024, 24, 2849 9 of 24

Substituting λ̈11 and λ̈21 with their new expressions defined by Equations (35) and (36)
in the control defined by (17), the flatness-based sliding mode tracking controller (FSMC)
applied to WMR is defined as follows:
   
u FSMCx −1 λ̈ xd − β 1 ė1 − k 1 sgn ( s x )
= Brob (39)
u FSMCy λ̈yd − β 2 ė2 − k2 sgn(sy )

The FSMC defined by Equation (39) contains a discontinuous control term due to the
function sgn(σ). Although selecting sufficiently large values for k1 and k2 can achieve con-
vergence to sliding variable in limited time and provide robustness against perturbations, it
also causes the phenomenon of chattering. Thus, to avoid this problem, the function sgn(σ )
can be replaced by the function Sat defined as follows:
(
σr
if |σr | ≤ as
Sat(σr ) as (40)
sgn(σr ) i f |σr | > as

where as is the width of the threshold of the saturation function.

The thickness of the boundary layer, denoted as as , within the saturation function
stands as a pivotal parameter influencing the efficacy of the sliding mode controller. As the
value of as increases, the approximation diverges more from the ideal sgn function, resulting
in enhanced reduction of chattering. However, this improvement comes at the cost of
diminished robustness. Conversely, if the value of the parameter as is reduced, the change
of the control signal will be too frequent, which leads to inevitable chatter of the control
signal. Therefore, a variable-thickness boundary layer as is tailored to strike a balance
between mitigating chattering and upholding system robustness amid uncertainties. In
the upcoming section, the FSMC described by Equation (39) will be integrated with active
disturbance rejection control to enhance the robustness lost by the Sat function and maintain
the advantage of reducing chattering.

4. Proposed Robust Tracking Controller

In this section, we introduce a novel cascade control strategy that utilizes a combination
of flatness property, active disturbance rejection control (ADRC), and boundary layer
sliding mode control to solve the problem of reduced robustness obtained when replacing
the function sgn by the function sat in the FSMC defined by Equation (39). Given the
uncertain kinematic model (25), we can obtain the following relationship by differentiating
λ11 and λ21 until the input terms u1 and u2 become evident:
     
λ̈11 ur1 u
= Brob + Crob + Drob r1 (41)
λ̈21 ur2 ur2

where Crob and Drob are defined as follows:

   
cos(θ )(vs ws + v̇t ) + sin(θ )(v̇s − vws − vt ws ) + ṗ x 0 −vt sin(θ ) + vs cos(θ )
Crob = , Drob = (42)
sin(θ )(vs ws + v̇t ) − cos(θ )(v̇s − vws − vt ws ) + ṗy 0 vt cos(θ ) + vs sin(θ )

By utilizing the control input described in Equation (17) on system (41), we achieve:

λ̈ = v + δ (43)
−1
where λ̈ = [λ̈11 , λ̈21 ] T ,v = [v1 , v2 ] T and δ = [δ1 , δ2 ] T = drob Brob v + Crob .
Rewriting Equation (43) in terms of two linear integrator systems subject to perturba-
tion yields the following expressions:
 
λ̇11 = λ12
 λ̇21 = λ22

MBF1 λ̇12 = v1 + δ1 MBF2 λ̇22 = v2 + δ2 (44)
 
Y1 = λ11 Y2 = λ21
 
Sensors 2024, 24, 2849 10 of 24

Consider ∆1 and ∆2 as the differentials of δ1 and δ2 with respect to time t, respectively.

We assume that both δi and ∆i (i = 1, 2) are bounded. In practical applications, determining
the actual values of the lumped disturbances δ1 and δ2 that affect the system is considered
a challenging problem. Hence, an observer is required to estimate these values.

4.1. ESO Design

The extended state observer (ESO) plays a vital role in system control by simultane-
ously estimating the system states and uncertainties. This capability enables the ESO to
effectively reject or compensate for disturbances, enhancing the system’s robustness and
performance. The ESO takes into account all factors that affect the system and treats param-
eter uncertainties and external perturbations as a single observed disturbance. The ESO is
named as such because it estimates uncertainties as an extended state. Its benefits include
not being reliant on the mathematical model of the system, as well as having a straightfor-
ward implementation and demonstrating good performance. Consider λ13 = δ1 , α23 = δ2
as an extended state for system (44). The latter can be expressed as follows:
 


 λ̇ 11 = λ 12 

 λ̇21 = λ22

λ̇ = λ + v 
λ̇ = λ + v
12 13 1 22 23 2
(45)


 λ̇ 13 = ∆ 1 

 λ̇ 23 = ∆ 2

Y = λ 
Y = λ
1 11 1 21

We can express systems (45) in matrix form as follows:

(
λ̇1 = A x λ1 + Bx v1 + Ex ∆1
(46)
Y1 = Cx λ1

(
λ̇2 = Ay λ2 + By v2 + Ey ∆2
(47)
Y2 = Cy λ2
   
0 1 0 0
where λ1 = [λ11 , λ12 , λ13 ] T , λ2 = [λ21 , λ22 , λ23 ] T , A x = Ay = 0 0 1, Bx = By = 1,
0 0 0 0
   T
Cx = Cy = 1 0 0 , Ex = Ey = 0 0 1 . The expression for the Extended State
Observer (ESO) corresponding to each extended system (46) and (47) can be given as follows:

λ̂˙ 1 = A x λ̂1 + Bx v x + α gx Cx (λ1 − λ̂1 ) (48)

λ̂˙ 2 = Ay λ̂2 + By vy + α gy Cy (λ2 − λ̂2 ) (49)
where α gx = [α11 , α12 , α13 ] T , α gy = [α21 , α22 , α23 ] T . To determine the observer gains αij
(i = 1, 2, 3), ( j = 1, 2, 3), we can adopt the methodology proposed by Gao [39] outlined in
the following manner:
s3 + α11 s2 + α12 s + α13 = (s + γxo )3 (50)
s3 + α21 s2 + α22 s + α23 = (s + γyo )3 (51)
The choice of γxo and γyo is made to ensure that Equations (50) and (51) form Hurwitz
polynomials with respect to the complex variable. The observer gain can be formulated as a
function of the ESO bandwidth by utilizing Equations (50) and (51), as demonstrated below:
α11 = 3γxo , α12 = 3γ2xo , α13 = γ3xo
2 3
(52)
α21 = 3γyo , α22 = 3γyo , α23 = γyo .
Sensors 2024, 24, 2849 11 of 24

The observer error associated with each ESO can be defined by employing
Equations (46)–(49) as follows:

ê˙x = λ̇1 − λ̂˙ 1 = ( A x − α gx Cx )êx + Ex ∆1 (53)

ê˙y = λ̇2 − λ̂˙ 2 = ( Ay − α gy Cy )êy + Ey ∆2 (54)

It is possible to express Equations (53) and (54) in matrix form as shown below:

ê˙ = Ĥ ê + Ed (55)
   
  −α11 1 0 −α21 1 0
Ĥ1 03
where ê = [êx , ê˙ x , ê¨ x , êy , ê˙ y , ê¨ y ]T , Ĥ = , Ĥ1 = −α12 0 1, Ĥ2 = −α22 0 1
03 ˆH2
−α13 0 0 −α23 0 0
T
, Ed = 0 0 ∆1 0 0 ∆2 .


Lemma 1. In Equation (55), the boundedness of limt→∞ ê(t) can be guaranteed if at least one of
the following two conditions is satisfied:
• δi < n1 , i = 1, 2 for all time t;
• ∆i < n2 , i = 1, 2 for all time t.
Asymptotic stability of the estimated error dynamics can be achieved when the values of
δi , i = 1, 2, are either directly obtained or assumed to be constant, leading to ∆i = 0, i = 1, 2.
In this scenario, the positive constants n1 and n2 play a vital role in ensuring the system’s sta-
bility. Lemma 1, as stated in Zhang et al. [40] , establishes that the roots of the matrix Ĥ in
Equation (55) reside in the left half plane. This result is ensured by the nonnegativity of the band-
widths γxo and γyo . Consequently, it can be deduced that the estimated error dynamics described by
Equations (53) and (54) are asymptotically stable.

4.2. New Robust Feedback Controller

The feedback controller presented in Equations (19) and (20) relies on state measure-
ments, but except for λ11 and λ21 , the remaining states cannot be accurately measured.
To solve this problem, the state estimation obtained through the two ESOs defined in
Equations (48) and (49) are used instead. Furthermore, in order to simplify the compen-
sation of the lumped disturbances δ1 and δ2 , they are replaced by their approximations,
δ̂1 and δ̂2 . By incorporating the results of the extended state observers (ESOs), a robust
feedback controller can be developed in the following manner:

vSADRCx = λ̈ xd − β 1 ê˙1 − k1 sat(ŝ x ) − δ̂1 (56)

vSADRCy = λ̈yd − β 2 ê˙2 − k2 sat(ŝy ) − δ̂2 (57)

according to the sliding mode active disturbance rejection control feedback given in
Equations (56) and (57), we can obtain the new robust tracking controller named Flatness-
Sliding-Active-Disturbance-Rejection Control (FSADRC), defined as follows:

λ̈ − β 1 ê˙1 − k1 sat(ŝ x ) − δ̂1

   
u FSADRCx
= Br−1 xd (58)
u FSADRCy λ̈yd − β 2 ê˙2 − k2 sat(ŝy ) − δ̂2

where êr1 = λ̂11 − λ xd and êr2 = λ̂21 − λyd . The schematic diagram presented in Figure 3
illustrates the principle of trajectory tracking control for a mobile robot.
Sensors 2024, 24, 2849 12 of 24

Figure 3. Mobile robot trajectory tracking control principle scheme.

4.3. Stability Analysis of the Closed-Loop System

This section will address the stability analysis of the tracking error systems for x and
y, utilizing the estimation error defined by Equations (53) and (54). In order to prove the
stability of the error dynamics of position x, Lyapunov’s function is chosen as follows:

1 2
Vsx = s (59)
2 x

where s x = ė1 + β 1 e1 = λ̇11 − λ̇ xd + β 1 (λ11 − λ xd ), λ11 = x, λ xd = xd .

We can define the derivative of the Lyapunov function Vsx as follows:

V̇sx = s x ṡ x = s x (λ̈11 − λ̈ xd + β 1 (λ̇11 − λ̇ xd )) (60)

When replacing λ̈11 by its Equation (44) defined by λ̇11 = v1 + δ1 , we obtain:

V̇sx = s x ṡ x = s x (v1 + δ1 − λ̈ xd + β 1 (λ̇11 − λ̇ xd )) (61)

When v1 represents the feedback controller, substituting it with the proposed robust
feedback tracking control, denoted as vSADRCx defined by Equation (56), yields:

V̇sx = s x ṡ x = s x (λ̈ xd − β 1 (λ̂˙ 11 − λ̇ xd ) − k1 sat(ŝ x ) − δ̂1 + δ1 − λ̈ xd + β 1 (λ̇11 − λ̇ xd ))

(62)
V̇sx = s x ṡ x = s x ( β 1 (λ̇11 − λ̂˙ 11 ) + δ1 − δ̂1 − k1 sat(ŝ x ))

where s x is defined as follows:

(
sx
asx if |s x | ≤ asx
Sat(s x ) = (63)
sgn(s x ) i f |s x | > asx

Concerning the stability and boundedness of the ESO defined by Equation (53), it can
be achieved by choosing α gx in such a way that the eigenvalues of A x − α gx Cx are negative,
indicating poles in the left-hand plane, and ensuring that uncertainty is bounded. As a
Sensors 2024, 24, 2849 13 of 24

˙ → 0. This implies that λ̂11 → λ11 , δ̂1 → δ1 , and ŝ x → s x . In this scenario,

result, the error êx
the Lyapunov function defined by Equation (62) is formulated as follows:

V̇sx = −s x (k1 sat(s x )) (64)

Since Sat(s x ), defined by Equation (63), is divided into two segments, the proof process
will be analyzed in two cases. In the first scenario, when the saturation function is defined
as described by:
sx
Sat(s x ) = (65)
asx
Moreover, the Lyapunov function is defined as follows:

V̇sx = − aksx (s2x ) ≤ 0 (66)

Alternatively, when the saturation function is given by:

Sat(s x ) = sgn(s x ) (67)

the Lyapunov function takes the form:

V̇sx = −k1 s x sgn(s x ) ≤ 0 (68)

Thus, based on Equations (66) and (68), it can be concluded that the Lyapunov function
V̇sx is negative regardless of the definition of the function Sat(s x ). As a result, the tracking
error of the position x is stable. Similarly, the same conclusions about the stability of the
closed-loop system y can be drawn.

5. Simulation Results
This section presents simulation tests to validate the efficacy and superiority of the
suggested controller, flatness sliding active disturbance rejection control (FSADRC), as de-
fined by Equation (58). The proposed control is evaluated against flatness sliding mode
control (FSMC), represented by Equation (39), and flatness-based tracking control (FBTC),
as defined in Equation (24), using computer simulation results. The parameters of the WMR
are r = 0.1 m, b = 0.15 m. To enhance the observation and comparison of the simulation
results, we have chosen two types of reference trajectories: a circular path and a Bézier
curve. Additionally, we also consider two different scenarios of perturbation. The controller
design parameters of FBTC, FSMC, and FSADRC are chosen as m x = my = 1, ϵxc = ϵyc = 2,
β 1 = β 2 = 5, and k1 = k2 = 10. As suggested by Gao [39], it is advisable to select the
observer bandwidth to be sufficiently higher than the controller bandwidth. This ensures
that the observer dynamics remain faster than the system dynamics, enabling effective dis-
turbance estimation and compensation. In our case, we have chosen observer bandwidths
of γxo = γyo = 6 rad/s to fulfill this requirement and ensure robust performance of the
control system. To ensure that the sliding mode control system achieves both satisfactory
dynamic and steady-state performance, and to prevent chatter in the control signal, the cut
and dry method is frequently employed to establish the thickness of the boundary layer.
Specifically, in this case, asx = asy = 0.3 is chosen.

5.1. First Scenario

In this simulation, we consider that slip velocities vt and vs can be up to 30% of the
forward speed. Thus, κ1 = 0.3. In addition, the WMR is subjected to constant wind
perturbation defined as follows:

p x = py = 3 m/s, ws = 0.5 rad/s (69)

Sensors 2024, 24, 2849 14 of 24

The reference trajectory considered in this scenario is a circle, which is defined by the
following equation:
xr = cos(t), yr = sin(t) (70)
The performance of the uncertain WMR systems under different control strategies,
namely FBTC, FSMC, and FSADRC, is depicted in Figure 4. Figure 5 shows the results
of the estimated lumped disturbance affecting the x and y channels obtained using the
extended state observer (ESO). Figure 6 illustrates the control input applied to the wheeled
mobile robot under the conditions of the first scenario. The simulation results indicate
that the uncertainty caused by slow wind perturbation and slip decreases the tracking
performance in trajectory following, rendering FBTC ineffective as a controller. On the other
hand, both FSMC and FSADRC demonstrate robustness in handling the overall disturbance
affecting the WMR model. These controllers exhibit the ability to mitigate disturbances and
successfully maintain the desired trajectory of the WMR system. Consequently, it can be
inferred that controllers that disregard uncertain models, despite being feedback controllers,
may exhibit unsatisfactory performance. The fundamental distinction between the FSMC
and FSADRC controllers lies in their design methodologies and approaches. FSMC relies
on finely-tuned gains to achieve disturbance rejection, which can lead to chattering due
to the relatively high gain values. In contrast, FSADRC combines the advantages of the
boundary layer method to minimize chattering and an ESO to estimate and eliminate
lumped disturbance.
1.5 FBTC 1.5 FBTC
FSMC FSMC
FSADRC FSADRC
Reference trajectory Reference trajectory
1 1

0.5 0.5
x[m]

y[m]

0 0

-0.5 -0.5

-1 -1

-1.5 -1.5
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time[s] Time[s]

1.5 6
FBTC FBTC
FSMC FSMC
FSADRC FSADRC
1 Reference trajectory 4 Reference trajectory

0.5 2
[rad]
y(m)

0 0

-0.5 -2

-1 -4

-1.5 -6
-1.5 -1 -0.5 0 0.5 1 1.5 0 1 2 3 4 5 6 7 8 9 10
x(m) Time[s]

Figure 4. Simulation tracking results of the wheeled mobile robot under the conditions of the
first scenario.
Sensors 2024, 24, 2849 15 of 24

2.5 7
Estimated Estimated
Lumped Disturbance Lumped Disturbance
6

Lumped disturbance of x position channel

Lumped disturbance of y position channel

2
5

4
1.5
3

1 2

1
0.5
0

-1
0
-2

-0.5 -3
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time[sec] Time[sec]

Figure 5. Lumped disturbance affecting the x and y position channels in the context of the
first scenario.

15 8
FBTC FBTC
FSMC FSMC
FSADRC 6 FSADRC
10

4
5
U 2 [rad/s]
U 1 [m/s]

0
0

-5
-2

-10 -4
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time[s] Time[s]

Figure 6. Control input applied to the wheeled mobile robot under the conditions of the first scenario.

5.2. Second Scenario

The objective of this simulation is to create and follow a trajectory for a robot, start-
ing from an initial state where x (0) = y(0) = 0, and reaching a final state specified by
x (10) = 3.5 and y(10) = 5. This trajectory must navigate through a room containing ob-
stacles, while also considering time-varying wind disturbances and slipping. The desired
trajectory should meet the following criteria: minimizing energy consumption, maneuver-
ing around static obstacles, and adhering to the specified state constraints as follows:

0 m ≤ λ xd ≤ 4 m, 0 m ≤ λyd ≤ 6 m (71)

The optimal trajectory generation method proposed in [17] offers a solution to obtain
the desired trajectory by solving a nonlinear optimization problem. By integrating the
principles of flatness, the collocation method, and B-spline functions, this method efficiently
generates trajectories while guaranteeing constraint satisfaction. To ensure consistency in
the simulation results, the parameters for all three controllers remain unchanged from the
previous simulations. Considering an uncertain initial condition of x̂ (0) = 1 and ŷ(0) = 1
for the wheeled mobile robot (WMR), we further specify that the slip velocities vt and vs
can potentially reach up to 50% to 70%. In addition, we take into account the influence of
sinusoidal wind disturbances. In contrast to the initial scenario, the disturbance signals
Sensors 2024, 24, 2849 16 of 24

consist of combinations of multi-frequency sinusoidal signals representing time-varying

disturbances, particularly wind, defined as follows:

p x = py = 1.5 + 2.5sin(4t) + 4.5cos(2t), ws = 1.5 + 3cos(2t) (72)

The simulation results regarding trajectory tracking performance of the second sce-
nario are depicted in Figure 7. Based on these figures, it can be observed that the WMR
system experiences significant divergence from the desired trajectory when affected by
slippage and external disturbances, rendering FBTC ineffective as a controller. The FSMC
controller’s intervention through the sliding mode’s discontinuous term eliminates uncer-
tainty effects and maintains the stability of the closed-loop control. However, as shown
in Figure 8, the presence of chattering in the FSMC control signals negatively impacts the
system’s behavior. Hence, it can be inferred that while FSMC is a robust control approach,
its practical applicability is quite restricted. Therefore, developing a control approach
capable of mitigating the chattering effect while maintaining the robustness advantage
provided by SMC is necessary. The results of the lumped disturbance estimation for this
simulation are illustrated in Figure 9. According to the simulation findings, the mobile
robot satisfactory trajectory tracking performance when confronted with model distur-
bances and uncertain initial conditions while employing the FSADRC controller. Of greater
significance, the proposed control methodology achieves superior tracking of the desired
trajectory, devoid of the chattering phenomenon, and enhances tracking performance
against aggressive disturbances.
4.5 6
FBTC FBTC
4 FSMC FSMC
FSADRC 5 FSADRC
Reference trajectory Reference trajectory
3.5

3 4

2.5
x[m]

y[m]

3
2

1.5 2

1
1
0.5

0 0
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time[s] Time[s]

150
FBTC
FSMC
FSADRC
Reference trajectory
100

50
[degres]

-50

-100
0 1 2 3 4 5 6 7 8 9 10
Time[s]

Figure 7. Simulation tracking results of the wheeled mobile robot in the conditions of the
second scenario.
Sensors 2024, 24, 2849 17 of 24

12 0.12
FBTC FBTC
11 FSMC FSMC
0.1
FSADRC FSADRC
10
0.08
9
0.06
8

U2[rad/s]
U1[m/s]
7 0.04

6
0.02
5
0
4
-0.02
3

2 -0.04
0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Time[s] Time[s]

Figure 8. Control input applied to the wheeled mobile robot under the conditions of the
second scenario.

6 5
Estimated Estimated
Lumped Disturbance Lumped Disturbance
Lumped disturbance of x position channel

Lumped disturbance of y position channel

5 4

4
3

3
2
2
1
1

0
0

-1 -1

-2 -2
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Time[sec] Time[sec]

Figure 9. Lumped disturbance affecting the x and y position channels in the context of the
second scenario.

6. Tracking the Experimental Results of a Wheeled Mobile Robot

This section outlines experiments conducted with the TurtleBot3, a Wheeled Mobile
Robot (WMR), to evaluate a proposed methodology. The TurtleBot provides a cost-effective
platform for researchers to explore and validate control algorithms without requiring expen-
sive robotic systems. Its compatibility with the Robot Operating System (ROS) enhances its
functionalities, offering resources for algorithm development and experimentation. With Li-
DAR, IMU, and wheel encoders onboard, the TurtleBot3 provides precise environmental
feedback, facilitating algorithm optimization. Researchers can augment the system with
additional sensors or hardware components to evaluate various control algorithms across
diverse scenarios. To facilitate the observation and comparison of experimental results,
we have selected two types of reference trajectories: an eight-shaped path and a Bézier
curve. Additionally, we have considered two different scenarios of perturbation: the first in-
volves slowly time-varying disturbances, while the second entails aggressive time-varying
disturbances. For further validation, the performance of the proposed control method
is compared with other state-of-the-art control techniques such as backstepping tracking
control (BTC) [41], flatness active disturbance rejection control (FADRC) introduced in [42],
flatness-based tracking control (FBTC) as defined by Equation (39), and backstepping slid-
ing active disturbance rejection control (BSADRC) [43]. The controller design parameters
selected for the experimental results are identical to those chosen for the simulation results.
Sensors 2024, 24, 2849 18 of 24

6.1. First Experiment with Slowly Time-Varying Disturbances

In this experiment, eight shapes were chosen for the reference trajectory, as
outlined below:
xr = 2cos(t), yr = −2sin(t) (73)
To replicate real-world navigation conditions for the WMR, high-speed fans are uti-
lized in the laboratory to simulate windy environments. Additionally, a stick is employed
to disturb the castors of the WMR, creating slipping incidents, thus adding further realism
to the testing environment. Figure 10 illustrates the real-time tracking of the eight-shaped
reference trajectory of the WMR using the proposed control method described in this paper.

Figure 10. Real-time trajectory tracking experiment.

Simulation of the experiment under identical conditions reveal tracking trajectories in

Figure 11. Figure 12 illustrates lumped disturbance estimation, while Figure 13 displays
control torques. Based on the experimental results shown in Figure 11, it is evident that
FADRC, FSADRC, and BSADRC methods excel at tracking trajectories even in the face
of genuine uncertainty. Conversely, the FBTC and BTC methods demonstrate significant
shortcomings when it comes to handling uncertainties. To assess the superiority of the pro-
posed control, we will conduct a thorough study in the subsequent section. This study will
include a quantitative analysis of the controllers under more severe disturbance conditions.
6 8
BTC
FBTC
FSMC
6
BSARDC
4
FBARDC
Reference trajectory
4

2
2
x[m]

y[m]

0 0

-2
-2

-4 BTC
FBTC
-4 FSMC
-6 BSARDC
FBARDC
Reference trajectory
-6 -8
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time[s] Time[s]

Figure 11. Cont.

Sensors 2024, 24, 2849 19 of 24

4 14
BTC BTC
FBTC FBTC
FSMC 12 FSMC
3 BSARDC BSARDC
FBARDC 10 FBARDC
Reference trajectory Reference trajectory
2 8

6
1

[rad]
y(m)
4
0
2

-1 0

-2
-2
-4

-3 -6
-3 -2 -1 0 1 2 3 0 2 4 6 8 10 12 14 16 18 20
x(m) Time[s]

Figure 11. Results of the wheeled mobile robot’s tracking under the conditions of the first experi-
ment scenario.

10 5

4
Lumped disturbance of x position channel

Lumped disturbance of y position channel

3

5 2

0 -1

-2

-3

-5 -4
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time[sec] Time[sec]

Figure 12. Estimation values of the lumped disturbances under the conditions of the first experi-
ment scenario.

2.5 2
FSADRC FSADRC
2
1.5
1.5
1
1
0.5
0.5
w l [N*m]

w r[N*m]

0 0

-0.5
-0.5
-1
-1
-1.5
-1.5
-2

-2.5 -2
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time[s] Time[s]

Figure 13. Control torques applied to the right and left wheels to track the eight-shaped
reference trajectory.

6.2. Second Experiment with Aggressive Time-Varying Disturbances

In this experiment, we intensify the frequency of disturbance variation generated by
the industrial ventilator and subject the robot to aggressive impacts with a stick to assess
Sensors 2024, 24, 2849 20 of 24

the effectiveness of the proposed controller. Additionally, we adopt the eighth-order Bézier
curve as a reference trajectory for both the x and y positions, defined as follows:

λ xd = xr = Px0 (1 − t)8 + 8Px1 (1 − t)7 t + 28Px2 (1 − t)6 t2 P + 56Px3 (1 − t)5 t3 + . . .

70Px4 (1 − t)4 t4 + 56Px5 (1 − t)3 t5 + 28Px6 (1 − t)2 t6 + 8Px7 (1 − t)t7 + Px8 t8 .
(74)
λyd = yr = Py0 (1 − t)8 + 8Py1 (1 − t)7 t + 28Py2 (1 − t)6 t2 P + 56Py3 (1 − t)5 t3 + . . .
70Py4 (1 − t)4 t4 + 56Py5 (1 − t)3 t5 + 28Py6 (1 − t)2 t6 + 8Py7 (1 − t)t7 + Py8 t8 .

where Pxj , Pyj , and j = 0 . . . 8 represent the control parameters of the reference trajectory.
These parameters may vary depending on several factors, including the robot’s initial
position, the desired final position, and constraints such as obstacle avoidance. As an
example, we select control parameters that allow the WMR to transition from its initial
state qr (0) = [0, 0, 0] T to the desired final state qr (20) = [2, 2, 0] T . The tracking experiment
results of the WMR under aggressive time-varying disturbances are depicted in Figure 14.
In Figure 15, the lumped disturbance affecting the WMR within the context of the second
experimental scenario is displayed. Similarly, Figure 16 illustrates the proposed control
input applied to the wheeled mobile robot within the same context.

4.5 4
BTC BTC
FBTC FBTC
4 FSMC FSMC
BSARDC BSARDC
3
3.5 FBARDC FBARDC
Reference trajectory Reference trajectory

3
2
2.5
x[m]

y[m]

2 1

1.5
0
1

0.5
-1
0

-0.5 -2
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time[s] Time[s]

5 0.6
BTC
FBTC
FSMC
BSARDC 0.4
4 FBARDC
Reference trajectory
0.2
3

0
[rad]
y[m]

2 BTC
FBTC
-0.2
FSMC
BSARDC
1 FBARDC
-0.4 Reference trajectory

0
-0.6

-1 -0.8
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 0 2 4 6 8 10 12 14 16 18 20
Time[s] Time[s]

Figure 14. Results of the wheeled mobile robot’s tracking under the conditions of the second
experiment scenario.
Sensors 2024, 24, 2849 21 of 24

200 400

Lumped disturbance of x position channel

Lumped disturbance of y position channel

300
150

200
100

100

50
0

0
-100

-50 -200
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time[sec] Time[sec]

Figure 15. Estimated values of the lumped disturbances under the conditions of the second experi-
ment scenario.

0.2 20
FSADRC FSADRC
0.15 15

0.1
10

0.05
5
w l [N*m]

w r[N*m]

0
0
-0.05
-5
-0.1

-10
-0.15

-0.2 -15

-0.25 -20
0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 20
Time[s] Time[s]

Figure 16. Control torques applied to the right and left wheels to track the Bézier reference trajectory.

To quantitatively assess the tracking performance of the WMR, we employed the

integral absolute error (IAE) and the control effort performance index as comparison
metrics. The IAE is computed for each of the control strategies in the following manner:
Z t
f
IAEi = |ei (t)|dt. ei (t) = λi (t) − λid (t), (75)
0

where t f is the total simulation duration and i = 1, 2, represents the position in the x and y
direction, respectively. The control effort is given as follows:

N
1
Pavg =
N ∑ u2 ( k ) (76)
k =1

where N indicates the total count of samples. The associated key performance indicators
IAE and Pavg for both strategies are provided in Table 1.

Table 1. Performance indexes IAE and Pavg .

Index BTC FBTC FADRC FSADRC BSADRC

IAE 5.5351 4.2654 0.07 0.0127 0.02
Pavg 2.5351 0.261 0.1266 0.13 1.253
Sensors 2024, 24, 2849 22 of 24

Examining the data in Table 1, it is evident that the FSADRC controller outperforms
the BTC, FBTC, FADRC, and FSMC methods in terms of tracking performance. Although its
tracking performance is nearly comparable to that of the BSADRC, the FSADRC requires
minimal effort to accomplish its task compared to the BSADRC. This characteristic is
particularly crucial in contexts where energy resources are limited, such as in mobile or
autonomous applications. The enhanced efficiency of the FSADRC over the BSADRC
is explained by the advantage of flatness control, which simplifies controller design by
transforming the nonlinear system into a linear one. This feature makes all control based
on the concept of flatness less complex than control based on backstepping. Ultimately,
the experiment and table findings show that the disturbance rejection function simplifies the
system model by addressing real-time modeling uncertainties. Consequently, the FSADRC
method relies less on an exact analytical model description, treating unknown dynamics
as internal disturbances compensated for by the rejection function. This enhances the
robustness of FSADRC, which also incorporates the boundary layer technique to alleviate
chattering effects.

7. Conclusions
This paper aims to introduce a robust control methodology for uncertain wheeled
mobile robots (WMR). By employing flatness-based control, the nonlinear kinematic model
of the WMR undergoes transformation into a canonical form, enabling the implementation
of a robust feedback controller that incorporates boundary layer sliding mode control and
extended state observer techniques. Simulation results conducted under various scenarios
of uncertainties illustrate the effectiveness of FSADRC in enhancing the trajectory tracking
performance of the WMR when compared to BTC, FBTC, and FADRC, even amid varia-
tions in slipping and external wind disturbances. Furthermore, within the same context,
FSADRC demonstrates comparable efficiency to BSADRC in terms of trajectory tracking,
while exhibiting an advantage in effort usage due to its flatness property. The smooth
operation of FSADRC, coupled with its resilience against parameter variations and external
disturbances, renders it a practical choice for real-world applications. Moreover, exper-
imental findings using the TurtleBot3 validate the efficacy of the proposed FSADRC in
real-world navigational tasks. In future studies, the application of FSADRC will extend to
other robotic systems, such as quadrotors and arm manipulators, to assess its effectiveness
and explore its potential for broader deployment.

Author Contributions: Methodology, A.A. (Amine Abadi); Validation, A.A. (Amine Abadi) and A.A.
(Amani Ayeb); Resources, D.F.; Writing—original draft, A.A. (Amine Abadi); Writing—review &
editing, M.L. and H.M.; Visualization, D.F.; Project administration, T.B.; Funding acquisition, T.B. All
authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Data are contained within the article.
Conflicts of Interest: The authors declare no conflicts of interest.

Abbreviations
The following abbreviations are used in this manuscript:

WMR Wheeled mobile robot

ADRC Active disturbance rejection control
ESO Extended state observer
BTC Backstepping tracking control
FBTC Flatness-based tracking control
FSMC Flatness sliding mode control
Sensors 2024, 24, 2849 23 of 24

FSADRC Flatness sliding active disturbance rejection control

BSADRC Backstepping sliding active disturbance rejection control

References
1. Al, A.S.M.A.O.; Al-Qassa, A.; Nasser, A.R.; Alkhayyat, A.; Humaidi, A.J.; Ibraheem, I.K. Embedded design and implementation
of mobile robot for surveillance applications. Indones. J. Sci. Technol. 2021, 6, 427–440. [CrossRef]
2. Ebel, H.; Rosenfelder, M.; Eberhard, P. Cooperative object transportation with differential-drive mobile robots: Control and
experimentation. Robot. Auton. Syst. 2024, 173, 104612. [CrossRef]
3. Yépez-Ponce, D.F.; Salcedo, J.V.; Rosero-Montalvo, P.D.; Sanchis, J. Mobile robotics in smart farming: Current trends and
applications. Front. Artif. Intell. 2023, 6, 1213330. [CrossRef]
4. Garaffa, L.C.; Basso, M.; Konzen, A.A.; de Freitas, E.P. Reinforcement learning for mobile robotics exploration: A survey. IEEE
Trans. Neural Netw. Learn. Syst. 2021, 34, 3796–3810. [CrossRef]
5. Zangina, U.; Buyamin, S.; Abidin, M.S.Z.; Azimi, M.S.; Hasan, H.S. Non-linear PID controller for trajectory tracking of a
differential drive mobile robot. J. Mech. Eng. Res. Dev. 2020, 43, 255–269. [CrossRef]
6. Benchouche, W.; Mellah, R.; Bennouna, M.S. The Impact of the dynamic model in feedback linearization trajectory tracking of a
mobile robot. Period. Polytech. Electr. Eng. Comput. Sci. 2021, 65, 329–343. [CrossRef]
7. Yousuf, B.M.; Saboor Khan, A.; Munir Khan, S. Dynamic modeling and tracking for nonholonomic mobile robot using PID and
backstepping. Adv. Control. Appl. Eng. Ind. Syst. 2021, 3, e71. [CrossRef]
8. Fliess, M.; Lévine, J.; Martin, P.; Rouchon, P. A lie-backlund approach to equivalence and flatness of nonlinear systems. IEEE
Trans. Autom. Control. 1999, 44, 922–937. [CrossRef]
9. Rigatos, G.; Zervos, N.; Siano, P.; Wira, P.; Abbaszadeh, M. Flatness-based control for steam-turbine power generation units using
a disturbance observer. IET Electr. Power Appl. 2021, 15, 1013–1028. [CrossRef]
10. Rigatos, G.; Hamida, M.A.; Abbaszadeh, M.; Siano, P. Flatness-based disturbance observer for condition monitoring of marine
power generation units. Proc. Inst. Mech. Eng. Part I J. Syst. Control. 2023, 237, 1620–1634. [CrossRef]
11. Jing, C.; Xu, H.; Song, X.; Lu, B. Adaptive extended state observer-based flatness nonlinear output control for torque tracking of
electrohydraulic loading system. Trans. Inst. Meas. Control 2018, 40, 2999–3009. [CrossRef]
12. Rigatos, G.; Wira, P.; Abbaszadeh, M.; Pomares, J. Flatness-based control in successive loops for industrial and mobile robots.
In Proceedings of the IECON 2022–48th Annual Conference of the IEEE Industrial Electronics Society, Brussels, Belgium,
17–20 October 2022; pp. 1–6. [CrossRef]
13. Lin-Shi, X.; Simon, T.; Trégouët, J.F.; Morel, H. Flatness-Based Control of an m-Branch Power Flow Controller for Meshed DC
Microgrids. In Proceedings of the IEEE 1st International Power Electronics and Application Symposium (PEAS), Shanghai, China,
13–15 November 2021; pp. 1–6. [CrossRef]
14. Rigatos, G.; Abbaszadeh, M.; Pomares, J.; Wira, P.; Cuccurullo, G. Flatness-based control in successive loops for robotic
manipulators and autonomous vehicles. Int. J. Syst. Sci. 2024, 55, 954–979. [CrossRef]
15. Yu, X.; Zhou, X.; Guo, K.; Jia, J.; Guo, L.; Zhang, Y. Safety flight control for a quadrotor UAV using differential flatness and
dual-loop observers. IEEE Trans. Ind. Electron. 2021, 69, 13326–13336. [CrossRef]
16. Abadi, A.; El Amraoui, A.; Mekki, H.; Ramdani, N. Guaranteed trajectory tracking control based on interval observer for
quadrotors. Int. J. Control 2020, 93, 2743–2759. [CrossRef]
17. Abadi, A.; Mekki, H.; Brahim, A.B.H.; El Amraoui, A.; Ramdani, N. Optimal trajectory generation and flatness tracking control
for a mobile robot. In Proceedings of the 18th International Conference on Sciences and Techniques of Automatic Control and
Computer Engineering (STA), Monastir, Tunisia, 21–23 December 2017; pp. 223–228. [CrossRef]
18. Kaaniche, K.; El-Hamrawy, O.; Rashid, N.; Albekairi, M.; Mekki, H. Mobile Robot Control Based on 3D Visual Servoing: A New
Approach Combining Pose Estimation by Neural Network and Differential Flatness. Appl. Sci. 2022, 12, 6167. [CrossRef]
19. Nasr, S.; Mekki, H.; Bouallegue, K. A multi-scroll chaotic system for a higher coverage path planning of a mobile robot using
flatness controller. Chaos Solitons Fractals 2022, 118, 366–375. [CrossRef]
20. Yakovlev, K,S.; Andreychuk, A.; Belinskaya, J.; Makarov, D. Safe interval path planning and flatness-based control for navigation
of a mobile robot among static and dynamic obstacles. Autom. Remote Control 2022, 83, 903–918. [CrossRef]
21. Khalesi, R.; Yousefi, M.; Pishkenari, H.N.; Vossoughi, G. Robust independent and simultaneous position control of multiple
magnetic microrobots by sliding mode controller. Mechatronics 2022, 84, 102776. [CrossRef]
22. Lian, S.; Meng, W.; Shao, K.; Zheng, J.; Zhu, S.; Li, H. Full attitude control of a quadrotor using fast nonsingular terminal sliding
mode with angular velocity planning. IEEE Trans. Ind. Electron. 2022, 70, 3975–3984. [CrossRef]
23. Li, J.; Wang, J.; Peng, H.; Hu, Y.; Su, H. Fuzzy-torque approximation-enhanced sliding mode control for lateral stability of mobile
robot. IEEE Trans. Syst. Man Cybern. 2021, 52, 2491–2500. [CrossRef]
24. Utkin, V.; Lee, H. Chattering problem in sliding mode control systems. In Proceedings of the International Workshop on Variable
Structure Systems, Alghero, Sardinia, 5–7 June 2006; pp. 346–350. [CrossRef]
25. Utkin, V. Discussion aspects of high-order sliding mode control. IEEE Trans. Autom. Control 2015, 61, 829–833. [CrossRef]
26. Boiko, I.M. Chattering in sliding mode control systems with boundary layer approximation of discontinuous control. Int. J. Syst.
Sci. 2015, 44, 1126–1133. [CrossRef]
Sensors 2024, 24, 2849 24 of 24

27. Guo, R.; Ding, Y.; Yue, X. Active adaptive continuous nonsingular terminal sliding mode controller for hypersonic vehicle. Aerosp.
Sci. Technol. 2023, 137, 108279. [CrossRef]
28. Belguedri, M.; Benrabah, A.; Khoucha, F.; Benbouzid, M.; Benmansour, K. An improved uncertainty and disturbance estimator-
based speed control for grid-connected pumping kite wind generator. Control Eng. Pract. 2024, 143, 105795. [CrossRef]
29. Han, J. From PID to active disturbance rejection control. IEEE Trans. Ind. Electron. 2023, 56, 900–906. [CrossRef]
30. Liu, L.; Wang, D.; Peng, Z. State recovery and disturbance estimation of unmanned surface vehicles based on nonlinear extended
state observers. Ocean Eng. 2019, 171, 625–632. [CrossRef]
31. Zhao, L.; Li, Z.; Li, H.; Liu, B. Backstepping integral sliding mode control for pneumatic manipulators via adaptive extended state
observers. ISA Trans. 2024, 144, 374–384. [CrossRef] [PubMed]
32. Chowdhury, D.; Al-Nadawi, Y.K.; Tan, X. Dynamic inversion-based hysteresis compensation using extended high-gain observer.
Automatica 2022, 135, 109977. [CrossRef]
33. Xie, T.; Li, Y.; Jiang, Y.; An, L.; Wu, H. Backstepping active disturbance rejection control for trajectory tracking of underactuated
autonomous underwater vehicles with position error constraint. Int. J. Adv. Robot. Syst. 2020, 17. [CrossRef]
34. Qi, G.; Hu, J.; Li, L.; Li, K. Integral Compensation Function Observer and Its Application to Disturbance-Rejection Control of
QUAV Attitude. IEEE Trans. Cybern. 2024, 1–12. [CrossRef]
35. Aole, S.; Elamvazuthi, I.; Waghmare, L.; Patre, B.; Meriaudeau, F. Improved active disturbance rejection control for trajectory
tracking control of lower limb robotic rehabilitation exoskeleton. Sensors 2020, 20, 3681. [CrossRef] [PubMed]
36. Hu, Y.; Li, B.; Jiang, B.; Han, J.; Wen, C.Y. Disturbance Observer-Based Model Predictive Control for an Unmanned Underwater
Vehicle. J. Mar. Sci. Eng. 2024, 12, 94. [CrossRef]
37. Ryu, J.C.; Agrawal, S.K. Differential flatness-based robust control of mobile robots in the presence of slip. Int. Robot. Res. 2011, 30,
463–475. [CrossRef]
38. Mauledoux, M.; Mejia-Ruda, E.; Aviles Sanchez, O.; Dutra, M.S.; Rojas Arias, A. Design of Sliding Mode Based Differential
Flatness Control of Leg-wheel Hybrid Robot. Appl. Mech. Mater. 2011, 835, 681–686. [CrossRef]
39. Gao, Z. Scaling and bandwidth-parameterization based controller tuning. In Proceedings of the 2003 American Control
Conference, Denver, CO, USA, 4–6 June 2003; pp. 4989–4996. [CrossRef]
40. Zhang, Y.; Jiang, Z.; Yang, H.; Cheng, J.; Zhang, W. High-order extended state observer-enhanced control for a hypersonic flight
vehicle with parameter uncertainty and external disturbance. Proc. Inst. Mech. Eng. Part G J. Aerosp. Eng. 2015, 229, 2481–2496.
[CrossRef]
41. Rudra, S.; Barai, R.K.; Maitra, M. Design and implementation of a block-backstepping based tracking control for nonholonomic
wheeled mobile robot. Int. J. Robust Nonlinear Control 2016, 26, 3018–3035. [CrossRef]
42. Abadi, A.; El Amraoui, A.; Mekki, H.; Ramdani, N. Flatness-Based Active Disturbance Rejection Control For a Wheeled Mobile
Robot Subject To Slips and External Environmental Disturbances. IFAC-PapersOnLine 2020, 53, 9571–9576. [CrossRef]
43. Dou, J.X.; Kong, X.X.; Wen, B.C. Backstepping sliding mode active disturbance rejection control of quadrotor attitude and its
stability. J. Northeast. Univ. (Nat. Sci.) 2016, 37, 1415. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.