Control Engineering Practice 40 (2015) 81–92

Contents lists available at ScienceDirect

Control Engineering Practice

journal homepage: www.elsevier.com/locate/conengprac

Design and validation of a robust immersion and invariance controller

for the lateral dynamics of intelligent vehicles$
Gilles Tagne n, Reine Talj, Ali Charara
Heudiasyc Laboratory, Sorbonne universités, Université de technologie de Compiègne, CNRS, Heudiasyc UMR 7253, CS 60 319, 60 203 Compiègne, France

art ic l e i nf o a b s t r a c t

Article history: This paper focuses on the lateral control of intelligent vehicles; the aim is to minimize the lateral
Received 15 May 2014 displacement of the autonomous vehicle with respect to a given reference path. The control input is the
Accepted 6 March 2015 steering angle and the output is the lateral displacement error. We present the design and validation of a
robust lateral controller based on the Immersion and Invariance (I&I) principle in order to ensure robust
Keywords: stability and good performances with respect to parametric variations and uncertainties that are
Lateral control encountered in driving applications. To validate our control law, tests were performed on SCANeR Studio,
Immersion and invariance control a driving simulation engine, according to several real driving scenarios. Simulations were also performed
Reference tracking using experimental data acquired by the DYNA vehicle (a Peugeot 308) belonging to the Heudiasyc
Autonomous vehicles
laboratory. The validation demonstrates the robustness and good performances of the proposed control
Intelligent vehicles
approach and clearly shows the improvement due to the I&I controller.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction cameras, lasers, radars and satellite receivers is usually used to

achieve this goal. It provides a dynamic map of the autonomous
1.1. Motivations vehicle's immediate environment.

Path planning: generating and choosing one trajectory (the
In recent years, technological advances have favoured the reference path) in the navigable space, according to several
emergence of intelligent vehicles. These vehicles have the capacity criteria.
to anticipate and compensate failures (which can be of the driver,
Vehicle control: controlling the vehicle using actuators such as
the vehicle, or the environment) and, more ambitiously, to drive the brakes, the accelerator and the steering wheel so that it
themselves autonomously. adheres to the reference path. Vehicle control can be separated
The “DARPA (Defence Advanced Research Projects Agency) into two subtasks: longitudinal control and lateral control.
Challenges” (2004, 2005, 2007) have provided a stimulus to
research into the development of autonomous vehicles (Siciliano, This paper focuses on the third of these components, that is to say
Khatib, & Groen, 2009), resulting in a growing interest by research vehicle control, and more precisely on lateral control, which is a
laboratories and commercial companies in the development of very active research field dating back to as far as the 1950s.
autonomous driving applications. See Wei et al. (2013) and Lateral control is concerned with steering the vehicle automatically
Bertozzi, Bombini, Broggi, and Buzzoni (2011) for some examples. to follow the reference path. In recent years a lot of research have been
One of the major challenges today for this growing area of devoted to the lateral guidance of autonomous vehicles. Several con-
research is autonomous driving at high speed. trol strategies have been developed in the literature. Simple Propor-
Autonomous driving has three essential components: tional (P) and Proportional-Integral (PI) controllers were proposed in
Broggi, Bertozzi, and Fascioli (1999) and Zhao, Chen, Mei, and Liang

Perception and localization: detecting the dynamic environment (2011) respectively. A nested PID controller was presented in Marino,
of the vehicle, which includes the road, fixed and mobile Scalzi, and Netto (2011). Adaptive controllers have also been developed
obstacles. A vision system composed of sensors such as for this application, such as in Netto, Chaib, and Mammar (2004).
Alternative classical techniques have also been applied: H 1 control in
Hima, Lusseti, Vanholme, Glaser, and Mammar (2011), state feedback
☆
This work was carried out in the framework of Labex MS2T, which was funded control in Rajamani (2006), Lyapunov stability based control in Enache,
by the French Government through the Investments for the future program,
managed by the National Agency for Research (Reference ANR-11-IDEX-0004-02).
Netto, Mammar, and Lusetti (2009), Model Predictive Control (MPC) in
n
Corresponding author. Levinson et al. (2011), Falcone, Borrelli, Tseng, Asgari, and Hrovat
E-mail address: gilles.tagne@hds.utc.fr (G. Tagne). (2008), and Klan and Skrjanc (2007), artificial intelligence and fuzzy

http://dx.doi.org/10.1016/j.conengprac.2015.03.002
0967-0661/& 2015 Elsevier Ltd. All rights reserved.
82 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92

logic in Onieva et al. (2011), and Naranjo, Gonzalez, Garcia, and de reformulation of the stabilization problem is implicit in sliding
Pedro (2008), a fuzzy Takagi–Sugeno LQ controller in Soualmi, mode control (SMC), where the target dynamics are the dynamics
Sentouh, Popieul, and Debernard (2012) and Kladis, Economou, of the system on the sliding surface, which is made attractive by a
Knowles, Lauber, and Guerra (2011), linear quadratic optimal control discontinuous control law. The main difference between I&I and
in Kim, Kang, and Yi (2011) and Menhour, Charara, and Lechner (2014), SMC is that in I&I the manifold does not necessarily have to be
a backstepping-based approach in Nehaoua and Nouvelière (2012), reached, whereas in sliding mode the manifold must be reached in
and a host of other techniques that are simply too numerous to be finite time. The advantage of the I&I principle is the fact that it
listed here. allows a better generalization compared to the sliding mode. It is
Advances in this domain have been the motivation for a more flexible in the choice of the target dynamics and the law of
number of performance comparisons of the proposed controllers, convergence towards this dynamics. It also avoids the use of a
such as in Chaib, Netto, and Mammar (2004), which compared discontinuous term in the control law keeping off the chattering
proportional, adaptive, H 1 and fuzzy controllers. Different com- (which is one of the main drawbacks of the SMC). In addition, the
parisons showed adaptive controllers to be a very promising I&I allows us to take greater account of the model during the
technique for this kind of uncertain non-linear application. Given controller design.
on the one hand the high non-linearity of the vehicle system, and The main contribution of the present paper is the use of the
on the other hand the uncertainties and disturbances of this kind principle of immersion and invariance to design a robust and
of system, robustness is a very important issue in the control adaptive controller which allows us to have comparable or better
design. The controller should be able to tolerate disturbances performances than a SMC controller. A preliminary work related to
(caused by wind, the coefficient of friction of the road, etc.) and the I&I controller design was presented in Talj, Tagne, and Charara
be able to deal with parameter uncertainties and variations. For (2013). In this paper (Talj et al., 2013), we have developed a
example, Levinson et al. (2011) is a recent presentation of controller based on the principle of Immersion and Invariance for
Stanford's autonomous research vehicle ‘Junior’ (second place at autonomous lateral trajectory tracking. This controller is based on
the DARPA Urban Challenge) in the context of robust autonomous a bicycle model with the following state variables: the yaw angle
driving. error, the yaw rate error, the lateral displacement error and its
derivative. During the validation with real data, we noticed that an
1.2. Problem statement and contributions offset appeared in the lateral displacement error after an impor-
tant solicitation of the vehicle. This offset is due to the use of the
Over the years, extensive control strategies based on Sliding yaw angle error in the control input, which is estimated by
Mode Control (SMC) method have been proposed to design better integrating the yaw rate error which is noisy in measurement. To
and more reliable systems for the lateral control of intelligent solve this problem, we have previously added a discontinuous
vehicles. The first-order SMC is used to control the vehicle term of robustness δrob in the control law (as in SMC) to cancel the
with small displacement error by compensating the uncertainties lateral error. This paper proposes a solution to the problem in the
and encountered disturbances (Ackermann, Guldner, Sienel, design of the I&I controller, where other state variables are used to
Steinhauser, & Utkin, 1995; He, Crolla, Levesley, & Manning, represent the vehicle dynamics. The proposed model reformula-
2006a; Hingwe & Tomizuka, 1997; Zheng, Tang, Han, & Zhang, tion avoids the use of the yaw angle by replacing it with the
2006). It also provides experimental results comparable to or sideslip angle. This reformulation decouples the lateral dynamics,
better than linear auto-tuning controllers. This method also which allows us to show that the closed-loop system can be
possesses the advantage of producing low complexity control laws decomposed into two cascaded subsystems. The new proposed I&I
compared to other approaches of robust control. The major draw- controller guarantees the stability of the system for all its positive
backs of the SMC method are the following: (1) it needs knowl- gains. This was not the case in the first version in Talj et al. (2013),
edge about the bounds of the disturbances and uncertainties in where we showed stability using a Lyapunov function. The result
advance, (2) it is not robust outside the sliding surface, and (3) the was restrictive, hiding some interesting characteristics of the
chattering. The chattering phenomenon often leads to undesirable system. Furthermore, we have used a different off-the-manifold
results, for example, high wear of moving mechanical parts and variable z to bring up an integral term in the control law so as to
even passengers' discomfort (Hingwe & Tomizuka, 1997). For those improve the robustness with respect to parametric uncertainties
reasons, it is necessary to control and maintain the amplitude of and disturbances.
oscillations at a low level. Thus, Tagne, Talj, and Charara (2013) To design the controller, we consider that the vehicle is
presented a higher-order SMC method to alleviate the chattering. equipped with sensors and/or observers to measure sideslip angle,
The super-twisting algorithm was utilized to minimize the lateral yaw rate, lateral error and its derivative. To validate the control
displacement with respect to a given reference of an autonomous strategy, the closed-loop system was simulated using SCANeR
vehicle at high speed. In Imine and Madani (2011), the super- Studio (Scaner_url, 2014), a driving simulation engine constructed
twisting algorithm was used to ensure the assistance of active to simulate several real driving scenarios. Simulations were also
direction to heavy vehicles at low speed. performed using experimental data acquired by the DYNA vehicle
The I&I principle is a relatively new method for designing non- (a Peugeot 308) belonging to the Heudiasyc laboratory. Our tests
linear and adaptive controllers. The method is based on system showed that the proposed control approach was robust and that
immersion and manifold invariance. The basic idea of the I&I performances were good.
approach is to achieve the control objective by immersing the
plant dynamics in a (possibly lower-order) target system that
captures the desired behaviour (Astolfi, Karagiannis, & Ortega,
2008). This is achieved by finding a manifold in state-space that 1.3. Paper structure
can be rendered invariant and attractive – with internal dynamics
that reflect the desired closed-loop dynamics – and by designing a This paper is organized as follows. Section 2 presents the
control law that takes the state of the system towards the dynamic models of the vehicle. Section 3 presents the control
manifold. In other words, the I&I theory consists of defining target problem definition, and the design of the I&I controller is pre-
dynamics and designing a control law that renders the manifold of sented in Section 4. Section 5 presents the results. Finally, we
the target dynamics attractive and invariant. Indeed, the I&I conclude in Section 6, with some remarks and prospects.
 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92 83

2. Dynamic vehicle models

2.1. Bicycle model

The controller is based on a simple and widely used dynamic

bicycle model (Rajamani, 2006) (see Fig. 1). This model is used to
represent the lateral vehicle behaviour (lateral acceleration y, € yaw
rate ψ_ , and sideslip angle β). It assumes that the vehicle is
symmetrical and that tire sideslip angles on the same axle are
equal. The roll and pitch dynamics are neglected and angles
(steering δ, sideslip β, yaw ψ) are assumed to be small. With a
linear tire force model we obtain a linear parameter varying (LPV)
model, where the longitudinal velocity Vx is considered as a
varying parameter. Dynamic equations in terms of sideslip angle
and yaw rate of the bicycle model are given by Fig. 2. 4-Wheels vehicle model.
8 !
> β_ ¼  μðC f þ C r Þ β  1 þ μðLf C f Lr C r Þ ψ_ þ μC f δ
>
>
>
> !
< mV x mV 2x mV x 1 X 4
ψ€ ¼ €
M þ I xz ϕ
ð1Þ I z i ¼ 1 zi v
>
> μðLf C f  Lr C r Þ μðLf C f þLr C r Þ
2 2
μLf C f
>
>
>
: ψ
€ ¼  β  ψ
_ þ δ !
I I V I 1 X 4
ϕ€ v ¼
z z x z
M xi  ms hðy€ þ ψ_ V x Þ þ I xz ψ€
I xx i ¼ 1
where β, ψ and δ represent the sideslip angle, the yaw angle of the
vehicle and the steering wheel angle (control input) respectively. ax ¼ V_x  V y ψ_
Table 1 presents the vehicle's parameters and nomenclature.
ay ¼ V_y þ V x ψ_ ð2Þ

2.2. 4-Wheels model where

X
2
For simulation with real data, we used a more representative F xij ¼ F x11 þ F x12 þ F x21 þ F x22 ;
model, namely the 4-wheels model to represent the behaviour of i ¼ 1;j ¼ 1

the vehicle and Dugoff's tire model for longitudinal and lateral tire X
2

forces. Fig. 2 presents the 4-wheels model. The equations of the F yij ¼ F y11 þ F y12 þ F y21 þF y22 ;
i ¼ 1;j ¼ 1
different dynamics (longitudinal, lateral, yaw and roll) used here
are given as follows (Kiencke & Nielsen, 2005; Jazar, 2008):
0 1 X
4
E
M zi ¼ ½ðF x11 þ F x21 Þ  ðF x12 þ F x22 Þ
_ 1 @X 2 X 2
2
V x ¼ ψ_ V y þ F  F aero þms hψ_ ϕv A i¼1
m i ¼ 1 j ¼ 1 xij þ Lf ðF y11 þ F y12 Þ  Lr ðF y21 þ F y22 Þ;
0 1
_ 1 @X 2 X 2
€ X
V y ¼  ψ_ V x þ F þ F bank ms hψ ϕ v A
4
m i ¼ 1 j ¼ 1 yij _ ;
M xi ¼ ½ms gh  ðK ϕf þ K ϕr Þϕv ðC ϕf þ C ϕr Þϕ v
i¼1

F aero ¼ 12 ρSC x V 2x ;
F bank ¼ mg sin ðϕr Þ:

with ax and ay being the longitudinal and lateral accelerations

respectively; Vx and Vy the longitudinal and lateral velocities
respectively; Fxij and Fyij the longitudinal and lateral tire forces
respectively where i ¼ 1; 2 for front/rear and j ¼ 1; 2 for left/right;
Ixx the vehicle moment of inertia of the sprung mass with respect
to the roll axis; Ixz the roll/yaw moment of inertia; E the vehicle
track; g the gravitational constant; h the height of the centre of
gravity; ms the sprung mass; ϕv the roll angle; K ϕf and K ϕr the
Fig. 1. Bicycle model. front and rear roll frictional coefficients respectively; C ϕf and C ϕr
the front and the rear roll stiffness respectively; ϕr the road bank
angle; Cx the aerodynamic drag coefficient; S the frontal area of the
vehicle and ρ the mass density of the air.
Table 1
Vehicle parameters and terminology (bicycle model).
The vertical forces at each tire are calculated by the equations
Vx Longitudinal velocity – (m/s)    
β Sideslip angle – (rad) 1 Lr h Lr h h
F z11 ¼ m g  ax  m g  ax ay
ψ_ Yaw rate – (rad/s) 2 L L L L Eg
δ Steering wheel angle – (rad)    
μ –
1 Lr h Lr h h
Road friction coefficient 1 F z12 ¼ m g  ax þ m g  ax ay
m Mass 1421 (kg) 2 L L L L Eg
Iz Yaw moment of inertia 2570 (kg m2)    
1 Lf h Lf h h
Lf Front axle-COG distance 1.195 (m) F z21 ¼ m g þ ax m g þ ax ay
Lr Rear axle-COG distance 1.513 (m) 2 L L L L Eg
   
Cf Cornering stiffness of the front tire 170,550 (N/rad) 1 Lf h Lf h h
Cr Cornering stiffness of the rear tire 137,844 (N/rad) F z22 ¼ m g þ ax þm g þ ax ay ð3Þ
2 L L L L Eg
84 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92

To determine the lateral and longitudinal forces of each tire, we where

used Dugoff's tire model (Dugoff, Fancher, & Segel, 1970) given by 2 μðC f þ C r Þ μðLf C f  Lr C r Þ 3
8  1 0 0
> σx 6
mV x mV 2x
7
> F x ¼  C σ 1 þ σ x f ðλ d Þ
>
>
6
6 μðLf C f  Lr C r Þ μðL2f C f þ L2r C r Þ 7
7
>
> A¼6   0 07
>
> tan ðαÞ 6 Iz Iz V x 7
>
> 6 7
>
> F y ¼  Cα f ðλd Þ 4  μðC fmþ C r Þ 
μðLf C f  Lr C r Þ
05
>
> 1 þ σx mV x 0
< (
ð2  λd Þλd for λd o 1; ð4Þ 0 0 1 0
>
> f ðλ Þ ¼
>
> 1 for λd Z 1 2 μC 3
>
> 2 3
>
f
>
> μF z ð1 þ σ x Þ 6 mV x 7
0
>
> λd ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 6 μ Lf C f 7 6 0 7
>
> 6 7 6 7
>
: 2 ðC σ σ x Þ2 þðC α tan ðαÞÞ2 B1 ¼ 6 Iz 7; B2 ¼ 6 7 ð10Þ
6 μC f 7 4  V 2x 5
4 m 5
where C σ and C α are the longitudinal and the cornering stiffness of 0
0
the tire respectively, α the tire slip angle, σx the longitudinal slip
ratio, μ the tire/road friction coefficient, and Fz the vertical load on The control input is the steering angle and the output is the lateral
the tire. displacement error. Ideally, the lateral control will eliminate the lateral
The tire slip angle of each tire αij is given by (Kiencke & Nielsen, error. Then, for a given curvature ρ and longitudinal velocity Vx, the
2005) desired behaviour corresponds to e_ ¼ e ¼ 0. Hence, it is easy to prove
0 1 that the desired equilibrium point is (Rajamani, 2006)
BV y þ Lf ψ_ C
⋆
ðβ ; ψ_ ; e;
_ eÞ > ¼ ðβ ; ψ_ ⋆ ; 0; 0Þ >
α11 ¼ δ11  arctan@ A;
E
V x  ψ_ with
0 2 1 !
Lf mV 2x
V y þ Lf ψ_ C β ⋆ ¼ Lr  ρ
α12 ¼ δ12  arctanB
@ A; μC r ðLf þ Lr Þ
E
V x þ ψ_
21 ψ_ ⋆ ¼ V x ρ ð11Þ
0
V y  Lr ψ_ C At the equilibrium point, the control input is
α21 ¼  arctanB
@ A;
E 2 2
V x  ψ_ Lf C f  Lr C r ⋆ Lf C f þ Lr C r ⋆
0 2 1 δ⋆ ¼ β þ ψ_ ð12Þ
Lf C f Lf C f V x
V y  Lr ψ_ C
α22 ¼  arctanB
@ A: ð5Þ We define the new error variables:
E 8
V x þ ψ_ > ~ ⋆
2 < β ¼ β β
>
Lastly, the longitudinal slip ratio is defined as >
ψ~_ ¼ ψ_  ψ_ ⋆ ð13Þ
8 >
:~
> Reff ω  V ωx δ ¼ δ  δ⋆
>
> during acceleration;
< Reff ω The error dynamics of the system (9) whose origin is the equilibrium
σx ¼ ð6Þ
>
>
>
Reff ω  V ωx point ðβ~ ; ψ~_ ; e;
_ eÞ > ¼ ð0; 0; 0; 0Þ > become
: during braking:
V ωx
x~_ ¼ Ax~ þ B1 δ~ ð14Þ
with V ωx being the actual longitudinal velocity at the wheel axle; where A and B1 have been defined above in (10).
Reff ω the equivalent rotational velocity of the tire, where ω is the
We will now use the vehicle lateral error model (14) to design
wheel rotational velocity and Reff the effective tire radius. the lateral controller, with δ~ as the control input.

3. Control problem definition 4. Immersion and Invariance (I&I) controller design

The lateral control's aim is to minimize the lateral displacement In this section the description of the I&I principle, and the
of the autonomous vehicle with respect to a given reference path. design of the I&I controller are presented.
The lateral error dynamics at the centre of gravity of the vehicle,
with respect to a reference path, are given by 4.1. I&I main principle
e€ ¼ ay  ayref ð7Þ
The developed controller is based on the following theorem,
where ay and ayref represent respectively the lateral acceleration of representing the main stabilization result of the Immersion and
the vehicle and the desired lateral acceleration along the reference Invariance method.
path. Assuming that the desired lateral acceleration of the vehicle
can be written as ayref ¼ V 2x ρ, where ρ is the curvature of the road Theorem 1 (Astolfi et al., 2008). Consider the system
which we assume constant. x_ ¼ f ðxÞ þ gðxÞu; ð15Þ
Given that ay ¼ V x ðβ_ þ ψ_ Þ (Rajamani, 2006), we have n m ⋆
with x A R , u A R , and an equilibrium point x to be stabilized.
e€ ¼ V x ðβ_ þ ψ_ Þ  V 2x ρ ð8Þ Assume that there exist smooth mappings α : Rp -Rp , π : Rp -Rn ,
Adding the equation of the lateral error dynamics (8) in the system ϕ : Rn -Rn  p , c : Rp -Rm and v : Rnxðn  pÞ -Rm , with p o n, such
(1), the new system state variables become x ¼ ðβ ; ψ_ ; e; _ eÞ > , that the following hold.
corresponding to the sideslip angle, the yaw rate, the lateral error

(A1) The target system
and its derivative. The system has the following dynamics:
x_ ¼ Ax þ B1 δ þ B2 ρ ð9Þ ξ_ ¼ αðξÞ; ð16Þ
 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92 85

with ξ A Rp has a globally asymptotically stable equilibrium at equilibrium at the origin. The system (14) is I&Istabilizable with
ξ⋆ A Rp and target dynamics (23).
⋆
x⋆ ¼ π ðξ Þ: ð17Þ
Proof. We now define the off-the-manifold variable

(A2) For all ξ A Rp , Z T
∂π z ¼ e_ þ λ1 e þ λ2 e dt; s:t: λ1 4 0; λ2 4 0 ð26Þ
f ðπ ðξÞÞ þ gðπ ðξÞÞcðπ ðξÞÞÞ ¼ αðξÞ: ð18Þ
∂ξ 0

We have to select a control input δ~ such that the trajectories of the

(A3) The set identity closed-loop system remain bounded and z converges to zero. It
fx A Rn j ϕðxÞ ¼ 0g ¼ fx A Rn j x ¼ π ðξÞ; ξ A Rp g ð19Þ should be remarked that when z-0, e converges asymptotically to
holds. zero, yielding the desired result. This is analogous to the principle

(A4) All trajectories of the system of SMC, in which the system trajectories are constrained so as to
∂ϕ reach the sliding surface s in finite time, and to remain on it. In
z_ ¼ ðf ðxÞ þ gðxÞvðx; zÞÞ; ð20Þ
∂x SMC a convergence in finite time is guaranteed by a discontinuous
control law. In the I&I controller, however, convergence might be
x_ ¼ f ðxÞ þ gðxÞvðx; zÞ; ð21Þ asymptotic. As a way of addressing this problem, let
are bounded and (20) has a uniformly globally asymptotically z_ ¼  Kz with K 40; ð27Þ
stable equilibrium at z¼ 0.
where the scalar K represents the rate of exponential convergence
Then x⋆ is a globally asymptotically stable equilibrium of the closed- of z to zero. The off-the-manifold variable z converges to 0, e and e_
loop system: converge to 0. The system is thus immersed in the target dynamics
then converges to the origin.
x_ ¼ f ðxÞ þ gðxÞvðx; ϕðxÞÞ: ð22Þ
Given the principle of the controller, it will be remarked that
there are two manifolds. First, an outer manifold M1 is reached
when the off-the-manifold variable z converges exponentially to
Any trajectory x(t) of the closed-loop system x_ ¼ f ðxÞ þ
zero with the rate of convergence K 4 0. Then, an inner manifold
gðxÞvðx; ϕðxÞÞ is the image through the mapping π ðÞ of a trajectory
M2 can be reached from any point in M1 with the off-the-
ξðtÞ of the target system. Note that the mapping π : ξ-x is an
manifold variable e that converges to zero.
immersion, i.e., the rank of π is equal to the dimension of ξ. The
Replacing z_ and z by their expressions in (27), and after some
approach consists in applying a control law that renders the manifold
calculations, the corresponding control input can be expressed as
x ¼ π ðξÞ attractive and keeps the closed-loop trajectories bounded.
follows:
Z
4.2. Application to lateral control mðK þ λ1 Þ mðK λ1 þ λ2 Þ mK λ2 T
δ~ ¼  e_  e e dt
μC f μC f μC f 0
Consider the vehicle lateral dynamics model (14). The main C f þ C r ~ Lf C f  Lr C r _
objective of the steering controller is to eliminate the lateral þ βþ ψ~ ð28Þ
Cf Cf V x
displacement error with respect to a given path, so that e ¼ e_ ¼ 0
at the equilibrium.
The target system is consequently chosen to be the image of the Defining
system (14) when e ¼ e_ ¼ 0. More precisely, the target state vector Z T
is ðξ1 ; ξ2 Þ, where ξ1 and ξ2 represent β~ and ψ~_ respectively, when u~ ¼ ðK þ λ1 Þe_ þ ðK λ1 þ λ2 Þe þ K λ2 e dt; ð29Þ
e ¼ e_ ¼ 0. Note that for e ¼ e_ ¼ 0, we also have e€ ¼ 0. The dynamics 0

can be expressed as follows: the closed-loop system becomes

"_ # " #" # " # " # 2 3
ξ1 0 1 ξ1 _
β~ ¼ C β~ þ 4  V x 5u~
1
¼ μC r ðLf þ Lr Þ μLr C r ðLf þ Lr Þ ð23Þ ð30Þ
_ξ  ξ2
ψ~_
L m
2 Iz Iz V x
ψ~_  fIz

Proposition 1. The target model (23) has a globally asymptotically

stable equilibrium at the origin ð0; 0Þ. The closed-loop system (30) can be interpreted as the inter-
connection of two subsystems: S1 with the state variables β~ and ψ~_
Proof. The dynamics of the state vector ξ ¼ ½ξ1 ; ξ2  can be written (23), and S2 with the state variables e, _ e and the integral of e (29).
in the form ξ_ ¼ Cξ, with The subsystem S2 combines and represents the interactions
" #
0 1 between the dynamics of the two off-the-manifold variables z and e,
C ¼ μC r ðLf þ Lr Þ  μLr C r ðLf þ Lr Þ ð24Þ which ensure convergence to the manifolds M1 and M2 respectively.
Iz Iz V x
The lateral error dynamics in closed-loop does not depend on
Then, after some simple calculations, we obtain the variables β~ and ψ~_ , it is given by
μLr C r ðLf þ Lr Þ μC r ðLf þ Lr Þ Z T
detðsI  CÞ ¼ s2 þ sþ ð25Þ e€ ¼  ðK þ λ1 Þe_  ðK λ1 þ λ2 Þe  K λ2 e dt ð31Þ
Iz V x Iz 0

For any uncertainties and variations encountered, the model's

parameters (C f ; C r ; Lf ; Lr ; I z ; μ) are always positive. The matrix C Proposition 3. The subsystem S2 has a globally asymptotically
satisfies the Routh–Hurwitz stability criterion (V x 4 0), which stable equilibrium at the origin.
leads to the desired result. □
Proof. The dynamics of the subsystem S2 have the characteristic
Proposition 2. Consider the system (14) whose equilibrium point is equation:
at the origin. Moreover, the subsystem (23), which is the image of the
system (14) for e ¼ e_ ¼ 0, has a globally asymptotically stable PðsÞ ¼ s3 þðK þ λ1 Þs2 þ ðK λ1 þ λ2 Þs þ K λ2 ð32Þ
86 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92

This equation satisfies the Routh–Hurwitz stability criterion for all 4.3. Analysis of the target dynamics
the gains λ1 4 0, λ2 4 0 and K 4 0, and hence the attractivity of
both manifolds is ensured. Analysing the target dynamics reveals the behaviour of the
Finally it is clear that when S2 converges to zero, the input u~ of internal dynamics. If modification is necessary, this can then be
the subsystem S1 converges to zero. The system then converges to achieved by acting on the command.
the target subsystem, whose dynamics (23) are stable and con- The target dynamics (23) is quite simply the dynamics of the
verge to ð0; 0Þ. We have already proven above that this target sideslip angle and yaw rate with the curvature ρ as input. These
model has a globally asymptotically stable equilibrium at the dynamics are given by the following equation:
origin ð0; 0Þ. Consequently, the trajectories of the closed-loop " # " # 2 3
β_ β Vx
system are bounded, yielding the desired result. For more details ¼C þ 4 Lf mV 2x 5ρ ð34Þ
on the proof of global asymptotic stability of the closed-loop ψ€ ψ_ I z

system, see Kokotovic and Marino (1986), where it is shown that

the cascade interconnection of Global Asymptotic Stable (GAS) where, for a given curvature ρ, β and ψ_ converge respectively to
systems with a bounded trajectory gives a globally asymptotically β⋆ and ψ_ ⋆ . In this section we study the response of the transfer
stable system. □ function between the input (the curvature) and the output (the
dynamic variables). We will see the influence that the speed, the
The proposed I&I controller guarantees the stability of the type of road, and the mass of the vehicle have on the response of
system, for all its gains λ1 4 0, λ2 4 0 and K 40. This is a very the system in the time domain.
strong result that reveals some interesting characteristics of the The transfer functions of β ðsÞ and ψ_ ðsÞ with respect to the input
system. The parameters K, λ1 and λ2 have to be chosen based on ρ are second-order functions with a relative degree of 1. ψ_ ðsÞ=ρðsÞ
other practical considerations related to the system performance. is minimum phase for all possible parameter variations. β ðsÞ=ρðsÞ is
Finally, the control input applied to the system (9) is minimum phase for all V x oV 0 with
Z sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
mðK þ λ1 Þ mðK λ1 þ λ2 Þ mK λ2 T μLr C r ðLf þ Lr Þ
δ ¼ δ~ þ δ⋆ ¼  e_  e e dt V0 ¼ ð35Þ
μC f μC f μC f 0 Lf m

Cf þ Cr L C L C mV 2 Using the parameters given in Table 1, the speed limit of Eq. (35)
þ β þ f f r r ψ_ þ x ρ ð33Þ corresponds to V x o 16:5 m=s C 60 km=h.
Cf Cf V x μC f
Figs. 3–5 are the step responses to a step of curvature
This I&I control law can be interpreted as a dynamic state ρ ¼ 0:002 m  1 , with zero initial condition. Fig. 3 shows the step
feedback controller plus a PID controller (with gains depending on responses of the slip angle and yaw rate for velocity variations
system parameters), for robustness. The dynamic state feedback ranging from 5 m/s to 40 m/s, on a dry road (μ ¼ 1). For V x 4 V 0 ,
provides a sort of compensation in order to force the system to we note a large variation in damping (decrease). Responses βðtÞ
reach the desired state. On the other hand, the PID controller helps and ψ_ ðtÞ have very similar dynamic behaviours (damping oscilla-
us to ensure convergence and to cancel the effect of uncertainties tion, response time). At low speed, vehicle dynamics behave well,
and/or disturbances. To apply our control law, the yaw rate, lateral without overshoots or oscillations. When speed increases, or road
error and its derivative are measured, and the sideslip angle is adhesion decreases (Fig. 4), step responses tend to have large
estimated. overshoots and oscillations. The increased tendency to overshoot
In the autonomous driving applications, the lateral stability of and oscillate indicates the increasing influence of zeros and the
the vehicle is already considered in the trajectory planning step decrease in damping of poles. The variation of road adhesion
which is an input to the control design. Hence, driving in linear affects the amplitude of the oscillations and the final value of β ðtÞ.
regions is guaranteed by the choice of the trajectory. It is This study shows that the target dynamics remains stable,
important to underline that we use the bicycle model to design whatever the parameter variations are.
the controller. However, the controller is designed to deal with
some model errors with the presence of the robust integral term in
5. Results
the off-the-manifold variable z. The structure of the controller
guarantees a convergence to a near neighbourhood of the desired
The proposed controller is validated by simulation on SCANeR
state, even though the model presents some weaknesses to Studio (Scaner_url, 2014). Simulations were also performed using
represent the system in some regions.

Fig. 3. Step responses: variations of speed. Fig. 4. Step responses: variations of road adhesion for V x ¼ 20 m=s.
 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92 87

experimental data acquired by the DYNA vehicle (a Peugeot 308). closed-loop system is no more guaranteed and the performances
The two environments are complementary and are used to assess of the SMC can also be significantly deteriorated.
the performance of the controllers in different driving scenarios. Considering the sliding surface variable s ¼ e_ þ λe, the equiva-
For the control law, we used the gains λ1 ¼ 8, λ2 ¼ 2 and K ¼2, lent term δeq corresponding to s_ ¼ 0 is
with the nominal vehicle parameters (see Table 1). To highlight the
Cf þ Cr L C L C mV 2 mλ
improvements achieved by the I&I controller, we compared it with δeq ¼ β þ f f r r ψ_ þ x ρ  e_ ð39Þ
a previously developed higher order sliding mode controller Cf Cf V x Cf Cf
(Tagne et al., 2013). The control input of the SMC controller is We observe that both control laws are very similar and can be
given by interpreted as linear/non-linear PIDs with a linear equivalent term
δSMC ¼ δST þ δeq ð36Þ (state feedback). For the SMC controller, we used the gains
λ ¼ 8; α1 ¼ 0:008; α2 ¼ 0:008. Note that the same measures are
where δST is the super-twisting algorithm: used for both controllers.
(
u1 ¼  α1 jsj1=2 signðsÞ
δST ¼ u1 þu2 _ ð37Þ
u 2 ¼  α2 signðsÞ 5.1. Validation on SCANeR™ Studio

with α1 and α2 being positive constants. The finite time conver-

To validate our control law, tests were performed on SCANeR
gence to the sliding surface is guaranteed for all:
Studio (Scaner_url, 2014), a driving simulation engine, correspond-
8
> C ing to several real driving scenarios on the track circuit given by
>
>
> α2 4 0
< bmin Fig. 6. This simulator uses the vehicle full model (a more repre-
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð38Þ
4C 0 ðbmax α2 þ C 0 Þ sentative model) to represent the vehicle dynamics and behaviour.
>
>
>
> α 1 Z The tests were performed on a road with a curvature varying
: b ðbmin α2 C 0 Þ
2
min
between  0.02 and þ0:02 m  1 . Longitudinal speed varied
These conditions (38) depend on the maximum considered mag- between 5 m/s and 30 m/s (see Fig. 7). A number of tests of
nitude (C 0 ; bmin ; bmax ) of the system's variables (due to noises). If robustness with respect to parametric uncertainties and variations
these conditions are no more verified, the convergence of the were carried out to show the performance of the controller.

5.1.1. Robustness of the controller to speed and curvature variations

Fig. 8 presents the controllers in normal driving conditions
with μ ¼ 1. The path followed by the controlled vehicle is shown
for both controllers, along with the errors, i.e., the lateral deviation
and the yaw rate error. These non-linear controllers are robust
against speed variations. The lateral error remains small for Vx
variations. We observe that the SMC and the I&I controllers can
ensure the convergence of the lateral error. The convergence time
of the SMC controller is lower than that of the I&I. Both controllers
have comparable maximum errors and ensure that the path is
followed. The displacement from the guideline does not exceed
10 cm in this test conditions. We observe a peak in the lateral error
around t ¼25 s. This is due to a large variation of the curvature
with the vehicle at a high speed (see Fig. 7). This allows us to test
the robustness of the controller against abrupt changes of curva-
ture at variable high speed.
Fig. 9 presents the steering angle of both controllers. The I&I
controller provides a continuous steering angle while avoiding
Fig. 5. Step responses: variations of mass for V x ¼ 20 m=s. chattering.

Fig. 6. Test 1: track circuit.

88 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92

This test therefore shows the good performance of both control important to assess the robustness of the controller against
strategies during normal driving (at high and varying speed with cornering stiffness uncertainties. Fig. 10 presents lateral errors
varying curvature) when nominal parameters are known. The I&I for different values of cornering stiffness. For uncertainties of the
controller provides the same performance while having a order of 710% regarding the value of cornering stiffness, both
smoother steering angle. controllers are able to follow the path with acceptable errors (for
safe driving the maximum error should not exceed 715 cm). The
SMC controller has small errors, similar to nominal conditions
5.1.2. Robustness of the controller to vehicle parameter uncertainties
where curvature is constant. However, for large variations in
The test described above shows the performance of the con-
curvature, the I&I controller is more robust. The peaks of the I&I
troller during normal driving. For trajectory tracking, one of the
are lower than those of the SMC controller for large curvatures at
major challenges is the robustness of the controller against
high speed (around t¼25 s). The behaviour of the closed-loop
uncertainties in the vehicle's parameters or the driving environ-
system with SMC can be described by two phases (Utkin, 1992):
ment. We next evaluate the performance of the controller against
parameter uncertainties.
The convergence phase: this phase corresponds to the time
It is difficult to estimate accurately the cornering stiffness of the
interval during which the system state trajectories are not on
tire. Moreover, this parameter varies significantly according to the
the sliding surface. In this phase, the system is sensitive to
type of road, the vertical load, the camber, etc. It is therefore
changes in parameters.

The sliding phase: this phase corresponds to the time interval
during which the state trajectories are confined throughout

Fig. 7. Test 1: road informations and longitudinal speed. Fig. 9. Test 1: steering angle.

Fig. 8. Test 1: paths; reference and control laws.

 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92 89

sliding surface. The sliding system has the property of insensi- conditions of the SMC are respected as it is the case in this
tivity with respect to disturbances. validation (small and almost constant errors of the SMC controller
during the steady state), the I&I approach provides several
improvements. It provides smaller transient errors and a smoother
In fact, the system with the SMC controller is insensitive to
steering angle.
disturbances only in sliding phase, but remains sensitive for the
transient phase, i.e., before the sliding surface is reached. This is
the reason why the behaviour of the system with the SMC 5.2. Validation using experimental data
controller is not predictable for the transient phase and depends
on the level of disturbances (see Figs. 10 and 11). However, during For the simulation with real data, we have used the model
steady state this controller provides a lower error. presented in Section 2.2, namely the 4-wheels model to represent
With the I&I controller, the error depends on the value of the behaviour of the vehicle and Dugoff's tire model for long-
parametric uncertainty and remains acceptable even for large itudinal and lateral tire forces. Simulations have also been per-
variations. We note that large variations in the parameters formed using experimental data acquired by the DYNA vehicle, a
diminish performance, but that stability is maintained. Peugeot 308 (Fig. 12). The experimental vehicle is equipped with
The mass of the vehicle may vary or be poorly estimated. It several sensors: an Inertial Measurement Unit (IMU) measuring
depends on the number of passengers in the vehicle and the accelerations (x, y, z) and the yaw rate, CORREVIT measuring the
amount of fuel in the tank. We have performed a test to evaluate sideslip angle and longitudinal velocity, torque hubs measuring
the robustness of the control law against the uncertainties on the tire-road efforts and vertical loads on each tire, four laser sensors
vehicle mass (Fig. 11). For uncertainties of the order of 710% of measuring the height of the chassis, GPS and a CCD camera. Data
the mass, controller errors were seen to remain acceptable. provided via the CAN bus of the vehicle were also used, including
Finally, it is well known that sliding mode control (SMC) is data on the steering angle and the rotational speed of the wheels.
robust against disturbances if the matching condition is satisfied. Fig. 13 shows the scheme of validation in simulation using real
Since the vehicle's lateral dynamics to be controlled do not satisfy data:
this condition, the behaviour of the system is not only governed by
sliding surface, but also determined by mismatched disturbances.
The first step is to acquire the data. A human driver drives the
It is important to underline that this drawback of SMC also shows experimental vehicle according to well-defined scenarios. Dur-
the benefit provided by the I&I controller. Even when the ing this step, the dynamic variables of the vehicle are stored.
These actual data are used to develop the reference trajectory.
Measurements of the yaw rate and lateral acceleration are used
as references for computing the errors.

Fig. 10. Test 1: robustness against uncertainties of cornering stiffness.

Fig. 12. Experimental vehicle (DYNA).

Fig. 11. Test 1: robustness against uncertainties of vehicle mass. Fig. 13. Validation scheme.
90 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92

Fig. 14. CERAM test circuit.

The second step is to test the closed-loop controller with the

full vehicle model to follow the reference trajectory.

The experimental data used here were acquired on the CERAM1

test circuits (Fig. 14). The following tests complement the previous
ones described above. Note that in these new scenarios we will
have wide bends and large speed variations that test the capacity
of the controllers to ensure proper path tracking for hard man-
oeuvres. All the parameters of the vehicle are considered to
be known.
In the test shown in Fig. 15, lateral acceleration is less than
4 m/s2. Longitudinal velocity is almost constant (13.5 m/s). Fig. 15
(a) shows the reference path and the paths followed by the
controlled vehicle (with both controllers). Fig. 15(b) presents the
longitudinal speed and errors; the lateral error and the yaw rate
error. Fig. 15(c) presents the vehicle's dynamic variables; the
steering angle, the yaw rate and the lateral acceleration. We
compare real data with data given by the simulated closed-loop
system. Dynamic variables are very close to measured ones. To
assess the robustness of controllers with respect to strong non-
linearities we computed the stability index given by (He, Crolla,
Levesley, & Manning, 2006b)
 
1 4 
SI ¼  β_ þ β  ð40Þ
24 24

When SI o1 the vehicle is in a stable region.

The test shown in Fig. 16 was carried out with the goal of
verifying the robustness of the controller during normal driving at
high and varying speed. Longitudinal speed varies between 5 m/s
and 25 m/s. Note that the maximal lateral acceleration is 5 m/s2.
Although in this scenario we have some manoeuvres at low speed
(wide cornering) and at high speed, the lateral displacement is
smaller than 10 cm in this test condition.
In these last two scenarios, the assumptions were not always
respected; low angles (the steering angle is greater than 151
during a bend) and non-linear forces. Indeed, several tests solicit-
ing the vehicle in non-linear zones have been done (SI 4 1).
However, the controller provides a robust trajectory tracking. Both
non-linear controllers are robust to variations in speed. These tests
therefore demonstrate the good performance of both strategies
during normal driving at high and varying speed.
According to the validation on SCANeR Studio and the valida-
tion using experimental data, we observe that the SMC controller
produces small errors for small values of parametric uncertainty at
constant curvature during the steady state. Performances of the Fig. 15. Test 2: (a) Paths, (b) longitudinal speed and errors, and (c) dynamics
variables.
SMC controller diminish significantly for large curvature variations
and for large parametric uncertainties. If the values of the SMC
controller gains are increased in order to reduce the maximum might wish to use a lower frequency, which would have the effect
errors for large parametric uncertainties, this will increase of the of diminishing the performance of SMC.
chattering to an excessive level. Furthermore, we use a steering With the I&I controller, the error depends on the value of
actuator with a frequency of 10 Hz. For reasons of comfort we parametric uncertainty and remains acceptable even for large
variations. Therefore with sensitivity studies, we can estimate in
1
CERAM - “Centre d'Essais et de Recherche Automobile de Mortefontaine” is advance the maximum error according to the knowledge of the
an automotive testing and research centre located in France. parametric uncertainty. We note that large variations in the
 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92 91

significant changes of curvature. The improvement due to the I&I

controller was verified out of simulation results. With this new
controller, we obtain performances that are similar to or better
than SMC. In addition, we obtain the kind of robust stability of the
controlled system that is required for high speed safe driving.
Although the conditions of the SMC are respected, we see the
improvements brought by the new strategy.

6. Conclusion

In Tagne et al. (2013), we developed a controller based on

higher-order sliding mode control (SMC). Given the implicit
resemblance between SMC and the Immersion and Invariance
(I&I) principle, we have developed a new controller based on the
I&I approach intended to overcome the drawbacks of SMC.
Chattering discontinuity was to be reduced or eliminated, greater
account was to be taken of the structure of the model, and
sensitivity to actuator dynamics was to be lessened.
The design of the proposed I&I controller is presented in this
paper. Our controller guarantees the stability of the system, for all
its gains λ1 4 0, λ2 4 0 and K 40. This is a very strong result that
reveals some interesting characteristics of the system. A study of
the passivity characteristics of all the system variables and their
importance in designing a robust controller is about to be under-
taken by the authors.
A validation and comparison with the SMC controller was
performed according to several scenarios representing different
driving situations. The different tests performed highlighted the
robustness of the developed control law. The robustness of the
controlled system was tested with respect to speed, curvature
variations and parametric uncertainties/variations (cornering stiff-
ness and vehicle mass). The SMC controller gives better perfor-
mances for small parameter variations, when the condition of
convergence in finite time is satisfied. This condition depends on
the value of the system parameters and noise. It is therefore
difficult to tune this controller. The design of the I&I controller
allowed us to prove that the stability of this controller is guaran-
teed for all positive gains (λ1 4 0, λ2 4 0 and K 4 0). This controller
provides equivalent performances without chattering. By taking
greater account of the system, control by I&I can clearly provide a
significant improvement over the performances of SMC for auton-
omous driving applications and ADAS.

References

Ackermann, J., Guldner, J., Sienel, W., Steinhauser, R., & Utkin, V. I. (1995). Linear
and nonlinear controller design for robust automatic steering. IEEE Transactions
on Control Systems Technology, 3(March (1)), 132–143.
Astolfi, A., Karagiannis, D., & Ortega, R. (2008). Nonlinear and adaptive control with
applications. London: Springer.
Bertozzi, M., Bombini, L., Broggi, A., Buzzoni, M., Cardarelli, E., Cattani, S., et al.
(2011). VIAC: An out of ordinary experiment. In International IEEE conference on
intelligent vehicles symposium (IV) (pp. 175–180), Baden-Baden.
Broggi, A., Bertozzi, M., & Fascioli, A. (1999). The ARGO autonomous vehicle's vision
and control systems. International Journal of Intelligent Control and Systems, 3(4),
409–441.
Chaib, S., Netto, M., & Mammar, S. (2004). H inf, adaptive, PID and fuzzy control: A
comparison of controllers for vehicle lane keeping. In International IEEE
intelligent vehicles symposium (IV) (pp. 139–144).
Dugoff, H., Fancher, P., & Segel, L. (1970). An analysis of tire traction properties and
Fig. 16. Test 3: (a) Paths, (b) Longitudinal speed and errors, and (c) dynamics
their influence on vehicle dynamic performance. SAE.
variables.
Enache, N. M., Netto, M., Mammar, S., & Lusetti, B. (2009). Driver steering assistance
for lane departure avoidance. Control Engineering Practice, 17(6), 642–651.
Falcone, P., Borrelli, F., Tseng, H. E., Asgari, J., & Hrovat, D. (2008). Linear time
parameters reduce performance, but that stability is maintained. varying model predictive control and its application to active steering systems:
This clearly shows that we have a robust stability that does not Stability analysis and experimental validation. International Journal of Robust and
depend on the value of the system parameters. This controller Nonlinear Control, 18, 862–875.
He, J., Crolla, D. A., Levesley, M. C., & Manning, W. J. (2006a). Coordination of active
yields acceptable errors for large parametric uncertainties and steering, driveline, and braking for integrated vehicle dynamics control. Journal
guarantees a comfortable ride (no chattering). It is more robust to of Automobile Engineering, 220(January (10)), 1401–1420.
92 G. Tagne et al. / Control Engineering Practice 40 (2015) 81–92

He, J., Crolla, D. A., Levesley, M. C., & Manning, W. J. (2006b). Coordination of active Naranjo, J. E., Gonzalez, C., Garcia, R., & de Pedro, T. (2008). Lane-change fuzzy
steering, driveline, and braking for integrated vehicle dynamics control. control in autonomous vehicles for the overtaking maneuver. IEEE Transactions
Proceedings of the Institution of Mechanical Engineers, Part D: Journal of on Intelligent Transportation Systems, 9(September (3)), 438–450.
Automobile Engineering, 220, 1401–1421. Nehaoua, L., & Nouvelière, L. (2012). Backstepping based approach for the combined
Hima, S., Lusseti, B., Vanholme, B., Glaser, S., & Mammar, S. (2011). Trajectory longitudinal-lateral vehicle control. In International IEEE conference on intelli-
tracking for highly automated passenger vehicles. In International federation of gent vehicles symposium (IV) (pp. 395–400), Alcalá de Henares.
automatic control (IFAC) world congress (pp. 12958–12963), Milano. Netto, M., Chaib, S., & Mammar, S. (2004). Lateral adaptive control for vehicle lane
Hingwe, P., & Tomizuka, M. (1997). Experimental evaluation of a chatter free sliding keeping. In American control conference (ACC) (Vol. 3, pp. 2693–2698).
mode control for lateral control in AHS. In American control conference (ACC) Onieva, E., Naranjo, J. E., Milanés, V., Alonso, J., García, R., & Pérez, J. (2011).
(pp. 3365–3369). Automatic lateral control for unmanned vehicles via genetic algorithms. Applied
Imine, H., & Madani, T. (2011). Sliding-mode control for automated lane guidance of
Soft Computing, 11(January (1)), 1303–1309.
heavy vehicle. International Journal of Robust and Nonlinear Control, 23(1),
Rajamani, R. (2006). Vehicle dynamics and control. US: Springer.
67–76. http://dx.doi.org/10.1002/rnc.1818.
Scaner_url. 〈http://www.sera-cd.com〉, 〈http://www.scanersimulation.com〉.
Jazar, R. N. (2008). Vehicle dynamics; theory and application. New York: Springer-
Siciliano, B., Khatib, O., & Groen, F. (2009). The DARPA urban challenge: Autonomous
Verlag.
vehicles in city traffic, Vol. 56. Berlin Heidelberg: Springer-Verlag.
Kiencke, U., & Nielsen, L. (2005). Automotive control systems. Berlin Heidelberg:
Soualmi, B., Sentouh, C., Popieul, J. C., & Debernard, S. (2012). Fuzzy Takagi–Sugeno
Springer-Verlag.
LQ controller for lateral control assistance of a vehicle. In International IEEE
Kim, D., Kang, J., & Yi, K. (2011). Control strategy for high-speed autonomous
driving in structured road. In International IEEE conference on intelligent conference on intelligent vehicles symposium (IV) (pp. 377–382), Alcalá de
transportation systems (ITSC) (pp. 186–191). Henyres.
Kladis, G. P., Economou, J. T., Knowles, K., Lauber, J., & Guerra, T. M. (2011). Tagne, G., Talj, R., & Charara, A. (2013). Higher-order sliding mode control for lateral
Engineering applications of artificial intelligence energy conservation based dynamics of autonomous vehicles, with experimental validation. In Interna-
fuzzy tracking for unmanned aerial vehicle missions under a priori known wind tional IEEE conference on intelligent vehicles symposium (IV) (pp. 678–683), Gold
information. Engineering Applications of Artificial Intelligence, 24(2), 278–294. Coast.
Klan, G., & Skrjanc, I. (2007). Tracking-error model-based predictive control for Talj, R., Tagne, G., & Charara, A. (2013). Immersion and invariance control for lateral
mobile robots in real time. Robotics and Autonomous Systems, 55, 460–469. dynamics of autonomous vehicles, with experimental validation. In European
Kokotovic, P., & Marino, R. (1986). On vanishing stability regions in nonlinear control conference (ECC) (pp. 968–973), Zurich.
systems with high-gain feedback. IEEE Transactions on Automatic Control., 31 Utkin, V. I. (1992). Sliding modes in control and optimization. Berlin Heidelberg:
(October (10)), 967–970. Springer-Verlag.
Levinson, J., Askeland, J., Becker, J., Dolson, J., Held, D., Kammel, S., et al. (2011). Wei, J., Snider, J. M., Kim, J., Dolan, J. M., Rajkumar, R., & Litkouhi, B. (2013). Towards
Towards fully autonomous driving: Systems and algorithms. In International a viable autonomous driving research platform. In International IEEE conference
IEEE conference on intelligent vehicles symposium (IV) (pp. 163–168), Baden- on intelligent vehicles symposium (IV) (pp. 763–770).
Baden. Zhao, P., Chen, J., Mei, T., & Liang, H. (2011). Dynamic motion planning for
Marino, R., Scalzi, S., & Netto, M. (2011). Nested PID steering control for lane autonomous vehicle in unknown environments. In International IEEE conference
keeping in autonomous vehicles. Control Engineering Practice, 19(December on intelligent vehicles symposium (IV) (pp. 284–289).
(12)), 1459–1467. Zheng, S., Tang, H., Han, Z., & Zhang, Y. (2006). Controller design for vehicle stability
Menhour, L., Charara, A., & Lechner, D. (2014). Switched LQR/H infinity steering enhancement. Control Engineering Practice, 14, 1413–1421.
vehicle control to detect critical driving situations. Control Engineering Practice,
24, 1–14.