Control Engineering Practice 24 (2014) 1–14

Contents lists available at ScienceDirect

Control Engineering Practice

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

Switched LQR/H1 steering vehicle control to detect critical

driving situations
Lghani Menhour a,n, Ali Charara b, Daniel Lechner c
a
Mines-ParisTech, CAOR-Centre de Robotique, Mathématiques et systèmes, 60 boulevard Saint-Michel, 75272 Paris Cedex 06, France
b
HEUDIASYC Laboratory, CNRS UMR 7253, Université de Technologie de Compiégne, 60205 Compiégne, France
c
IFSTTAR Laboratory: Chemin de la Croix Blanche, 13300 Salon de Provence, France

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

Article history: This paper proposes a switching steering vehicle control designed using the linear quadratic regulator
Received 22 May 2012 (LQR) problem, the Linear Matrix Inequality (LMI) framework and the H 1 norm. The proposed switched
Accepted 4 November 2013 control law comprises two levels: the first level is a switched Proportional–Integral-Derivative controller
Available online 4 December 2013
of lateral deviation (PIDy) and the second is a switched Proportional-Derivative controller of yaw angle
Keywords: (PDψ ). These two levels are used to ensure an accurate tracking of the vehicle's lateral deviation y and
Switched linear systems yaw angle ψ. This control strategy makes use of a common Lyapunov function design method used for
Common Lyapunov function the stability analysis of switched continuous-time systems. Sufficient conditions for global convergence
LMI/LQR control problem of the switched control law are presented and proved under arbitrary switching signals. All these
H∞ control
conditions are expressed in terms of LMIs. The switched steering control was developed for an
Vehicle dynamics
application seeking to identify approximately the maximum achievable speed in a bend. This application
Switched steering vehicle control
requires a steering control for simulating a realistic nonlinear four-wheel vehicle model and for
performing a speed extrapolation test to evaluate the physical limits of a vehicle in a bend. This study
includes the performance tests using experimental data from the Peugeot 307 prototype vehicle
developed by IFSTTAR Laboratory.
& 2013 Elsevier Ltd. All rights reserved.

1. Introduction super-twisting. Other steering controllers based on a predictive LQ

control and the second-order decoupling controller are presented
Knowledge of a vehicle's dynamic behavior is essential for in Cole, Pick, and Odhams (2006) and Marino and Cinili (2009)
improving vehicle safety, vehicle handling and passenger comfort. respectively. The problem of improving ground-vehicle handling is
Active safety systems such as ABS and ESP can considerably reduce addressed in Poussot-Vassal et al. (2011) using braking and
the number of accidents, and these safety systems can be suspension control; this approach is extended through the intro-
enhanced by the knowledge of the vehicle's dynamic behavior. duction of steering control in Zheng, Ho, Han, and Zhang (2006).
This study presents a switched steering vehicle control strategy A steering control based on PID multi-controllers with two
for simulating nonlinear vehicle behavior in a closed loop with degrees of freedom is proposed in Menhour, Lechner, and
high accelerations in a bend. The steering control problem has Charara (2011). Other studies address the steering control problem
been addressed by a number of studies. Two controllers for vehicle via a steering torque control approach. The vehicle steering torque
steering control presented in Ackermann, Guldner, Sienel, control problem is also covered in a number of publications
Steinhauser, and Utkin (1995) are based on a linear and a non- including Fujiwara and Adachi (2002), Chong, Namgoong, and
linear control respectively. An inventory of driver models and their Sul (1996), and Enache, Sebsadji, Mammar, Lusetti, and Glaser
application in automobile dynamics is given in Plochl and (2009). In these studies, there has been relatively little focus on
Edelmann (2007). Recently, Cerone, Milanese, and Regruto the performance of controllers under high dynamic loads.
(2009) developed a driver model combining two tasks, namely The aim of this work is the design of switched steering vehicle
automatic lane-keeping and driver steering for either obstacle control for an uncertain switched linear vehicle model. The
avoidance or lane-change maneuvers. In Nouvelière and Mammar proposed control law comprises two levels, the first containing
(2007), a second-order sliding mode control is described, based on a switched PIDy control for lateral deviation, and the second
containing a switched PDψ control for yaw angle. The control
n
Corresponding author.
strategy makes use of a convex LMI (Linear Matrix Inequality)
E-mail addresses: [email protected] (L. Menhour), framework, the LQR control problem, the H 1 norm and the
[email protected] (A. Charara), [email protected] (D. Lechner). common Lyapunov function approach. It should be noted that

0967-0661/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.conengprac.2013.11.007
2 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

the common Lyapunov function approach is currently used for the

stability analysis of switched linear continuous-time systems
(Liberzon & Morse, 1999; Lin & Antsaklis, 2009; Sun & Ge, 2005).
This control law is used to control the system with a single input
(steering angle input) and two outputs (lateral deviation and yaw
angle). In our study longitudinal speed serves as a switching rule
to determine the PIDy and PDψ controllers to be used.
Switching systems are generally used when the knowledge of
nonlinear models or the design of control laws using the nonlinear
model becomes a hard or impossible task. Consequently, a finite
number of local models and switching rules are used to define the
switched model. Switched systems are a class of hybrid systems,
and stability and design problems relating to switched systems
have also been addressed in Branicky (1998), Liberzon and Morse
(1999), Daafouz, Riedinger, and Iung (2002), Sun and Ge (2005),
Koenig, Marx, and Jacquet (2008), and Lin and Antsaklis (2009).
In fact, the stability analysis of linear continuous-time and
discrete-time switched systems have been covered quite widely
in the literature: for switched continuous-time systems, a com-
mon Lyapunov function approach is used (Liberzon & Morse, 1999;
Lin & Antsaklis, 2009; Sun & Ge, 2005), while for switched
discrete-time systems, switched Lyapunov functions approach is Fig. 1. Steering control coupled with a nonlinear model and a road trajectory.
applied (Branicky, 1998; Daafouz et al., 2002; Koenig et al., 2008).
Several similar approaches for designing robust controllers,
including LPV=H 1 control (Apkarian, Gahinet, & Becker, 1995), embedded simulation function, executing a dynamic model with
gain scheduling state feedback control (Stilwell & Rugh, 1999), and incremented speeds, in order to extrapolate the vehicle behavior,
fuzzy systems (Castillo & Melin, 2008; Mendel, 2004; Sugeno & starting from a normal driving situation and moving towards
Kang, 1988) have been developed. In Apkarian et al. (1995) a virtual loss of control. This work was carried out on the
a method is proposed for solving the control problem in LPV laboratory vehicle (Lechner, 2008).
systems via a standard H 1 norm and Linear Matrix Inequalities The block diagram in Fig. 1 shows the system comprising the
(LMIs). An alternative approach based on a gain scheduling state reference trajectory of the road, the switched steering vehicle
feedback control for LPV systems is proposed in Stilwell and Rugh control and the nonlinear vehicle model. The proposed switched
(1999). This approach uses an interesting structure involving less steering control has two degrees of freedom: a single input
number of interpolation coefficients and readability of the algo- (steering angle) and two outputs. Fig. 1 shows the control law
rithm using physical variables. with two levels: the first level controls the lateral deviation via the
As regards the design of PID and PD controllers, several switched PIDy control, while the second controls the yaw angle via
methods have been developed and published. These controllers the PDψ control. The switched control approach is able, on the one
are designed using gain and phase margin methods (Fung, Wang, hand, to overcome the uncertainties of the linear vehicle (2) used
& Lee, 1998), convex and non-convex optimization methods to design the control law and, on the other hand, to perform the
(Astrom, Panagopoulos, & Hagglund, 1998; Ge, Chiu, & Wang, speed extrapolation tests in a closed loop and to estimate the
2002; Lin, Wang, & Lee, 2004), relay-feedback auto-tuning PID maximum speed of the vehicle in a bend. Such tests require
methods (Hang, Astrom, & Wang, 2002; Ho, Honga, Hanssonb, a robust control law capable of operating under high dynamic
Hjalmarssonc, & Denga, 2003), as well as oscillation and step loads and which can be used to characterize the nonlinear
response (Ziegler & Nichols, 1942). behavior of the vehicle. It should be noted that a definition of
The outline of this paper is as follows: the following section the speed extrapolation problem along with the different steps
formulates the problem and introduces the concept of speed that yield an approximation of the maximum achievable speed are
extrapolation. The reference trajectory and the vehicle models presented in Menhour et al. (2011).
used are presented in Sections 3 and 4 respectively. The different
steps in the design of the switching steering vehicle control (SSVC)
are described in Section 5. Section 6 compares the steering control 3. Reference trajectory of the road
with real data acquired under high loads using a Peugeot 307
laboratory vehicle. The SSVC is also tested for low friction The parameters characterizing the trajectory are the yaw angle
coefficients and parameter variations. The obtained Results prove ψd, the curvature ρd, the path length coordinate sd and the xd and
the effectiveness of the control law. Section 7 contains concluding yd coordinates computed by
8 R
remarks. >
> ψ ðs Þ ¼ ψ d0 þ ssd0d ρd ðsd Þ ds
> d d
< Rs
xd ðsd Þ ¼ xd0 þ sd0d cos ðψ d ðsd ÞÞ ds ð1Þ
>
> R
2. Problem formulation >
: y ðs Þ ¼ y þ sd sin ðψ ðs ÞÞ ds
d d d0 sd0 d d

The control law is used to estimate the maximum achievable ψ d0, xd0, yd0, sd0 and ρd ðsd Þ ¼ ay ðsd Þ=V 2x ðsd Þ are the original terms of
speed in a bend. For this purpose, the speed extrapolation tests are the trajectory.
performed using experimental data acquired in normal driving
situations. Extrapolation is based on several instances of nonlinear 4. Vehicle models used
vehicle models coupled with the same number of control laws,
which are executed to extrapolate the behavior of the vehicle in In this study two vehicle models (Jazar, 2008; Milliken &
a bend. In other words, this real-time application makes use of an Milliken, 1995; Rajamani, 2006) are used: a linear two-wheel
 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14 3

2 2C f þ 2C r 2C f Lf  2C r Lr 2C f þ 2C r
3
vehicle model (L2WVM) is used to design the switching control,   0
mV x ðtÞ mV x ðtÞ m
and a nonlinear four-wheel model (NL4WM) is used as a vehicle 6 7
6 2C f Lf  2C r Lr 2C f L2f þ 2C r L2r 2C f Lf  2C r Lr 7
6 7
simulator to perform the speed extrapolation tests. The experi- A~ ¼ 6  Iz V x ðtÞ  I z V x ðtÞ 0 Iz 7;
6 7
mental validation of these models is described in detail in 4 1 0 0 0 5
Menhour, Lechner, and Charara (2012). 0 1 0 0
 
2C f 2Lf C f
B~ ¼ 0 0
m Iz
4.1. Linear two-wheel vehicle model for control design
The matrices A~ and B~ of (2) depend on Vx, Cf and Cr. Therefore,
for each value of these parameters, a local linear model can be
The vehicle lateral dynamics is determined by considering a
obtained. In fact, when the vehicle operates under high dynamic
bicycle vehicle model. In this model we assume that the symmetry
loads, several parameters become variables: for example, the tire
of the vehicle is sufficient to give an approximation of the lateral
cornering stiffnesses Cf and Cr become nonlinear functions of
dynamics (the vehicle model notation is given in Table 1). In
sideslip angles, the longitudinal slip ratio, vertical forces and other
normal driving situations lateral tire forces can be modeled as
dynamic parameters. Unfortunately, if we take into account all
proportional to the slip angles of each axle (F yf ¼ C f αf , F yr ¼ C r αr ),
variations and uncertainties, (2) becomes nonlinear. Consequently,
that is to say
the design of the control law becomes a hard or impossible task.
    To better understand some uncertainties and the simple model,
V y þ ψ_ Lf V y  ψ_ Lr
F yf ¼ C f δ  ; F yr ¼ C r  we define a set of sub-models and switching rules. Model (2) can
Vx Vx
then be treated as a switched continuous-time linear system:
The state space of this model can be written as follows: x~_ ðtÞ ¼ A~ sðtÞ ðtÞxðtÞ
~ þ B~ sðtÞ uðtÞ ð3Þ

x~_ ðtÞ ¼ AðtÞ

~ xðtÞ ~
~ þ BuðtÞ ð2Þ where sðtÞ is a switching rule that will be used as a supervisor to
determine the appropriate local model.
With
4.2. Nonlinear four-wheel vehicle simulation model
~ ¼ ½yðtÞ;
xðtÞ _ ψ_ ðtÞ; yðtÞ; ψ ðtÞ; uðtÞ ¼ δðtÞ

Eq. (4) presents the four degrees of freedom, namely the

longitudinal, lateral, yaw and roll movements of the nonlinear
Table 1 four-wheel vehicle model in a plane. All these movements are
The vehicle model variables.
illustrated in Fig. 2, and explanations of all the symbols are given
Symbol Variable name in Table 1. The rotational speeds of the wheels are assumed to be
known, and are used to compute the longitudinal slip ratio and the
ax, ay Longitudinal and lateral accelerations (m/s2) longitudinal force. The vehicle model movements are governed by
V x, V y Longitudinal and lateral speeds (m/s)
the following equations:
ψ_ , ψ Yaw angle (rad) and yaw rate (rad/s)
ϕ, ϕ_ Roll angle (rad) and roll rate (rad/s) 8
>
> Longitudinal motion
δ Steering angles (rad) >
>
> _
> mðV x  ψ V y Þ  ms hψ ϕ ¼ ðF ωx1 þ F ωx2 Þ cos δ
Fxi Longitudinal tire forces of the wheel i in-vehicle coordinate system >
> _ _
>
>
(N) >
>  ðF þ F Þ sin δ þF ωx3 þ F ωx4  12ρSC x V 2x
>
> ωy1 ωy2
F xωi Longitudinal tire forces of the wheel i in wheel coordinate system >
>
>
> Lateral motion
(N) >
>
>
> € ¼ ðF
mðV_ y þ ψ_ V x Þ þ ms hϕ
Fyi Lateral tire forces of the wheel i in-vehicle coordinate system (N) >
>
> ωx1 þ F ωx2 Þ sin δ
F yωi Lateral tire forces of the wheel i in-wheel coordinate system (N) >
>
τi >
> þ ðF ω þ F ω Þ cos δ þ F ω þ F ωy4
Longitudinal tire slip ratio of the wheel i >
< y1 y2 y3
αi Tire slip angle of the wheel i (rad) Yaw motion ð4Þ
ζi Camber angle of the wheel i >
> € ¼ ðS =2ÞðF
> I z ψ€  I xz ϕ
> F Þ  L ðF þ F Þ
Fzi Vertical tire forces of the wheel i (N) >
> b ωx3 ωx4 r ωy3 ωy4
>
>
Fyf, Fyr Front (f) and rear (r) lateral tire forces (N) >
> þ ½ðSb =2ÞðF ωx1  F ωx2 Þ þ Lf ðF ωy1 þ F ωy2 Þ cos δ
αf, αr Front (f) and rear (r) tire slip angles (rad) >
>
>
> þ ½  ðSb =2ÞðF ωy1  F ωy2 Þ þ Lf ðF ωx1 þ F ωx2 Þ sin δ
g Gravitational constant (m/s2) >
>
>
>
μ Friction coefficient >
> Roll motion
>
>
xd, yd Desired longitudinal and lateral displacements (m) >
> € _
> I xx ϕ þ ms hðV y þV x ψ_ Þ I xz ψ€ ¼ ðK ϕf þ K ϕr Þϕ
ψd, ρd Desired yaw angle (rad) and road curvature
>
>
>
>
sd Desired path length co-ordinate (m) >
:  ðC ϕf þ C ϕr Þϕ _ þ m ghϕ
s
ey, eψ Errors on lateral deviation (m) and yaw angle (rad)
Te Sampling time (s)
Iz Yaw moment of inertia (kg m  2) The elastokinematic characteristics of the laboratory vehicle are
Ix Moment of inertia about the x-axis (kg m2) defined on a test rig and the main suspension compliances are
I xz Moment of inertia about the x- and z-axes (kg m2) taken into account (i.e. toe in/out with lateral forces).
m, ms Vehicle mass and sprung mass of the vehicle(kg)
Cf , Cr Front and rear cornering stiffnesses (N/rad)
C ϕðf ; rÞ Front and rear suspension roll damping (N m/rad s)
4.2.1. Tire model
K ϕðf ; rÞ Front and rear suspension roll stiffnesses (N m/rad)
Lf Distance from the CoG to the front axle (m) Several mathematical tire models have been developed and
Lr Distance from the CoG to the rear axle (m) published in recent decades (Burckhardt, 1993; Dugoff, Fanches, &
hcg Height of the center of gravity (m) Segel, 1970; Pacejka, 2005). These models are highly dependent on
Sb Wheel track (m) tire–road forces. The Magic formula tire model is widely used as the
Rω Tire radius (m)
CoG Center of gravity
steady-state tire force in studies of vehicle dynamics. The coupled
Pacejka (2005) model is used, represented by the nonlinear
4 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

4.2.3. Longitudinal slip ratio

For each wheel of the vehicle, the nonlinear function for the
wheel slip ratio is defined as
8
>
> Rω ωi
< τi ¼ V
>  1 if V xωi 4 Rω ωi
xωi
ð7Þ
>
> Rω ωi
: τi ¼ 1  V
> if V xωi o Rω ωi
xωi

where ωi and V xωi are respectively the rotational and the long-
itudinal speeds of the wheel.

4.2.4. Vertical forces

The vertical force applied at each wheel corresponds to the
static load added to the load transfer that results from the long-
itudinal and lateral accelerations. The load transfer due to the roll
motion is also considered. The equations of these forces are

Lr gm hcg max Lr hcg may _Þ

ðK ϕf ϕ þ C ϕf ϕ
F z1 ¼   
2ðLf þ LrÞ 2ðLf þ LrÞ 2Sb ðLf þ LrÞ Sb

Lr gm hcg max Lr hcg may _Þ

ðK ϕf ϕ þ C ϕf ϕ
F z2 ¼  þ þ
2ðLf þ LrÞ 2ðLf þ LrÞ 2Sb ðLf þ LrÞ Sb

Lf gm hcg max Lf hcg may _Þ

ðK ϕr ϕ þ C ϕr ϕ
F z3 ¼ þ  
2ðLf þ LrÞ 2ðLf þ LrÞ 2Sb ðLf þ LrÞ Sb

Lf gm hcg max Lf hcg may _Þ

ðK ϕr ϕ þ C ϕr ϕ
F z4 ¼ þ þ þ ð8Þ
2ðLf þ LrÞ 2ðLf þ LrÞ 2Sb ðLf þ LrÞ Sb

5. Switched steering vehicle control design

Fig. 2. Nonlinear vehicle model with longitudinal, lateral, yaw and roll movements.
This section provides a proof of the existence and the conver-
gence of the switched steering vehicle control, which is expressed
functions Fy and Fx: in terms of LMI constraints. The LMI constraints on the steering
( angle saturations are added into the algorithm control design. The
F ωxi ¼ F x ðζ i ; τi ; αi ; F zi Þ ði ¼ 1⋯4Þ design of the PIDy and PDψ controllers and the switching strategy
ð5Þ
F ωyi ¼ F y ðζ i ; τi ; αi ; F zi Þ ði ¼ 1⋯4Þ are also described. It should be remarked that in our case study
only the variation of longitudinal speed is considered in the design
where τi is the tire slip ratio, αi the tire sideslip angle, Fzi the vertical procedure.
force and ζi the camber angle. All the parameters of this model were
provided by MICHELIN and obtained from a test bed. These
parameters are used to simulate the limit behavior of the tires 5.1. Formulation of the switched steering control problem
where the longitudinal and the lateral slips are coupled, in
accordance with Pacejka (2005). This is a particularly important The proposed switched steering vehicle control uses a switch-
step impacting the accuracy of the results obtained. αi, τi and Fzi are ing control strategy (Branicky, 1998; Daafouz et al., 2002; Koenig
calculated for each wheel using the nonlinear functions descri- et al., 2008; Liberzon & Morse, 1999; Lin & Antsaklis, 2009; Sun &
bed below. Ge, 2005), LMI optimization (Boyd, Ghaoui, Feron, & Balakrishnan,
1994), the LQR control problem, a Linear Matrix Inequality frame-
work and the H 1 norm. More precisely, the design procedure uses
4.2.2. Sideslip angle a common Lyapunov function for the stability analysis of switched
The nonlinear formulations for the sideslip angles used in the continuous-time systems (Liberzon & Morse, 1999; Lin & Antsaklis,
tire model are as follows: 2009; Sun & Ge, 2005). The design of a switched steering vehicle
8 0 1 control1 comprising switched PIDy control and switched PDψ
>
> control is presented. The PIDy components are obtained by
>
>
> B V y þ Lf ψ_ C
>
> α 1 ¼ δ  arctan @ A
>
> S enlarging the state vector of the system (3) to include the integral
>
>
> V x  b ψ_ R
>
>
> 0 2 1 term of the lateral deviation (yI ¼ y dt). This term is introduced
>
>
>
>
>
> B V y þ Lf ψ_ C to attenuate the noise effects in the control process of the lateral
>
>
>
α2 ¼ δ  arctan@ A
> S deviation (see the discussion in Busawon & Kabore, 2001 about the
>
< V x þ b ψ_
2
0 1 ð6Þ disturbance attenuation using the integral action in the estimation
>
>
>
> B V y  Lr ψ_ C issues). The enlarged2 switched, uncertain system is as follows:
>
> α ¼  arctan @ A
>
> 3
S
>
> V x  b ψ_
>
> _ ¼ AsðtÞ xðtÞ þ BsðtÞ uðtÞ
xðtÞ ð9Þ
>
> 0 2 1
>
>
>
>
>
> B V y  Lr ψ_ C
>
> α4 ¼  arctan@ A
1
The notation used in the control law is given in Table 2.
>
> S
>
> V x þ b ψ_
2
Despite the modification of the state vector of the system (3), the new
:
2 enlarged system (10) remains controllable.
 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14 5

where The equations corresponding to (18) along the trajectory of the

" # " # switched closed-loop system (15) is then given by
A~ sðtÞ 041 B~ sðtÞ
AsðtÞ ¼ ; BsðtÞ ¼ ; x_ ¼ ½ x~_ y_ I 
012 1 0 0 0 ð18Þ 3 ðeTx ðAsðtÞ  BsðtÞ K sðtÞ ÞT þ ωT BTsðtÞ ÞPex

and uðtÞ ¼ δðtÞ þ eTx PððAsðtÞ  BsðtÞ K sðtÞ Þex þBωsðtÞ ωÞ

For our problem, let ex ðtÞ ¼ xðtÞ  xr ðtÞ be the state tracking þ eTx ðQ sðtÞ þ K TsðtÞ RsðtÞ K sðtÞ Þex  γ 2 ωT ω
error, where xr(t) is the reference signal. The system in (9) can then
3 eTx ððAsðtÞ  BsðtÞ K sðtÞ ÞT PÞex þ ωT BTsðtÞ Pex
be rewritten as
e_ x ðtÞ ¼ AsðtÞ ex ðtÞ þ BsðtÞ uðtÞ þ BωsðtÞ ωðtÞ ð10Þ þ eTx ðPðA sðtÞ  BsðtÞ K sðtÞ ÞÞex þ eTx PBωsðtÞ ω
þ eTx ðQ sðtÞ þ K TsðtÞ RsðtÞ K sðtÞ Þex  γ 2 ωT ω
System (10) can be considered as an H 1 control problem, where
BωsðtÞ ¼ I and ωðtÞ ¼ AsðtÞ xr ðtÞ  x_ r ðtÞ is the bounded disturbance. 3 eTx ððAsðtÞ  BsðtÞ K sðtÞ ÞT PÞex þ ωT BTsðtÞ Pex
In this problem, we wish to minimize the disturbance effects with
þ eTx ðPðA sðtÞ  BsðtÞ K sðtÞ ÞÞex þ eTx PBωsðtÞ ω
respect to any given reference signal, which is equivalent to
determining u(t) for the enlarged system (10) so as to minimize þ eTx ðQ sðtÞ þ K TsðtÞ RsðtÞ K sðtÞ Þex  γ 2 ωT ω o 0 ð19Þ
the following H 1 cost function:
Z 1 Now, (19) can be rewritten as follows:
2 3
JðuÞ ¼ ðeTx Q sðtÞ ex þ uT RsðtÞ u  γωT ωÞ dt ð11Þ βsðtÞ PBωsðtÞ  ex 
0 _ T 4
V ðex Þ r½ex ω  T
T 5 ð20Þ
BωsðtÞ P  γ 2 I ω
More precisely, the H 1 cost function (11) can be rewritten as
follows:
with
JzJ oγ JωJ ð12Þ βsðtÞ ¼ ðAsðtÞ BsðtÞ K sðtÞ ÞT P þ PðAsðtÞ  BsðtÞ K sðtÞ Þ þ Q sðtÞ þK TsðtÞ RsðtÞ K sðtÞ .
where V_ ðex Þ is negative definite for any given ½eTx ωT  a 0 if:
2 3
‖z‖2 ¼ eTx Q sðtÞ ex þ uT RsðtÞ u ð13Þ βsðtÞ PBωsðtÞ
4 5o0 ð21Þ
BTωsðtÞ P  γ2 I
where Q sðtÞ and RsðtÞ are symmetrical positive definite matrices,
and γ is a given positive constant of the H 1 norm. The optimal
Now, pre- and post-multiplying by Y ¼ diagðP  1 ; IÞ, (21) becomes
control solution for this problem can be expressed in the state
2 3
tracking error feedback as follows: βsðtÞ PBωsðtÞ
T4 5Y o 0
Y ð22Þ
un ðtÞ ¼ δ ðtÞ ¼  K sðtÞ ex ðtÞ
n
ð14Þ BTωsðtÞ P  γ 2 I

The main objective is to determine the switched control (14) to

which is equivalent to
control the switched system (10) in order that the following 2 1 3
switched closed-loop system P βsðtÞ P  1 BωsðtÞ
4 5o0 ð23Þ
e_ x ðtÞ ¼ ðAsðtÞ  BsðtÞ K sðtÞ Þex ðtÞ þ BωsðtÞ ωðtÞ ð15Þ BTωsðtÞ  γ2I

is stable and satisfies the H 1 cost function (11). To this end, with
consider the following proposition. P  1 β sðtÞ P  1 ¼ P  1 ðAsðtÞ  BsðtÞ K sðtÞ ÞT þ ðAsðtÞ  BsðtÞ K sðtÞ ÞP  1 þ P  1
Proposition 1. For the switched linear system (10) with the cost Q sðtÞ P  1 þ P  1 K TsðtÞ RsðtÞ K sðtÞ P  1 or, equivalently,
2 3 " #
function (11) and the switching signal sðtÞ, suppose that for all ΔsðtÞ BωsðtÞ
4 T 5 P  1 Q sðtÞ P  1 þ P  1 K TsðtÞ RsðtÞ K sðtÞ P  1 0
i A f1; …; Mg, the pair ðAi ; Bi Þ is controllable. If a level of attenuation þ o0
BωsðtÞ  γ 2 I
γ 4 1 exists, along with a common symmetric positive definite matrix 0 0
X A Rðn þ nI Þðn þ nI Þ and matrices Γ i A Rmðn þ nI Þ such that the following ð24Þ
inequality is satisfied for i ¼ 1; …; M:
with ΔsðtÞ ¼ P ðAsðtÞ  BsðtÞ K sðtÞ Þ þ ðAsðtÞ  BsðtÞ K sðtÞ ÞP
1 T 1
. Let us
2 3 define the following change of variables:
Πi Bωi X 0 (
6 T 7 X ¼ P 1
6 Bωi  γ2I Γi 0 7 : ð25Þ
6 7 Γ sðtÞ ¼  K sðtÞ P  1 ¼  K sðtÞ X
6 T 7 o0 ð16Þ
6X
4 Γ T
i  Q i 1 0 7
5
0 0 0  Ri 1 the LMI (24) is then equivalent to
2 3
XAsðtÞ þ Γ sðtÞ BsðtÞ þ AsðtÞ X þ BsðtÞ Γ sðtÞ BωsðtÞ
T
4 5
then uðtÞ ¼ K i eðtÞ is the switched H 1 state tracking control law, BTωsðtÞ  γ2 I
where Ki is the optimal gain of Eq. (14). Moreover, the gains of the " #" #" #
controller Ki are given by K i ¼  Γ i X  1 and Π i ¼ XATi þ X T Γ sðtÞ
T Q sðtÞ 0 X 0
þ
Γ Ti BTi þ Ai X þBi Γ i . 0 0 0 RsðtÞ Γ sðtÞ 0 o 0 ð26Þ

Proof. Let us define the Lyapunov function Using the Schur complement, (26) becomes
Vðex Þ ¼ eTx Pex ð17Þ
2 3
is positive definite. Sufficient conditions for the existence of (11) Π sðtÞ BωsðtÞ X 0
6 T 7
are related to the existence of a common Lyapunov function Vðex Þ 6 BωsðtÞ γ I 2
Γ sðtÞ 0 7
6 7
such that the following inequality is satisfied: 6 T 7 o0 ð27Þ
6 X
4
Γ TsðtÞ 1
 Q sðtÞ 0 7
5
1
V_ ðex Þ þ eTx Q sðtÞ ex þ uT RsðtÞ u  γ 2 ωT ω o 0 ð18Þ 0 0 0  RsðtÞ
6 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

Table 2 5.2. LMI saturation constraint on the steering angle

Variables of the control law.
The design of a control law via LMI optimization makes it easier
Symbol Variable name
to take specifications relative to the control signal into account.
xr Desired state of the enlarged system We now introduce a constraint on the limits of the steering angle
ex State tracking error to reproduce a real driving phenomenon. This constraint is
ω Disturbance expressed as an LMI constraint. In extreme driving conditions
Δ Speed between the operating points [m/s]
with high dynamic loads the steering angle applied on the vehicle
K yðP;I;DÞi Gains of the PIDy on the lateral deviation
model cannot exceed umax ¼ δmax . When the steering angle
K ψðP;DÞi Gains of the PDψ on the yaw angle
exceeds this limit the tracking error becomes too large and the
PIDyi Switched PID controller for lateral deviation
PDψi Switched PD controller for yaw angle anti-loop reaction becomes ineffective. Note that these limits are
δmax;min Maximum and minimum steer angles (rad) not related to the physical limits of the steering system, but are
Qi, Ri LQR weighting matrices related to the experimental conditions. Let us first consider the
ε Upper bound for the quadratic cost Schur complement and an invariant ellipsoid Ω (Blanchini &
γ Level of attenuation
s Switched signal
Miani, 2008), and assume that the control signal uðtÞ ¼
P Common Lyapunov function δðtÞ ¼  Γ i X  1 xðtÞ is the solution of the LMIs described by Eqs.
Ki Switched H 1 gains (30)–(32) and xT ð0ÞX  1 xð0Þ r ε. The quadratic stability can be
X,Γi Solution of the optimization problem interpreted in terms of the invariant ellipsoid, which implies that
ðÞT Stands for the transpose matrix
x(t) belongs to Ω for all t Z 0, and consequently:
ðÞ 4 0 Symmetric positive definite matrix
V ðmax;minÞ
x
Bounds of the speed interval (m/s)
max‖δðtÞ‖22 ¼ max‖Γ i X  1 xðtÞ‖22
Vxi0 Operating points (m/s) t 40 t 40

r max‖Γ i X  1 x‖22
xAΩ
Substituting sðtÞ ¼ i and K sðtÞ ¼ K i ¼  Γ sðtÞ X ¼  Γi X1 1
into
(27), (16) is obtained, with Π sðtÞ ¼ Π i ¼ XATsðtÞ þ Γ sðtÞ BTsðtÞ þAsðtÞ X þ
T
¼ λmax ðΓ i X  1=2 X  1=2 Γ i Þ
T

BsðtÞ Γ sðtÞ ¼ XATi þ Γ i BTi þ Ai X þ Bi Γ i . □

T

¼ λmax ðΓ i X  1 Γ i Þ
T
ð34Þ
Furthermore, in many practical situations, the cost function
(11) will satisfy the bound (Boyd et al., 1994):
J r xT ð0ÞPxð0Þ ð28Þ
where xð0Þ corresponds to the initial state conditions of the system
(10). Based on Proposition 1, the optimal solution to the LQR
problem can be obtained. Furthermore, the LQR problem described
above can be rewritten as an optimization problem that minimizes
a quadratic objective function over X and Γi subject to a set of LMI
constraints as follows:
min xT ð0ÞX  1 xð0Þ ð29Þ
X;Γ i

subject to: Fig. 3. Experimental vehicle.

2 3
Πi Bωi X 0
6 T 7
6 Bωi  γ  1 I Γi 0 7
6 7
6 T 7 o0 ð30Þ
6X
4 Γ Ti  Q i 1 0 7
5
0 0 0  Ri 1

X 40 ð31Þ

for i ¼ 1; …; M

where Π i ¼ XATi þ Γ i BTi þ Ai X þ Bi Γ i . It has also been shown that in

T

practical situations the quadratic objective function (29) can be

replaced by a quadratic constraint that minimizes the variable ε,
and the quadratic objective function (29) can be modified as
xT ð0ÞX  1 xð0Þ r ε, which is equivalent, by Schur complement, to
" #
ε xT ð0Þ
Z0 ð32Þ
xð0Þ X

where ε is an additional positive variable used as an upper bound

for the cost function.
The optimization problem given by (29)–(31) is reformulated as
a set of LMIs (30)–(32) to obtain an optimal solution ðX; Γ i Þ that
satisfies them. The state feedback gain is given by
Fig. 4. Longitudinal speeds of two tests and 15 operating points on
K i ¼  ΓiX  1 ð33Þ longitudinal speed.
 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14 7

which is equivalent, by Schur complement, to 5.3. Structure of the PIDy and PDψ controllers
" #
δmax Γ i The following equation gives the vector which contains the
Z0 ð35Þ
Γ Ti X different parameters of the PIDy and PDψ controllers obtained by
solving the optimization problem subject to a set of LMI con-
where δmax is the maximum steering angle (‖δðtÞ‖22 r δmax ). The
2
straints described by Eqs. (36). The following gains are obtained:
inequality (34) is a generalized eigenvalue problem and is true at
ψ ψ
all times t Z 0 if the LMI (35) is satisfied. K i ¼  Γ i X  1 ¼ ½K yDi ; K Di ; K yPi ; K Pi ; K yIi  ð37Þ
The optimal solution of the control problem is the solution that
minimizes the variable ε, taking into account the new LMI In the enlarged models (10), the PIDy and PDψ controllers
constraints given by Eqs. (32) and (35). The control optimization become static-state feedback controllers, and the static feedback
problem given by Eqs. (29)–(31) is transformed as follows: gain Ki contains all the PIDy and PDψ controller parameters. The
8 following equation illustrates these controllers:
< Eqs: ð30Þ; ð31Þ
> Z
ψ ψ
minimize ε subject to : Eqs: ð32Þ; ð35Þ ð36Þ uðtÞ ¼ δðtÞ ¼ K yPi ey þ K yIi ey dt þ K yDi e_ y þ K Pi eψ þ K Di e_ ψ ð38Þ
>
: for i ¼ 1; …; M: |fflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflffl}
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} PD
ψi
PIDyi

Fig. 5. Longitudinal speed and steering angle at Site 1: measured and obtained Fig. 7. Longitudinal speed and steering angle at Site 2: measured and obtained
switching PIDy and PDψ controllers coupled with (L2WVM and NL4WVM). switching PIDy and PDψ controllers coupled with (L2WVM and NL4WVM).

Fig. 6. Trajectories at Site 1: reference trajectory and trajectories of (L2WVM and NL4WVM) coupled with the switching PIDy and PDψ controllers and tracking errors (lateral
deviation error ΔY and yaw angle error Δψ ).
8 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

The two parts of Eq. (38) correspond to the PIDy and PDψ ðPIDyi ; PDψ i Þ controllers is requested are defined as ½V 0xi  Δ=2;
controllers that control respectively the lateral deviation and the V 0xi þ Δ=2½ (for i A f1; …; Mg). We consider that Vxi0 is the center of
yaw angle. each interval and the operating point of each local linear model
(see Eq. (40)). The range of variation of speed is V x A ½5; …; 50 m=s½;
5.4. Switching strategy consequently M values of the speed and M of PIDy and PDψ
controllers will be taken into account.
Fig. 1 shows the block diagram of the switching PIDy and PDψ The structure of the two switches for simultaneous selecting of
controllers. The two switches select PIDy and PDψ controllers PIDy and PDψ controllers is shown in Fig. 1. These switches are
using longitudinal speed. The speed intervals in which each pair of powered by the longitudinal speed. The operating points

Fig. 8. Trajectories at Site 2: reference trajectory and trajectories of (L2WVM and NL4WVM) coupled with the switching PIDy and PDψ controllers and tracking errors (lateral
deviation error ΔY and yaw angle error Δψ ).

Fig. 9. Experimental validation of dynamic parameters in closed loop of L2WVM and NL4WVM at Site 2.
 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14 9

represented by different values of the longitudinal speed are local linear model and are validated over the interval:
   
  Δ Δ Δ 0 Δ
V 0xi ¼ V min þ Δ i  12 ; i ¼ 1; …; M ð39Þ ½V min ; 4V max
 ¼ V 0
 ; 4 V 0
þ [ ⋯ [ V 0
xi  n  ; V þ
x x x x1
2 x1
2 2 xi  n 2
 
where Δ ¼ ðV max  V min
x Þ=M is the speed that separates the different
Δ Δ
x [ ⋯ [ V 0xM  ; V 0xM þ ð40Þ
operating points. The ith PIDy and PDψ controllers stabilize the 2 2

Fig. 10. Experimental validation of front longitudinal forces in closed loop of NL4WVM at Site 2.

Fig. 11. Experimental validation of lateral forces in closed loop of NL4WVM at Site 2.

Fig. 12. Experimental validation of vertical forces in closed loop of NL4WVM at Site 2.
10 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

Table 3
Maximum values for the normalized errors (ez ; %) of NL4WVM þ SSVC.

Errors Site 1 Site 2 Errors Site 1 Site 2

Maximum normalized mean errors for certain parameters

jeax j 3.5365 3.1964 jeay j 4.2361 4.2467
jeV x j 0.3422 0.4298 jeV y j 2.2125 2.8342
jeψ_ j 2.4532 2.1321 jeϕ_ v j 2.1489 2.2389

Maximum normalized mean errors for longitudinal forces

jeF x1 j 7.6543 7.4563 jeF x2 j 7.2376 8.0432

Maximum normalized mean errors for lateral forces

jeFy1 j 7.3231 7.1243 jeFy2 j 5.9687 4.9421
jeFy3 j 3.9867 4.2345 jeFy4 j 4.7854 5.2346

Maximum normalized mean errors for vertical forces

jeF z1 j 4.1234 2.9874 jeF z2 j 4.1243 3.1235
jeF z3 j 2.9876 4.0423 jeF z4 j 4.1236 3.9856

Fig. 13. Switching indices between the gains of switching steering control — Left: indices at Site 1 and Right: indices at Site 2.

6. Test and comparison of the switched steering vehicle vehicle models, L2WM and NL4WM. The goal of using the NL4W
control with experimental data model is to show the performance and efficiency of the steering
control with high nonlinearities, to perform speed extrapolation
The experimental data used in this work were acquired by the tests, and to evaluate the performance of this control law under
vehicle shown in Fig. 3, a Peugeot 307 test car belonging to the high dynamic loads (or high lateral acceleration) in the bend.
IFSTTAR - Département Mécanismes d'Accidents Laboratory. This Under these driving conditions L2WVM reaches its limits, making
vehicle is equipped with a number of sensors, including gyro- NL4WM necessary to overcome the limitations of L2WVM.
meters, accelerometers, steering angle sensors, wheel-force trans- This section presents the validation results described in Section
ducers and CORREVIT, that measure the longitudinal and lateral 4 using experimental data from the IFSTTAR Peugeot 307. Fig. 8
speeds. The sampling frequency of these sensors is 100 Hz. shows the three bends at the second site used for this study. The
The results shown below were obtained with the following second bend (on the left) has the smallest radius of curvature, and
parameters: the LQR cost weights Qi and Ri are fixed for each value is more dangerous than the first and third bends under high loads.
of the speed and ε, using M ¼15 local linear models for Figs. 5–8 show the experimental validation, at two sites, of the
V x ðtÞ A ½5; 50 m s  1 ½. Fig. 4 shows 15 equilibrium points used in “switching PIDy and PDψ controllers” using the L2W (dotted line)
the following simulation results, plotted with longitudinal speeds and the NL4W (solid line) vehicle models in terms of trajectory
from two tests performed at sites 1 and 2. For a continuous tracking. NL4W, providing a more realistic vehicle behavior,
steering angle to be guaranteed during switching, parameter M produces less errors than the Linear bicycle model in relation to
must be chosen judiciously, and so a compromise has to be made. lateral deviation and yaw angle (Figs. 6 and 8) and yields a steering
For example, a high value of M leads to a faster and a more input very close to the actual driver steering angle measured on
efficient control system but at the same time causes significant the vehicle (Figs. 5 and 7). These results prove that under high
switching between the operating regimes of the controller. The dynamic loads L2WVM reaches its limits, making the NL4W model
value of M is chosen with reference to experimental data acquired necessary to simulate the nonlinear behavior of the vehicle.
at several experimental sites. In this paper two sites are presented.
For all the results obtained using experimental data acquired at (footnote continued)

real sites3 the proposed steering control law was tested with two

 Site 1: Results shown in Figs. 5 and 6.

3
The sites used in this study, located in the Côtes d'Armor, were selected by  Site 2: Results shown in Figs. 7–12, 14 and 15. All the figures show the results of
the French SARI project in virtue of their difficult driving conditions. The experi- the speed extrapolation test performed using experimental data acquired at the
mental results shown in this section concerned the following sites: second site (Figs. 16–18).
 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14 11

Fig. 14. Parameter variations in the NL4WM test at Site 2: variation of the mass m and yaw inertia moment Iz — steering angles, lateral deviation error ΔY and yaw angle
error Δψ .

Fig. 15. Test at Site 2 of switching PIDy and PDψ controllers coupled with NL4WM for three values of the friction coefficient in the range μA ½0:7; 1: steering angles, lateral
deviation error ΔY and yaw angle error Δψ .

Figs. 9–12 compare the performances of the systems composed longitudinal acceleration when this control law is coupled
of L2WVM and NL4WM coupled separately with SSVC, using with NL4WM.
experimental data from Site 2:  Figs. 10–12 show for the NL4WVM coupled with steering
vehicle control the front longitudinal, vertical and lateral forces.
 Fig. 9 shows the main dynamic parameters of NL4WVM (solid
line) and L2WVM (dash-dot line) coupled with steering vehicle In these results all dynamic parameters are quite close to the
control separately. These parameters are lateral speed, yaw measurements for this trial involving high loads (lateral accelera-
rate, sideslip angle at the CoG, lateral acceleration and tion up to 6.4 m s  2). Table 3 summarizes these results using the
12 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

maximum absolute values of the normalized errors (41). The controllers used at Site 1 are 5, 6 and 7 (see the left-hand side of
normalized error ez for a computed variable z is defined as Fig. 13), while those used at Site 2 are 9 and 10 (the right-hand
side of Fig. 13).
ez ðiÞ ¼ 100jzðiÞ  zmeas ðiÞj=maxjzmeas j ð41Þ
Figs. 14 and 15 only show the performances of the switching
The normalized errors in Table 3 show that NL4WM þ SSVC steering control coupled with the NL4WM. Fig. 14 shows that the
produce satisfactory behavior close to the measurements acquired steering angle calculated by the control law remains realistic and
at Site 2. The maximum absolute values of the normalized errors that errors are limited to the lateral deviation and yaw angle when
are less than 8.1% for lateral and longitudinal forces, and less than using structured parametric uncertainties (the mass m and yaw
4.5% for other dynamic parameters. These good experimental inertia moment Iz are reduced (dash-dot line) or increased (solid
results confirm that an approach based on NL4WM þ control line) by 20%), proving the good level of robustness and the
laws is appropriate for performing the tests described in the performance of the proposed approach.
following subsection. The switching steering vehicle control was also tested with the
Fig. 13 shows the switching indices between the gains of the road friction coefficient set at different values. Fig. 15 illustrates
switching PIDy and PDψ controllers on two sites. The PIDy and PDψ the results obtained using three separate values for the friction
coefficient μ, namely 0.7, 0.8 and 0.9. Here the switching PIDy and
PDψ controllers provide satisfactory performances in terms of
tracking errors and the stability of the steering angles, except
when the adhesion value equals 0.7. The longitudinal speed is
95 km h  1 (see the speed profile in Fig. 7) and the radius of
curvature of the second bend equals R2 ¼ 110 m (see Fig. 8). The
tracking errors on the lateral deviation and yaw angle rise to 0.7 m
and 41 respectively in the bend, and at the same time the steering
angle oscillates. Note that the road adhesion μ is assumed to be
constant throughout the experimental trajectory.
Figs. 16–18 show the performance results in the case of speed
extrapolations for the steering vehicle control coupled with NL4WM.
These tests were carried out using the experimental data acquired at
Site 2, performed at 100 km h  1: three instances of the NL4W model
coupled with the switching PIDy and PDψ controllers were used with
speed increments of þ6 km h  1, þ12 km h  1 and þ18 km h  1. This
kind of test also proves the robustness and stability of the control
approach proposed here, since tracking errors remain moderate: even
in the most difficult left-hand bend, with an þ18 km h  1 speed
increment (i.e. when the speed reaches 118 km h  1), the lateral
deviation does not exceed 0.4 m (Fig. 17), when the lateral acceleration
Fig. 16. Site 2: 3 increases in speed for the NL4W model coupled with the
reaches 9 m s  2 (Fig. 17), and the front sideslip angles 71. Fig. 18 shows
switching PIDy and PDψ controllers: (a) increases in longitudinal speeds, that the tires are operating here in the nonlinear area, and have
(b) steering angles. reached their physical limits, since the lateral forces are saturated.

Fig. 17. Site 2: 3 increases in speed for the NL4W model coupled with the switching PIDy and PDψ controllers: (a) trajectories, (b) lateral deviation errors,
(c) yaw angle errors.
 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14 13

Fig. 18. Site 2: 3 increases in speed for the NL4W vehicle model coupled with the switching PIDy and PDψ controllers: tire lateral operating points.

Table 4
Geometrical characteristics of the bend at Site
2 and its maximum negotiating speed.

Site Site 2

Bend's road bank angle 8%

Bend's radius R2  110 m
Road friction coefficient μ 1
Speed limit found 118 km h  1

second site is approximately equal to 118 km h  1, as shown in

Fig. 16 between the abscissae 630 m and 750 m. This speed limit is
obtained when the road adhesion equal to 1.
Fig. 19 shows the activation of all criteria used to detect the
critical driving situation due to the high dynamic loads (see
Section 2 for the problem statement). These criteria are applied
1
on NL4WM þ SSVC, executed with V x þ 18 km h . The results
concerning the loss of control over the vehicle shown in Fig. 19 are
obtained using the following criteria:

 Understeer and oversteer driving situations using the Self-Steer

Fig. 19. Site 2: Detection of critical driving situation of NL4WM þ SSVC executed
1
with V x þ 18 km h . Gradient (oversteer if SSG 4 0, understeer if SSG o 0 and SSG ¼0
for neutral steer).
 Stability of the sideslip angle (j4β þ β_ j o 24Þ.
Fig. 18 shows the four tire lateral operating points for the trail  The maximum lateral acceleration is aymax . The detection is
performed with a speed increment of þ18 km h  1 and two other made if jay j Z aymax  0:8g.
models with speed increments of þ6 km h  1 and þ12 km h  1.  Load transfer ratio LTR ¼ ðF z1l  F z3l Þ=ðF z2r þ F z4r Þ ¼ ΔF zl =ðF z2r þ
Black curves represent the reference lateral forces obtained with F z4r Þ.
the Pacejka model for the 307 tire setting, for four values of the  The bound on the yaw motion that takes into account the
vertical load (1, 3, 5 and 7 kN). In this figure, the tires are operating physical limits of the vehicle aymax is ψ_ max ¼ aymax =V x . Detec-
in the nonlinear area, and the tires reach their physical limits, tion occurs where jψ_ j 4 ψ_ max .
when sideslip angles increase and lateral forces decrease. Figs. 17
and 18 prove that a virtual loss of control of the vehicle model Table 4 summarizes the geometrical characteristics and the speed
with the speed increment equal to þ18 km h  1 is imminent. From limit for negotiating the second bend at Site 2. The speed is obtained
the results obtained during the speed extrapolation test we may with the road friction coefficient μ equal to 1. The maximum speed is
deduce that the speed limit for negotiating the second bend at the identified through the evaluation of certain dynamic parameters of
14 L. Menhour et al. / Control Engineering Practice 24 (2014) 1–14

1
the extrapolated model at speed V x þ 18 km h using criteria Chong, U.-S., Namgoong, E., & Sul, S.-K. (1996). Torque steering control of 4-wheel
presented in the previous paragraph to detect the physical limits of drive electric vehicle. IEEE Power Electronics in Transportation, 21, 159–164.
Cole, D., Pick, A., & Odhams, A. (2006). Predictive and linear quadratic methods for
the vehicle. Note that all these criteria are reported and used in potential application to modelling driver steering control. Vehicle System
Koibuchi, Yamamoto, & Inagaki (1996), Odenthal, Bunte, & Ackerman Dynamics, 44(3), 259–284.
(1999), He, Crolla, Levesley, & Manning (2006), and Re, Allgower, Daafouz, J., Riedinger, P., & Iung, C. (2002). Stability analysis and control synthesis
for switched systems: A switched Lyapunov function approach. IEEE Transactions
Glielmo, Guardiola, & Kolmanovsky (2010). on Automatic Control, 47(11), 1883–1887.
Dugoff, H., Fanches, P., & Segel, L. (1970). An analysis of tire traction properties and
their influence on vehicle dynamic performance. In SAE (700377).
7. Conclusions Enache, N., Sebsadji, Y., Mammar, S., Lusetti, B., & Glaser, S. (2009). Driver's
influence on the performance of an integrated lane departure avoidance and
lane keeping assistance system. In IEEE-MSC (pp. 119–124), Saint Petersburg,
This work presents a switched steering vehicle LQR/H 1 control Russia.
strategy. This control strategy comprises switched PIDy control for Fujiwara, Y., & Adachi, S. (2002). Control design of steering assistance system for
lateral deviation and switched PDψ control for yaw angle. All PIDy driver characteristics. In IEEE proceedings of the 41st SICE annual conference.
Fung, H.-W., Wang, Q.-G., & Lee, T.-H. (1998). Pi tuning in terms of gain and phase
and PDψ controllers are designed using the proposed switched margins. Automatica, 34, 1145–1149.
LQR/H 1 control method. This method is based on the LQR control Ge, M., Chiu, M., & Wang, Q. (2002). Robust pid controller design via lmi approach.
problem, the H 1 norm, LMI optimization and a common Lyapunov Journal of Process Control, 12, 3–13.
Hang, C. C., Astrom, K. J., & Wang, Q. G. (2002). Relay feedback auto-tuning of
function method. Notice that the common Lyapunov function process controllers—a tutorial review. Journal of Process Control, 12, 143–162.
approach is generally used for the design and stability analysis He, J., Crolla, D. A., Levesley, M. C., & Manning, W. J. (2006). Coordination of active
of switched linear continuous-time systems. A comparison of the steering, driveline, and braking for integrated vehicle dynamics control. IMechE
Journal of Automobile Engineering, 220, 1401–1421.
results obtained with actual experimental data demonstrates:
Ho, W., Honga, Y., Hanssonb, A., Hjalmarssonc, H., & Denga, J. (2003). Relay auto-
tuning of pid controllers using iterative feedback tuning. Automatica, 39,
 the satisfactory performances of the control strategy with 149–157.
respect to nonlinearities and parameter variations, even where Jazar, R. N. (2008). Vehicle dynamics: Theory and application. US: Springer.
Koenig, D., Marx, B., & Jacquet, D. (2008). Unknown input observers for switched
the road friction coefficient is low (μ A ½0:7; 1Þ, nonlinear discrete time descriptor systems. IEEE Transactions on Automatic
 the accuracy in terms of lane keeping errors and computed Control, 53, 373–379.
steering angle (a positive experimental validation of the main Koibuchi, K., Yamamoto, M., & Inagaki, S. (1996). Vehicle stability in limit cornening
by active brake. In SAE international (pp. 26–29), Michigan.
dynamic parameters of the linear and nonlinear vehicle Lechner, D. (2008). Embedded laboratory for vehicle dynamic. In International
models), symposium on advanced vehicle control, Kobe, Japan.
 the performances of the switching steering control under high Liberzon, D., & Morse, A. S. (1999). Basic problems in stability and design of
switched system. IEEE Control Systems Magazine, 19(5), 59–70.
lateral accelerations (ay 4 8 m s  2 ) and sideslip angles Lin, C., Wang, Q. G., & Lee, T. H. (2004). An improvement on multivariable pid
(α1;2 4 61), where the tires reach their physical limits (or using controller design via iterative lmi approach. Automatica, 40, 519–525.
speed extrapolation tests). Lin, H., & Antsaklis, P. J. (2009). Stability and stabilizability of switched linear
systems: A survey of recent results. IEEE Transactions on Automatic Control, 54(2),
308–322.
The potential of this application that aims to identify the Marino, R., & Cinili, F. (2009). input-output decoupling control by measurement
maximum speed of a vehicle negotiating a bend through the use feedback in four wheel-steering-vehicles. IEEE Transaction on Control Systems
Technology, 17, 1163–1172.
of speed extrapolation and a switched steering vehicle control
Mendel, J. (2004). Computing derivatives in interval type-2 fuzzy logic systems.
appears very promising. The results obtained here demonstrate IEEE Transactions on Fuzzy Systems, 12, 84–98.
that it will be possible in the future to deploy this application to Menhour, L., Lechner, D., & Charara, A. (2011). Two degrees of freedom pid multi-
enhance road databases, or to provide information to be displayed controllers to design a mathematical driver model: Experimental validation and
robustness tests. Vehicle System Dynamics, 49, 595–624.
on road signs with variable messages. Menhour, L., Lechner, D., & Charara, A. (2012). Design and experimental validation
The design method of switched discrete-time steering vehicle of linear and nonlinear vehicle steering control strategies. Vehicle System
control using switched Lyapunov functions will be investigated in Dynamics, 50(6), 903–938.
Milliken, W. F., & Milliken, D. L. (1995). Race car vehicle dynamics. Society of
future work. Automotive Engineers International.
Nouvelière, L., & Mammar, S. (2007). Experimental vehicle longitudinal control
References using a second order sliding mode technique. Control Engineering Practice, 15,
943–953.
Odenthal, D., Bunte, T., & Ackerman, J. (1999). Nonlinear steering and breaking
Ackermann, J., Guldner, J., Sienel, W., Steinhauser, R., & Utkin, V. I. (1995). Linear control for vehicle rollover avoidance. In European control conference, Karlsruhe,
and nonlinear controller design for robust automatic steering. IEEE Transaction Germany.
on Control Systems Technology, 3(1), 132–143. Pacejka, H. B. (2005). Tire and vehicle dynamics. Netherlands: Elsevier.
Apkarian, P., Gahinet, P., & Becker, G. (1995). Self-scheduled h1 control of linear Plochl, M., & Edelmann, J. (2007). Driver models in automobile dynamics applica-
parameter-varying systems: A design example. Automatica, 31(9), 1251–1261. tion. Vehicle System Dynamics, 45, 699–741.
Astrom, K. J., Panagopoulos, H., & Hagglund, T. (1998). Design of pi controllers based Poussot-Vassal, C., Sename, O., Dugard, L., Gaspar, P., Szaboc, Z., & Bokor, J. (2011).
on non-convex optimization. Automatica, 34, 585–601. Attitude and handling improvements through gain-scheduled suspensions and
Blanchini, F., & Miani, S. (2008). Set-theoretic methods in control. Boston, Berlin: brakes control. Control Engineering Practice, 19, 252–263.
Springer, Basel. Rajamani, R. (2006). Vehicle dynamics and control. USA: Springer.
Boyd, S., Ghaoui, L. E., Feron, E., & Balakrishnan, V. (1994). Linear matrix inequalities Re, L., Allgower, F., Glielmo, L., Guardiola, C., & Kolmanovsky, I. (2010). Automotive
in system and control theory. Philadelphia: SIAM. model predictive control: Models, methods and applications. In Lecture notes in
Branicky, M. (1998). Multiple Lyapunov functions and other analysis tools for control and information sciences. Springer-Verlag.
switched and hybrid systems. IEEE Transactions on Automatic Control, 43(4), Stilwell, D. J., & Rugh, W. J. (1999). Interpolation of observer state feedback
475–482. controllers for gain scheduling. IEEE Transactions on Automatic Control, 44(6),
Burckhardt, M. (1993). Fahrwerktechnik, Radschlupfregel-systeme. Germany: Vogel- 1225–1229.
Verlag. Sugeno, M., & Kang, G. (1988). Structure identification of fuzzy model. Fuzzy Sets
Busawon, K. K., & Kabore, P. (2001). Disturbance attenuation using integral and Systems, 28(1), 15–33.
observers. International Journal of Control, 74(6), 618–627. Sun, Z., & Ge, S. S. (2005). Analysis and synthesis of switched linear control systems.
Castillo, O., & Melin, P. (2008). Type-2 fuzzy logic: Theory and applications. Springer Automatica, 41, 181–195.
Verlag. Zheng, S., Ho, Tang, Han, Z., & Zhang, Y. (2006). Controller design for vehicle
Cerone, V., Milanese, M., & Regruto, D. (2009). Combined automatic lane-keeping stability enhancement. Control Engineering Practice, 14, 1413–1421.
and driver's steering through a 2-dof control strategy. IEEE Transactions on Ziegler, J., & Nichols, N. B. (1942). Optimum setting for automatic controllers.
Control Systems Technology, 17, 135–142. Transactions of the ASME, 64, 759–768.