Vehicle longitudinal motion modeling for nonlinear

control
Khalid El Majdoub, Fouad Giri, Hamid Ouadi, Luc Dugard, Fatima Zara
Chaoui

To cite this version:

Khalid El Majdoub, Fouad Giri, Hamid Ouadi, Luc Dugard, Fatima Zara Chaoui. Vehicle longitudinal
motion modeling for nonlinear control. Control Engineering Practice, Elsevier, 2012, 20 (1), pp.69 -
81. <10.1016/j.conengprac.2011.09.005>. <hal-00638489>

HAL Id: hal-00638489

https://hal.archives-ouvertes.fr/hal-00638489
Submitted on 7 Nov 2011

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Vehicle Longitudinal Motion Modeling for nonlinear control

K. El Majdoub c, F. Giri a,*, H. Ouadic, L. Dugardb, F.Z. Chaouic

a,*
GREYC Lab, UMR CNRS, University of Caen Basse-Normandie, Caen, France
b
GIPSA Lab, UMR CNRS, INPG, Grenoble, France
c
ENSET, Universite´ de Rabat-Agdal, Rabat, Marocco

Abstract— The problem of modeling and controlling vehicle longitudinal motion is addressed for
front wheel propelled vehicles. The chassis dynamics are modeled using relevant fundamental laws
taking into account aerodynamic effects and road slop variation. The longitudinal slip, resulting from
tire deformation, is captured through Kiencke‟s model. A highly nonlinear model is thus obtained and
based upon in vehicle longitudinal motion simulation. A simpler, but nevertheless accurate, version of
that model proves to be useful in vehicle longitudinal control. For security and comfort purpose, the
vehicle speed must be tightly regulated, both in acceleration and deceleration modes, despite
unpredictable changes in aerodynamics efforts and road slop. To this end, a nonlinear controller is
developed using the Lyapunov design technique and formally shown to meet its objectives i.e. perfect
chassis and wheel speed regulation.

Keywords – vehicle longitudinal control, longitudinal slip, tire Kiencke‟s model, speed control,
Lyapunov stability.

1. INTRODUCTION
Vehicle longitudinal motion control aims at ensuring passenger safety and comfort. It is an important
aspect in dynamic collaborative driving i.e. when multiple vehicles should coordinate to share road
efficiently while maintaining safety. In this respect, several works have been devoted to what is
commonly referred to adaptive cruise control that consists in maintaining a specified headway between
vehicles (Ioannou and Chien, 1993; Moon et al., 2009). Different control techniques have been used in
these works including linear and adaptive control (You et al., 2009), genetic fuzzy control (Poursamad
and Montazeri, 2008), sliding mode control (Liang et al., 2003; Nouveliere and Mammar, 2007), and
scheduling gain control involving PIDs (Ren et al., 2008). However, most previous works on
longitudinal control were based on simple models neglecting important nonlinear aspects of the
vehicle such as rolling resistance, aerodynamics effects and road load. In some studies, the controller
performances were not formally analyzed (Ren et al., 2008). In (Yamakawa et al., 2007), longitudinal
vehicle control has been studied focusing on torque management for independent wheel drive. It is
worth noticing that in all previous studies on longitudinal vehicle control, the control design has been
based on simple models not accounting for tire-road interaction.

1
In the present study, the problem of longitudinal vehicle control is revisited, for front wheel propelled
vehicles, focusing on speed regulation. The aim is to design a controller that is able to tightly regulate
the chassis and wheel velocities, in both acceleration and deceleration driving modes, despite changing
and uncertain driving conditions. This problem has not been dealt with previously. A further
originality of the present paper is that the control design relies upon a more complete model that
accounts for most vehicle nonlinear dynamics including tire-road interaction. That is, the study
includes two major contributions. First, a suitable control model is developed for the vehicle
longitudinal behavior. In this respect, recall that a convenient model is one that is sufficiently accurate
but remains simple enough to be utilizable in control design. To meet the accuracy requirement, the
model must account not only for aerodynamic phenomena but also, and especially, for tire-road
friction. Modeling the tire/road contact is a quite complex issue involving multiple aspects relevant to
tire characteristics (e.g. structure, pressure) and to environmental factors (e.g. road load, temperature).
Several tire models have been proposed in the literature e.g. Guo‟s model (Guo and Ren, 2000),
Pacejka‟s model (Pacejka and Besselink , 1997), Dugoff‟s model (Dugoff and Segel, 1970), Gim‟s
model (Gim and Nikravesh, 1990), Kiencke‟s model (Kiencke and Nielsen, 2004). In the present work,
Kiencke‟s model is retained because it proves to be a good compromise between accuracy and
simplicity. The overall vehicle modeling is carried out according to the bicycle model principle. In
addition to tire equations, the model includes chassis dynamics equations (obtained from fundamental
dynamics and aerodynamics laws) and incorporates relevant practical prior knowledge e.g. the tire
longitudinal slip is physically limited. The overall vehicle model turns out to be a combination of two
nonlinear state-space representations describing, respectively, the acceleration and deceleration
longitudinal driving modes. Its high complexity makes it hardly utilizable for control design but, due
to its high accuracy, it proves to be quite suitable for simulator building. This model development is
one major achievement of the present study. The second contribution is the design of a nonlinear
controller that ensures global stabilization and longitudinal speed regulation during
acceleration/deceleration driving modes. This is carried out using the Lyapunov design technique
(Khalil, 2002), based on a simpler (but still accurate) version of the above simulation-oriented model.
It is formally proved that the developed controller actually achieves the stability and regulation
objectives it was designed to. Furthermore, it is observed through numerical simulations that the
controller is quite robust with respect to uncertainties on environmental characteristics.
The paper is organized as follows: Section II is devoted to modeling the acceleration/deceleration
vehicle longitudinal behavior; the obtained model is used in Section III to design a controller and to
analyze the resulting closed-loop system; the controller performances are illustrated in Section IV by
numerical simulations. A conclusion and reference list end the paper.

2
2. MODELLING OF CHASSIS LONGITUDINAL MOTION

Except for aerodynamic forces, all external efforts acting on a vehicle are generated at the wheel-road
contact. The understanding and modeling of the forces and torques developed at wheel-road contact is
essential for studying properly the vehicle dynamics. These are discussed in the forthcoming
subsections. In this respect, recall that the vehicle motion is composed of two types of displacements:
translations along the x, y, z axes and rotations around these same axes (Fig. 1).

2.1. Kiencke’s Tire Modeling

The tire is a main component of the wheel-road contact as it ensures three important functions
(Kiencke and Nielsen, 2004): (i) bearing the vertical load and absorbing road deformations; (ii)
producing longitudinal acceleration efforts and contributing to vehicle braking; (iii) producing the
required transversal efforts that help the vehicle turning.
The efforts generated at the wheel-road contact include longitudinal (acceleration/deceleration) forces,
lateral guiding forces and self alignment torque. The effect of these efforts on the vehicle behaviour is
determined by the tire-road adhesion. For small load variations, the longitudinal coefficient of friction
is characterized by the following ratio:
Ftx
(1)
Fv

where Ftx denotes the longitudinal effort and Fv the vertical load. The ratio is called longitudinal
adhesion or friction coefficient. The value of this coefficient depends on the tire slip resulting from the
deformation of the tire in contact with the road (Kiencke and Nielsen, 2004). The longitudinal slip is
characterized by the coefficient defined as follows:
. in acceleration mode, i.e. VV VW , one has:
VV VV
1 1 (2)
VW reff W

z Vertical
ψ motion

Roll
φ Pitch
Longitudinal x Transversal
motion y motion

Fig. 1. Degrees of freedom of a vehicle

3
. in deceleration mode, i.e. VV VW , one has:

VW reff W
1 1 (3)
VV VV

where reff denotes the effective wheel radius, W
designates the wheel angular velocity, Vw is the

speed of the tire-road contact, Vv is the linear velocity of the wheel centre (Fig. 2). A similar

deformation occurs when the wheel presents a slip angle i.e. the resulting lateral slip produces a
lateral force F ty .
Modeling the efforts at the wheel-road contact has been given a great deal of interest over the last
years. In this respect, several tire models have been developed with quite different properties, e.g.
(Guo and Ren, 2000; Pacejka and Besselink , 1997; Dugoff and Segel, 1990; Gim and Nikravesh,
1990; Kiencke and Nielsen, 2004). For control design use, the most suitable tire model is one that
presents the best accuracy/simplicity compromise. From this viewpoint, Kiencke‟s model turns out to
be a quite satisfactory choice (Kiencke and Nielsen, 2004). Indeed, this model is sufficiently accurate

as it accounts for the main features such as the vertical load Fv , slip angle , slip coefficient . On
the other hand, it has already proved to be useful in designing simple estimators for state variables like
slip angle and lateral efforts (You et al., 2009). In the present paper, this model will prove to be useful
in control design.

2.2. Kiencke’s Model

This was developed in (Kiencke and Nielsen, 2004) using the Burckhardt‟s extended model to
compute the friction coefficient . Accordingly, the latter is a function of the combined
longitudinal/lateral slip coefficient and the forces acting on the tire. Fig. 3 gives a schematic
representation of Kiencke‟s wheel model where:
C1 1 exp( C2 ) C3 exp( C4VG )(1 C5 Fv2 ) (4)

2 2
y
(5)
z y
Ft
F ty
W
VV
reff F tx
F tx VW
y
F ty z
Mtz
VV
Fig. 2. Forces applied on the wheel

4
 Fv Ftx

Fty

Fig. 3. Kiencke‟s wheel model

VV
1 and y (1 ) tan( ) (Acceleration) (6)
VW
VW
1 and y tan( ) (Deceleration) (7)
VV
The other parameters and variables contained in equations (4) to (7) have the following meanings:
C1 , C2 , C3 : parameters depending on the road state.

C4 : coefficient depending on the maximal driving speed.

C5 : coefficient depending on the wheel allowed maximal load.
: global (longitudinal/lateral) slip.

y : lateral slip.

VG : speed of the vehicle centre of gravity.

FV : vertical load.
In Kiencke‟s model, the longitudinal and lateral efforts are respectively described by (8)-(9):
Fv
Ftx cos( ) C t y sin( ) (8)

Fv
Fty C t y cos( ) sin( ) (9)

where C t denotes a weighting coefficient taking values in the interval 0.9 0.95 . In longitudinal
motion, the following simplifications are used:
(i) The tire behaviour is independent of the vehicle maximal speed. That is, the term in C4 is
neglected in equation (4).
(ii) The tire model depends linearly on the maximal load. Then, the term in C5 is neglected in (4).

(iii) The vehicle moves along a straight line. That is, the slip angle is null.
Using these remarks, equations (4) to (9) simplify to:
C1 1 exp( C2 ) C3 , , y 0 (10)

and y 0 (11)

Ftx Fv (12)

Fty 0 (13)

5
2.3. Rolling Resistance and Aerodynamic Resistance

It is obvious that tire deformation causes mechanical losses. Radial deformations are caused by the
vertical load. Fig. 4 shows that the vertical force distribution (along the deformation area) is not
uniform. The resulting force moves from the central point I to a point located at a distance d rr ,max (Fig

4). This is called vertical load maximal trail and is defined as follows:

d rr ,max rr reff (14)

When the wheel starts moving, a torque is generated provided that d rr ,max 0 . To bring back the

vertical load strength to the central point I , a compensating torque M rr must be applied (by the
driving motor) on the wheel. This is called rolling resistance torque and is given by:

M rr Fv d rr ,max Fv rr reff (15)

2.4. Aerodynamic Resistance

Aerodynamic resistance has naturally an impact on energy consumption onboard. Fluid mechanics
laws are resorted to explain air flow around a moving vehicle. Accordingly, vehicle forms are
continuously reinvented to improve aerodynamic performances (Power and Nicastri, 2000).
Aerodynamic efforts come in direct interaction with the vehicle, producing various forces (drag, lift
and lateral) which in turn generate torques (yaw, roll and pitch). The aerodynamic drag force is
significant when the vehicle moves at grand speed. The aerodynamic lift effort increases the rolling
resistance (because the surface of tire-road contact grows up) improving tire steerability. On the other
hand, the aerodynamic lift force decreases the vehicle adhesion which reduces its stability and security
(Milliken and Milliken, 1995). In presence of front-wind, the aerodynamic resistance is represented by

two forces: the aerodynamic drag force F aex and the aerodynamic lift force F aez . These are defined as
follows:
1
Faex C x S (VV Va ) 2
2 (16)
1
Faez C z S (VV Va ) 2
2

reff

I Fv
drr

Fig. 4. Rolling resistance

6
with:
: air density depending on atmosphere pressure and ambient temperature.
Cx : aerodynamic drag coefficient.
Cz : aerodynamic lift coefficient.
S : frontal projection area vehicle.
VV : vehicle speed.
Va : wind speed (positive in presence of front-wind, negative in presence of rear-wind).

2.5. Modeling of a Wheel Submitted to Driving Couple

Fig. 5 shows a one-wheel vehicle with mass M v . The wheel is driven by a couple M m . Let J denotes
the inertia resulting from the wheel, the transmission shaft and the driving motor. Invoking the
dynamic fundamental principle, one gets the following equation:
d W (18)
Mm J Ft reff M rr
dt
where equation (15) has been used to account for the rolling resistance. Using the relation VW reff W ,

one gets:
reff
Vw Mm Ft reff Fv reff rr (19)
J
Equation (19) together with (1), (2), (3) and (10) describe the one-wheel vehicle behavior. This is
further illustrated by the schematic representation of Fig. 6.

2.6. Model of Two-Wheel Vehicle with One Driving Wheel

Longitudinal and transversal behaviors can be assumed to be decoupled when the steering angle is
small. Then, we make use of the vehicle symmetry to perform a projection (of all forces) on the
longitudinal axis reducing thus the four-wheel model into a (two-wheel). Fig. 7 illustrates the forces
involved in a bicycle model. The involved notations are described in Table I. Now, the focus will be
made on the bicycle model considering that only one wheel is submitted to a motor torque M m . For
vehicle stability purpose, the front wheel is driving. The distribution of the vehicle load over the tires
is computed applying the two dynamic fundamental laws. Doing so, one gets the following equations:
Ftf ( M v g sin Faex ) M vVv (20)

Fvf Fvr ( M v g cos Faez ) 0 (21)

Fvf l f Fvrlr ( Ftf Ftr )h 0 (22)

Solving equations (21) and (22) for Fvf and Fvr yields:

7
 Body of mass Mv Mm VW

W
Fv
Fv
Vv Ft
Ft

Fig. 5. One-wheel vehicle model Fig. 6. Schematic representation of a one-wheel

vehicle submitted to a driving couple

h F aez
F aex
CoG

F tf 
F vf
Mg
lf
Mvg F tr
lr F vr

Fig. 7. The forces acting on a bicycle type vehicle

Vv Faez Faex

Fvf M v g (1 ) cos ( sin ) (1 ) (23)
g Mvg Mvg

Vv Faez Faex

Fvr Mvg cos ( sin ) (24)
g Mvg Mvg

h lf
where and . From equation (20) it is easily seen that the vehicle speed
lf lr lf lr

undergoes the following equation:

1
Vv Ftf ( M v g sin Faex ) (25)
Mv

TABLE I. NOTATIONS OF VEHICLE LONGITUDINAL MODEL.

lf : Distance between CoG and the front wheel base (m)
lr : Distance between CoG and the rear wheel base (m)
l : Distance between the bases of the two wheels (m)
h : Height of the gravity centre (m)
Faex : Aerodynamic drag force (N)
Faez : Aerodynamic carrying force (N)
g : Gravity acceleration (m.s-2)
Mv : Vehicle mass (kg)
Ftf, Ftr : Front and rear wheel drive force (N)
Fvf, Fvr : Load on the front and rear wheel (N)
θ : Road slop (rad)

8
2.7. State-Space Representation of Vehicle Longitudinal Behavior

The equations obtained so far are now combined together to build-up a state-space representation of
the vehicle longitudinal acceleration/deceleration behavior. The vehicle longitudinal dynamics are
characterized by two state variables, i.e. vehicle (chassis) speed Vv and front-wheel speed Vw . As the
slip coefficient depends on the current driving mode (acceleration or deceleration), the vehicle is
characterized by two state-space representations. Each representation describes the vehicle in the
corresponding operation mode.

2.7.1. State-Space Representation in Deceleration Mode Vw Vv

Combining (3), (10), (11), (13) and (15)-(18), one obtains the following state-space representation:
1
Vw Vw Vw
Vw 1 Mm 2 3 4 exp( ) 5 6 7 (Vv Va ) 2
Vv Vv Vv

Vw Vw Vw
8 (Vv Va ) 2 9 exp( ) 10 (Vv Va ) 2 exp( ) (26a)
Vv Vv Vv
1
Vw Vw Vw
Vv 11 12 (Vv Va ) 2
2 3 4 exp( ) 13 14 15 (Vv Va ) 2
Vv Vv Vv

Vw Vw Vw
16 (Vv Va ) 2 17 exp( ) 18 (Vv Va ) 2 exp( ) (26b)
Vv Vv Vv

where the various parameters are defined in Table II. A more compact representation of (26a-b) is
obtained introducing the notations: u M m , x1 Vw , x2 Vv and:
1
x1 x1 x1
f1 ( x1 , x2 ) 2 3 4 exp( ) 5 6 7 ( x2 Va ) 2
x2 x2 x2

x1 x1 x1
8 ( x2 Va ) 2 9 exp( ) 10 ( x2 Va ) 2 exp( ) (27a)
x2 x2 x2
1
2 x1 x1 x1
f 2 ( x1 , x2 ) 11 12 ( x2 Va ) 2 3 4 exp( ) 13 14 15 ( x2 Va ) 2
x2 x2 x2

x1 x1 x1
16 ( x2 Va ) 2 17 exp( ) 18 ( x2 Va ) 2 exp( ) (27b)
x2 x2 x2

With the above notations, the state-space representation (26a-b) is given the following usual compact
form:
x1 1 u f1 ( x1 , x2 )
(28)
x2 f 2 ( x1 , x2 )

9
 TABLE II. DEFINITION OF THE MODEL PARAMETERS
Vehicle parameters in deceleration Vehicle parameters in acceleration
: C2 ' : C2
reff reff
1 : '1 :
J
J
2 : 1 kv (C1 C3 ) '2 : 1 kv (C1 C3 )
3 : k v C3 '3 : k v C3

4 : k v C1 exp( C 2 ) '4 : k v C1 exp( C 2 )

2
: r
eff '5 : reff2
5
(1 ) M v g cos ( rr kv (C1 C3 )) (1 ) M v g cos ( rr kv (C1 C3 ))
J J
: (1 reff2 '6 : reff2
6 ) M v g cos k v C3 (1 ) M v g cos k v C3
J J
2
r reff2
7 : 1 (1 ) SC z ( rr k v (C1 C3 ))
eff '7 : 1
(1 ) SCz ( rr kv (C1 C3 ))
2 J 2 J
2 r2
: 1 reff '8 : 1 eff
8 (1 ) SC z kvC3 (1 ) SCz kvC3
2 J 2 J
2
: (1 r eff '9 : reff2
9 ) M v g cos k v C1 exp( C2 ) (1 ) M v g cos kv C1 exp( C2 )
J J
2 2
10
: 1 reff '10 : 1 reff
(1 ) SCz kv C1 exp( C2 ) (1 ) SC z kvC1 exp( C2 )
2 J 2 J
11 g sin( ) '11 : g sin( )
1 ' 1
12 : SC x 12 : SCx
2M v 2M v

13 : (1 ) g cos k v (C1 C3 ) '13 : (1 ) g cos k v (C1 C3 )

14 : (1 ) g cos k v C3 '14 : (1 ) g cos kv C3

1 '15 1
15 : (1 ) SC z kv ( C1 C3 ) : (1 ) SC z kv ( C1 C3 )
2M v 2M v
16 : 1 '16 : 1
(1 ) SC z k v C3 (1 ) SCz kv C3
2M v 2M v
17 : (1 ) g cos kv C1 exp( C2 ) '17 : (1 ) g cos kv C1 exp( C2 )

18 : 1 '18 : 1
(1 ) SCz k v C1 exp( C2 ) (1 ) SC z k v C1 exp( C 2 )
2M v 2M v

2.7.2. State-Space Representation in Acceleration Mode Vv Vw

Using (2), (10), (11), (13) and (15)-(18), one gets the following state-space representation:
1
Vv Vv Vv
Vw '
1 Mm '
2
'
3
'
4 exp( '
) '
5
'
6
'
7 (Vv Va ) 2
Vw Vw Vw

' Vv Vv Vv
8 (Vv Va ) 2 '
9 exp( '
) '
10 (Vv Va ) 2 exp( '
) (29a)
Vw Vw Vw

10
 1
Vv Vv Vv
Vv '
11
'
12 (Vv Va ) 2 '
2
'
3
'
4 exp( '
) '
13
'
14
'
15 (Vv Va ) 2
Vw Vw Vw

' Vv Vv Vv
16 (Vv Va ) 2 '
17 exp( '
) '
18 (Vv Va ) 2 exp( '
) (29b)
Vw Vw Vw

where the various parameters are defined in Table II. Let us introduce the notations:

1
' ' ' x2 ' ' x2 ' ' x2 ' x2
f ( x1 , x2 )
1 2 3 4 exp( ) 5 6 7 ( x2 Va ) 2 '
8 ( x2 Va ) 2
x1 x1 x1 x1

' ' x2 ' x2

9 exp( ) 10 ( x2 Va ) 2 exp( '
) (30a)
x1 x1
1
' ' ' 2 ' ' x2 ' ' x2 ' ' x2 '
f ( x1 , x2 )
2 11 12 ( x2 Va ) 2 3 4 exp( ) 13 14 15 ( x2 Va ) 2
x1 x1 x1

' x2 x2 x2
16 ( x2 Va ) 2 '
17 exp( '
) '
18 ( x2 Va ) 2 exp( '
) (30b)
x1 x1 x1

where u, x1 and x2 are as in (27a-b). With these notations, the state-space representation (30a-b) is
given the following more compact form:

'
x1 1 u f1' ( x1 , x2 )
(31)
x2 f 2' ( x1 , x2 )

2.8 Control- and Simulation-Oriented Models for Two-Wheel Vehicle With Single Driving Wheel

2.8.1 Control-Oriented Models

2.8.1.1 Control-Oriented Model Accounting For Tire Dynamics

Combining the mode-dependent state space representations (28) and (31), one gets a single model
representing the vehicle in all operation modes i.e.
*
x1 1 ( x1 , x2 ) u g1 ( x1 , x2 )
(32)
x2 g 2 ( x1 , x2 )
with:
g1 ( x1 , x2 ) ( x1 , x2 ) f1 ( x1 , x2 ) 1 ( x1 , x2 ) f1' ( x1 , x2 ) (33a)

g 2 ( x1 , x2 ) ( x1 , x2 ) f 2 ( x1 , x2 ) 1 ( x1 , x2 ) f 2' ( x1 , x2 ) (33b)
* '
1 ( x1 , x2 ) ( x1 , x2 ) 1 1 ( x1 , x2 ) 1 (33c)
1 sign( x1 x2 )
( x1 , x2 ) (33d)
2

11
In addition to vehicle aerodynamics, this model does account for tire-road contact effect. Furthermore,
it will prove to be useful in control design (Section III).

2.8.1.2 Control-Oriented Model Ignoring Tire Dynamics

To better appreciate the benefit of accounting of tire-road contact, a comparison will be performed in
Section IV between the controller obtained from (32)-(33) and the one obtained from a simplified
model neglecting tire-road contact. Specifically, the simplified model is obtained letting 0 (tire
sliding negligence) which immediately implies that Vv Vw (i.e. x1 x2 ). Ignoring also the rolling

resistance rr , one gets invoking the dynamic fundamentals principles for translation and rotation:
reff
Vw Mm Ftf reff (34)
J
1
Vv Ftf ( M v g sin Faex ) (35)
Mv

where Faex is defined by (16). It is readily seen from (34)-(35) that the vehicle speed undergoes the
following equation:
reff reff2 M v 1
Vv 2
Mm 2
g sin SCx (Vv Va ) 2 (36)
J r Mv
eff J r Mv
eff 2M v

where the different parameters are defined in Table V. Introduce the following notations:
reff
(37a)
J reff2 M v

reff2 M v 1
f ( x2 ) 2
g sin SCx ( x2 Va ) 2 (37b)
J r Mv
eff 2M v

Then, equation (36) can be given by the following more compact form where u M m and x2 Vv :

x 2 u f ( x2 ) (38)
The first-order equation (38) is a simplified version of the second-order model (32)-(33). As already
mentioned, both models will be based upon in control design and the obtained controllers will be
compared later in Section IV.

2.8.2 Simulation Model

Obvious physical considerations show that in real-life longitudinal motions, the vehicle and wheel
speeds are always quite close to each other. On the other hand, there is no guarantee that the model
(32)-(33) ensures always that x1 x2 . In the next lines, the model will be slightly modified so that
x1 x2 becomes a structural property of it. Doing so, one also will discard any risk of singularity in
'
(27a-b) or (30a-b) by ensuring that ( x1 , x2 ) 0 and ( x1 , x2 ) 0 , with:

12
 x1 x1
( x1 , x2 ) 2 3 4 exp
x2 x2
(39)
' ' ' x2 ' ' x2
( x1 , x2 ) 2 3 4 exp
x1 x1
'
Indeed, it is readily seen that the functions and are continuous and:
( x1 , x2 ) 2 3 4 exp 0
' ' ' ' '
if x1 x2
( x1 , x2 ) 2 3 4 exp 0
'
The continuity of and then guarantees the existence of h 0 and 0 l 1 , such that:

inf ( x1 , x2 ) 0
x1
1 l 1
x2
' (40)
inf ( x1 , x2 ) 0
x1
1 1 h
x2

' '
As a matter of fact, the size of h and l depends on the parameters , , i , i (i 2, 3, 4) . The

above result, shows that equations (32)-(33) are representative of the vehicle longitudinal behavior as
long as the state vector ( x1 , x2 ) stays in the following validity domain:

x1
Dv ( x1 , x2 ) IR 2 : 1 l 1 h (41)
x2

A more realistic vehicle longitudinal model is one that structurally enforces the state variables ( x1 , x2 )
x1
to stay always in the above domain. To this end, introduce the new variable x and its time-
x2
1 x
derivative x x1 x2 . Then, equation (32) can be rewritten in term of the couple ( x, x 2 ) as
x2 x2
follows:
x h( x, x2 , u )
(42)
x 2 g 2 ( x, x2 )

with:
1 *
h ( x, x 2 , u ) 1 u g1 ( x, x2 ) xg 2 ( x, x2 ) (43)
x2
It is readily seen from (42) that:
t
x(t ) h( x, x2 , u )d x(0) (44)
0

Then, the prior knowledge (41) is accounted for taking the saturated version of (44) i.e.:
x(t ) sat ( z (t )) (45)
with

13
 def t
z (t ) h ( x, x 2 , u ) d x ( 0) (46)
0

where sat (.) denotes the saturation function defined by:

1 l if z 1 l
sat ( z ) z if 1 l z 1 h (47)
1 h if z 1 h

It is readily seen from (46) that the auxiliary variable z (t ) undergoes the differential equation:

z(t ) h( x, x2 , u ) and z (0) x(0) (48)

Then, it follows from (45) that, see Fig. 8.
x r ( z )h( x, x2 , u ) (49)
where
d
r ( z) sat ( z ) (50)
dz
It is easily checked using (47) that r (.) is the unit rectangular function defined by:

0 if z 1 l
r ( z) 1 if 1 l z 1 h (51)
0 if z 1 h

Putting the second equation in (42) together with (48)-(49), one gets the new model:
z (t ) h( x, x 2 , u )
x r ( )h( x, x 2 , u ) with z (0) x(0) and x1 x x2 (52)
x 2 g 2 ( x, x 2 )
As it accounts for the prior knowledge (47), the new model (52) turns out to be more accurate than the
models (32)-(33) and (38). However, model (52) is too complex to be used for control design, due e.g.
to the nonsmooth nature of the function r (z ) defined by (51). Therefore, that model can (and presently
will) only be used to build up simulators of the vehicle longitudinal motion. Model (32)-(33) is a quite
satisfactory compromise between model (38) (simpler but less accurate) and model (52) (more
accurate but more complex).

Example 1. To illustrate previous and forthcoming results, the example of a Citroën-2CV car, with the
characteristics of Table VI, will be considered throughout in the rest of the paper. Using Table II and
'
Table VI, one gets the numerical values of the parameters in the functions and defined by (39),
see Table III.
It is readily seen, by simple checking, that conditions (40a) are fulfilled. Furthermore, it is seen from
x1
Fig. 9a that ( x1 , x2 ) 0 if 0.93 1 . On the other hand, Fig. 9b shows that ' ( x1 , x2 ) 0
x2

14
 x(0)

u x x1
h ( x, x2 , u ) sat (.)
.
z

x2
g 2 ( x, x2 ) . x2

Fig. 8. Synoptic scheme of the simulation vehicle model (52)

(a) (b)

Figs. 9.a. ( x1 , x2 ) vs x1 / x 2 Fig. 9.b. ' ( x1 , x2 ) vs x 2 / x1

TABLE III. VALUES OF THE PARAMETERS IN (34) FOR THE CITROËN-2CV CAR
Deceleration Acceleration
: -23.99 ' : 23.99
2
: 1.36002 '
2
: 1.15202
3
: -0.104 '
3
: 0.104
4
: -6714265556.98732 '
4
: -9.76223532472411 10-12

x2 x1
whenever 0 1 i.e. 1 . In the light of these observations, it is seen that the validity
x1 x2

x1
domain is 0.93 . That is, l1 0.07 and h1 . On the other hand, the tire cannot assume
x2

(in acceleration as well as in deceleration mode) a sliding larger than max 10% i.e. max .

x1
Then, one gets from (2) and (3) that 0.9 1.11 which gives l2 0.1 and h2 0.11 .
x2

x1
Letting l min l1 , l2 0.07 and h min h1 , h2 0.11, one gets that 0.93 1.11 . The
x2

15
resulting bounding on the sliding turn out to be l h with l l 0.07 and

h
h 0.1
1 h

3. VEHICLE LONGITUDINAL REGULATORS DESIGN

3.1. Reference Trajectory Generation

We seek a speed controller for the vehicle moving in longitudinal acceleration/deceleration modes.
The regulator design is based on the vehicle model (32) and the control objective is to enforce the
vehicle chassis and wheel speeds, x1 Vw and x2 Vv , to track their reference trajectories, denoted Vw*

and Vv* , respectively. The latter are required to be time-differentiable. This requirement that can
always be complied with by pre-filtering given (non-differentiable) speed setpoints:
1
Vv* Vvd
1 Tr s
(53a)
* 1
V w Vwd
1 Tr s

with Vw* (0) Vv* (0) 0 , where the filter time constant Tr is freely chosen by the user and Vwd , Vvd are

positive speed setpoints. The above references cannot both be freely chosen because they are linked by
(32). As a matter of fact, the desired reference of vehicle speed Vvd is first chosen by the user. Then,

the wheel reference Vwd is let to be of the form:

Vwd (1 *
)Vvd (53b)
*
where is a constant (representing the sliding) that is uniquely obtained from (32) letting there
x2 Vvd , x1 (1 *
)Vvd and setting x2 0 . Doing so, one gets g 2 (1 *
)Vvd ,Vvd 0 , due to (32).
Then (33b) yields:
*
(1 )Vvd ,Vvd f 2 (1 *
)Vvd ,Vvd 1 (1 *
)Vvd ,Vvd f 2' (1 *
)Vvd ,Vvd 0 (53c)
*
This is an algebraic equation that must be solved to get the adequate value of for any fixed Vvd .
*
Recall that (1 )Vvd ,Vvd 1, 1 . Then, it is readily checked that the desired speeds
*
Vvd ,Vwd Vvd , (1 *
)Vvd will be in the validity domain Dv if belongs to the interval

h
l , .
1 h

*
Example 2. Table IV shows the values of the parameter , obtained by solving (53c), for different
values of Vvd in both acceleration and deceleration modes. It is readily observed that *
belongs to the

16
 TABLE IV. VALUE OF PARAMETER * (%) IN DIFFERENT OPERATING CONDITIONS
(ALL SPEEDS ARE IN km/h)

Vvd Va = 0, θ = 0° Va = 10, θ =0° Va= 0, θ = 5°

I 65 0.1300 0.0980 0.7280

II 60 -6.6500 -6.6520 -6.6525
I: Acceleration, II: Deceleration

h *
interval l , . In Example 1 we got l 0.07 and h 0.11. Hence, the above interval for
1 h

turns out to be 7%, 10% .

3.2. Speed Control Law Design

As the control objective is to enforce the vehicle speeds ( Vw , Vv ) to track their reference trajectories

( Vw* , Vv* ), let us introduce the following tracking errors:

def
z1 x1 Vw*
def (54)
*
z2 x2 V v

From (32) it readily follows that the errors undergo the following equations:
z1 *
1 ( x1 , x2 ) u g1 x1 , x2 Vw*
(55)
z2 g 2 x1 , x2 Vv*
Let us consider the following positive definite Lyapunov function candidate:
V ( z1 , z2 ) c1 z1 c2 z 2 (56)

where c1 , c2 are any positive design parameters. Derive V ( z1 (t ), z2 (t )) with respect to time yields:

V c1 sign( z1 ) z1 c2 sign( z2 ) z2

c1 sign( z1 ) *
1 ( x1 , x2 ) u g1 x1, x2 Vw* c2 sign( z2 ) g2 x1, x2 Vv*

sign( z1 ) c1 *
1 ( x1 , x2 ) u c1 g1 ( x1 , x2 ) Vw* c2 sign( z1 ) sign( z2 ) g 2 ( x1 , x2 ) Vv* (57)
where equations (44) have been used in the second equality. Equation (46) suggests the following
control law:
1
u *
c1 g1 ( x1 , x2 ) Vw* c V sign( z1 ) c2 sign( z1 ) sign( z2 ) g 2 ( x1 , x2 ) Vv* (58)
c1 1 ( x1 , x2 )
with c 0 is a new design parameter. Indeed, substituting the right side of (58) to u in (57) yields
V c V which in turn implies:

17
 V ( t ) V ( 0 )e c t (59)
Equation (48) holds for any t 0 provided that (58) has been applied over the interval [0, t ) . On the
other hand, substituting (58) in (32) yields the closed-loop system representation in the ( x1 , x2 )
coordinates:

c1 Vw* sign ( z1 )(c2 sign ( z2 ) g 2 ( x1 , x2 ) Vv* cV )

x1
c1 (60a)
x2 g 2 ( x1 , x2 )

Note that the derivatives Vw* and Vv* in (58) can be obtained using (53a). Specifically:

Vwd Vw*
Vw*
Tr
(60b)
Vv Vv*
d
Vv*
Tr

Theorem 1 (Main result). Consider the control system, illustrated by Fig. 10, consisting of the state-
space model (32) in closed-loop with the regulator defined by the control law (58). Let the regulator
parameters be chosen as follows:

* V (0) V (0)
1 l (1 ) 1 *
/ 1 (61a)
c1Vw c2Vv*

* V (0) V (0)
1 1 *
/ 1 1 h (61b)
c1Vw c2Vv*

for some 0 1 , arbitrarily chosen. Then, one has the following properties:

1) The domain Dv defined by (41) is an invariant set of the closed-loop system (60a-b).

2) Whatever the initial condition ( x1 (0), x2 (0)) Dv , one has:

a) all signals of the closed-loop system remain bounded.
b) the tracking errors ( z1 (t ), z2 (t )) vanish asymptotically.

Proof. Part 1. First, note that the control law (58) is well defined (i.e. involves no singularity) as long
as ( x1 (t ), x2 (t )) Dv because, by definition of Dv , no singularity of g1 x1 , x2 and g 2 x1 , x2 can
* ' *
occurs therein. Furthermore, it is readily seen from (33c-d) that 1 ( x1 , x2 ) 1 , 1 i.e. 1 ( x1 , x2 )

Regulator u
Vv* (58) Vw
Vehicle
(32)
Va Vv
Fig. 10. Theoretical control system analyzed in Theorem 1. It include the
vehicle longitudinal model (32) in closed-loop with the regulator (58)

18
never vanishes. Now, let us show that Dv is an invariant set of the closed-loop system (49a-b). To this

end, suppose that ( x1 (0), x2 (0)) Dv . As the right sides of (60a-b) are piecewise continuous functions

of ( x1 , x2 ) , there exist a 0 such that, for all t (0 ) , one has ( x1 (t ), x2 (t )) Dv which in view of
(41) means that:
x1
1 l 1 h , for all t [0 ) (62)
x2
x1 (t )
lim 1 l ,1 h (63)
t x 2 (t )
On the other hand, one gets from (59) that V (t ) V (0) , for all t [0 ) , which together with (56)
yields:
V (0) V (0)
Vw* x1 (t ) Vw*
c1 c1
, for all t [0 )
* V (0) * V (0)
Vv x2 (t ) V v
c2 c2

This in turn gives:

V (0) V ( 0)
Vw* Vw*
c1 x1 (t ) c1
V (0) x2 (t ) V ( 0)
Vv* Vv*
c2 c2
or, equivalently:
V ( 0) V ( 0)
*
1 1
V c1Vw*
w x1 (t ) V w
*
c1Vw*
* *
, for all t [0, ) (64)
V 1 V ( 0)
v x2 (t ) V 1 V ( 0)
v
c2Vv* c2Vv*

Using (53b), it follows from (64) that, for all t [0, ) :

V (0) V (0)
1 1
* c1Vw* x1 (t ) * c1Vw*
(1 ) (1 )
V ( 0) x2 (t ) V ( 0)
1 1
c2Vv* c2Vv*

x1 (t )
which, together with (61a-b), gives 1 l 1 h , for all t [0 ) . But, this clearly
x 2 (t )

contradicts (63) because 0 1 . So, Dv is actually an invariant set of the system (60a-b).

Part 2. From Part 1, it follows that if ( x1 (0), x2 (0)) Dv then ( x1 (t ), x2 (t )) Dv , for all t 0 . Then,
equation (58)-(59) hold for all t 0 . Equation (59) implies that V (t ) is bounded and, consequently, so
are x1 (t ) and x2 (t ) , due to (54) and (56). Furthermore, since the right side of (58) is a piecewise

19
function of ( x1 , x2 ) and involves no singularity (because ( x1 (t ), x2 (t )) Dv ), it follows that the control

signal u (t ) is also bounded proving Part 2a. Part 2b is a direct consequence of (59). This completes the
proof of Theorem 3.1

Remarks 1. 1) Theorem 3.1 ensures that the closed-loop system is asymptotically stable (in the
( z1 , z2 ) -coordinates) with an attraction region containing the whole validity domain Dv .
2) A crucial step in the regulator development was to find out a suitable Lyapunov function. In
particular, quadratic like functions turned out to be helpless, due to the particular structure of the
controlled system.
* h
3) Inequalities in (61a-b) can easily be fulfilled letting l and c1 , c2 be sufficiently
1 h

large.
4) Theorem 1 shows that the regulator (58) performs well when applied to the control design model
(32). The question is whether such a good performance is preserved when the regulator is applied to
the more accurate model (52), see (Figs. 12a). This question is investigated by simulation in Section
IV.

3.3. Control Law Design Neglecting Tire Dynamics

Presently, the simplified control design model (38) is based upon. The control objective is to enforce
the vehicle speed Vv to follow the reference trajectory Vv* . Introduce the control error:
def
z x2 Vv* (65)
From (41) it is follows that this error undergoes the following equation:
z u f ( x2 ) Vv* (66)
Consider the Lyapunov function candidate:

z2
V ( z) (67)
2
Using (66)-(67), one gets the time-derivative of V (z ) with respect to time:

V z u f ( x2 ) Vv* (68)
This suggests the following control law:
1 c
u f ( x2 ) Vv* z (69)
2

with c 0 is a design parameter. Indeed, substituting (67) in (68) gives V cV which shows that
V is exponentially vanishing. Hence, the control objective (i.e. z 0 ) is ensured if the control law

20
(69) is applied to the simplified model (38). The question is whether such a good performance is
preserved when the control law (69) is applied to the more realistic model (52) (Fig. 12b). This
question is investigated in Section IV.

3.4 Further Discussion on Practical Implementation

First, notice that the control action generated by a vehicle longitudinal controller (like (58)) constitutes
the torque reference signal for the engine torque regulator (not discussed in this study). Accordingly,
the longitudinal speed regulator and the engine torque regulator constitute together a more global
cascade controller (Fig. 11). In cascade control jargon, the torque regulator is referred to „inner‟ or
„slave‟ while the speed regulator is called „outer‟ or „master‟. For the global cascade controller to
perform well the torque regulator must be designed so that the inner loop is much more rapid than the
outer loop. The previous remarks are independent of the engine technological nature (thermal, electric
or hybrid).
Recall also that the analytical design of the (outer) longitudinal speed regulator (58) (developed in
Section III) relies on the longitudinal vehicle model (32)-(33). The numerical values of the involved
parameters depend on the particular vehicle under study (for instance, Table VI shows those of a
Citroën-2CV). They should normally be provided by the vehicle manufacturer; otherwise they must be
determined using model identification methods.
The controller design parameters (c, c1 , c2 , Tr ) must be given suitable numerical values before online
running of the control algorithm. As shown by simulation (Section IV), suitable values can simply be
selected following the usual „try-an-error‟ search method.
The practical application of the longitudinal speed regulator necessitates the numerical implementation
of the control law (58) and speed measurements ( x1 Vw and x2 Vv ). As the controlled system have
relatively slow dynamics (due to its mechanical nature) a sampling frequency of 100 Hz would be
convenient for data acquisition. Following the usual practice, the conditioning of data acquisition is
made better if the signals are properly processed before sampling. Online pre-processing operations
include signal filtering, amplification, modulation, demodulation etc. Given the simplicity of involved
online algebraic operations (in control law and signal processing), the small number of online
input/output measurements and the relatively low data-acquisition frequency, the required computation
resources are relatively modest. A low-cost DSP and any one of today‟s microprocessors or
programmable logic controllers (PLC) would be sufficient for real-time implementation of the control
algorithm. The major numerical implementation tasks are described by Table V.

21
 Engine Engine & Torque Driveshaft
torque Gear Box & Tires
Torque
regulator
reference u

Chassis & wheel

speed
measurements
Longitudinal speed
regulator (58)

Fig. 11. Global cascade controller for vehicle longitudinal speed control. The grey part refers to the
(outer) speed control loop dealt with in the present study.

TABLE V. CONTROLLER IMPLEMENTATION ALGORITHM

Step #0# Choice of design parameter and sampling time
Step #1# Acquisition of parameters and states variables x1 (k ) and x 2 ( k )
Step #2# Generate the reference trajectories Vv* (k ) and Vw* (k )
Step #3# Compute the control law u (k )
Step #4# Apply the control value to the engine
Step #5# Set k k 1 and Go to step #1#

4. SIMULATION

The performances of the sophisticated controller (58), obtained from model (32), will be compared
using numerical simulations, to those of the simpler controller (69), obtained from the simpler and less
accurate model (38). Both controllers are applied to the most accurate (simulation-oriented) vehicle
longitudinal model (52), that not only account for tire dynamics but also for the validity domain (41)
so that the model operation is singularity-free. That is, the simulation study will be performed, with
Matlab-Simulink, according to the experimental setups illustrated by Figs 12a and 12b, respectively.
The characteristics of the vehicle are those of a Citroën-2CV of Table VI.

Regulator u Regulator u
Vv* (58) Vv Vv* (69) Vv
Vehicle Vehicle
model model Vw
(52) (52)
Va Vw Va

Fig. 12a. Vehicle longitudinal control involving the Fig. 12a. Vehicle longitudinal control involving the
simulation-oriented model (52) in closed-loop with the simulation-oriented model (52) in closed-loop with the
sophisticated speed regulator (58) simpler speed regulator (69)

22
 TABLE VI. NUMERICAL CHARACTERISTICS OF THE CITROËN-2CV CAR (FROM CHAIBET, 2006)
AD : Asphalt Dry, CW : Cobblestone Wet AD CW
C1 : Primary tire parameter 1.2801 0.5
C2 : Primary tire parameter 23.99 30
C3 : Primary tire parameter 0.52 0.2
Mv : Chassis mass 560 kg
J : Wheel inertia 1000 kg m2
reff : Effective wheel radius 0.28 m
rr : Rolling resistance coefficient 0.025
: Related height of the center of gravity 0.2
: Related position of center of gravity 0.43
Kv : Load correction factor 0.55
: Density of air 1.202 kg/m3
Cx : Aerodynamic drag coefficient 0.5
Cz : Aerodynamic lift coefficient 0.259
S : Frontal area vehicle 0.8 m2

As mentioned previously, there is no simple way to find the best choice for the design parameters
i.e. c, c1 , c2 , Tr . Following the usual practice, suitable numerical values are selected using the heuristic
„try-an-error‟ search method. Accordingly, the next values have been obtained:
. the model reference time constant in (53a) is set to Tr 1s ,
. the design parameters c, c1 , c2 for the controller (58) are given the values c 0.1 , c1 60 , c2 2,
. the best choice of the design parameter c in the controller (69) turned out to be c 2.

4.1. Control Performances in Easy Driving Conditions

Presently, the control performances of both controllers are illustrated in ideal driving conditions
characterized by dry and flat road and weak front wind. Specifically, the driving conditions are defined
by the model parameters 0 , Va 10km / h , (C1 , C2 , C3 ) (1.2801, 23.89, 0.52) .
Furthermore, it is supposed that all above values are perfectly known to the designer and used in the
controller design. Figs. 13a-e show the control system responses obtained with both controllers in
these easy driving conditions. For both, the speed responses, Vv and Vw , converge to their respective
references (with settling time less than 1.75s) and present no overshoot in the acceleration mode (Figs.
13a-b). But, as expected, the performances of the sophisticated controller (58) are clearly better than

23
those of (69). Indeed, the former is much more speedier, develops a smaller control effort (Fig. 13c-d)
and ensures a weaker sliding (Fig. 13e).

4.2. Control Performances in Hard Driving Conditions

Presently, the driving conditions are much harder compared to the previous subsection. That is, the
road is given the profile of Fig. 14 showing an ascendant stage followed immediately by a descent
stage. Analytically, the varying road slop is defined as follows:

0 for 0 t t1
1 t t1
(t ) max (1 cos( 2 )) for t1 t t2 (70)
2 t 2 t1
0 for t t 2

with max 10 , t1 8s , t 2 12s . Furthermore, the road is Asphalt Dry, characterized by:

(C1 , C2 , C3 ) (1.2801, 23.99, 0.52)

Fig. 15a-b show the vehicle and wheel velocities obtained with both controllers. It is seen that the
deviation (between wheel and chassis velocities) gets larger when the vehicle passes over the road
bump, between t1 8s and t2 12s . However, the deviation is much larger with the simpler controller
(69) than with the sophisticated one (58). This fact is better illustrated by Fig. 16 that show the
behavior of sliding is better with controller (58).
Fig. 17 shows the wheel torque developed by both controllers. Clearly, the regulator (69) develops a
huge torque (on the wheel) during the ascendant stage. In the descendant stage, the developed torque
decays drastically taking small values. As a matter of fact, such behavior is practically unacceptable
because it is harmful for (person and vehicle) safety and is costly from an energetic viewpoint. Clearly,
the regulator (58) is better. The supremacy of (58) over (69) is now quantified considering the average
torque and frictional work done by the front tires. These are respectively defined as follows:
1
TA u dt (71)
0

1
DA Ftf Vw dt (72)
0

where u is the torque applied to the wheels, Ftf is the component of the frictional force projected onto

the tangential direction of the contact surface, Vw is the speed of contact tire-road and is the entire
simulation time interval. Fig. 18a shows the variation of the average torque and frictional work with
the maximum slop max . The comparison between the two controllers is dealt with considering two
road states i.e.:
- Asphalt Dry (C1 , C2 , C3 ) (1.2801, 23.99, 0.52)

24
 - Cobblestone Wet (C1 , C2 , C3 ) (0.5, 30, 0.2) .
The figure indicates that, in all driving conditions, the average torque of the vehicle being driven by
the regulator (58) is much smaller than with the regulator (69). This confirms that energy consumption
is lower with (58) and safety is better. From Fig.18b it is seen that tire friction activity is much weaker
when the vehicle is driven with the regulator (58), especially when the vehicle is going through a bump
by cobblestone wet road condition. Then, a high tire friction activity is involved with the controller
(69) which means that the tires are then subject high pressure.

Fig 13a. Closed-loop chassis speed responses. Solid: reference speed Vv* . Dashed: vehicle speed response Vv obtained

with regulator (58). Dotted: vehicle speed response V v obtained with regulator (69)

Fig 13b. Wheel speed responses. Solid: reference V wd . Dashed: speed response Vw with regulator (58). Dotted: speed

response Vw with regulator (69)

25
Fig 13c. Control signal u . Solid: control signal with regulator (58). Dashed: control signal with regulator (69)

Fig 13d. Traction force Ftf . Solid: Ftf with regulator (58). Dashed: Ftf with regulator (69)

Fig 13e. Sliding ; solid: Sliding (58); dashed: Sliding (69)

26
 max

t1 t2

Fig. 14. Road profile considered in Subsection 4.2.1

Fig. 15a. Vehicle speed closed-loop response. Solid: true vehicle speed response Vv obtained with regulator (58). Dotted:

vehicle speed response V v obtained with regulator (69)

Fig. 15b. Wheel speed closed-loop responses. Solid: true wheel speed Vv obtained with regulator (58). Dotted: wheel speed

V v obtained with regulator (69)

Fig. 16. Sliding responses. Solid: sliding obtained with (58). Dashed: sliding obtained with regulator (69)

27
 Fig. 17. Control signal u . Solid: control signal with regulator (58). Dashed: control signal with regulator (69)

Regulator (69)
Cobblestone Wet
Regulator (69)
Asphalt Dry
Regulator (58)
Cobblestone Wet
Regulator (58)
Asphalt Dry

Fig. 18a. Average torque when going through the bump

Regulator (69)
Cobblestone Wet

Regulator (69)
Asphalt Dry

Regulator (58)
Cobblestone Wet
Regulator (58)
Asphalt Dry

Fig. 18b. Frictional effort done by the tire

5. CONCLUSION

The problem of vehicle longitudinal control is addressed, both in acceleration and deceleration modes,
based on the two new models defined by (32) and (52). The originality of these models lies in the fact
that they explicitly accounts for the longitudinal slip (resulting from tire deformation) using Kiencke‟s

28
model. Furthermore, the model (52) also accounts for real-life considerations that physically limit the
longitudinal slip value. Consequently, (52) turns out to be quite suitable for simulating the vehicle
longitudinal behavior. The control design is based on the slightly simpler model (32) which ignores the
slip limitation but still accounts for the tire longitudinal slip. It is shown that a speed controller can
actually be obtained from that model using a Lyapunov type design technique. It is formally shown
that the obtained controller, defined by (58), does meet its performances i.e. stability and perfect speed
reference. It is also checked by simulation that the controller (58) is much better than the simpler one
defined by (69) and ignoring longitudinal slip. The supremacy of (58) is especially appreciated in hard
driving conditions i.e. crossing rampant and wet roads. In all simulated experiments, the vehicle
longitudinal motion is represented by the most accurate and highly nonlinear model (52).

REFERENCES

Chaibet A., (2006). “Contrôle Latéral et Longitudinal pour le Suivi de Véhicule”, PhD, Université
d‟Evry Val d‟Essonne, France.

Dugoff P.F.H. and L. Segel, (1970). “An analysis of tire traction properties and their influence on
vehicle dynamic performance”, SAE Technical Paper 700377, vol 3, pp. 1219-1243,
doi:10.4271/700377.

Gim G. and P. Nikravesh, (1990). “An analytical model of pneumatic tyres for vehicle dynamic
simulations part1: Pure slips”, International Journal of Vehicle Design, Vol. 11, No. 6, pp. 589-618.

Guo K. and L. Ren, (2000). “A non-steady and non-linear tire model under large lateral slip condition”,
SAE Technical Paper 2000-01-0358, 2000, doi:10.4271/2000-01-0358.

Ioannou P.A. and C.C. Chien, (1993). “Autonomous Intelligent Cruise Control”. IEEE Transactions on
Vehicular Technology, Vol. 42, No. 4, pp. 657-672.

Khalil H.K., (2002). “Nonlinear Systems”, Prentice-Hall, 3rd edition.

Kiencke U. and L. Nielsen, (2005). “Automotive control system for engine, driveline, and vehicle”,
2nd edition, Springer.

Liang H., K.T. Chong and T.S. No, S.Y. Yi, (2003). “Vehicle longitudinal brake control using variable
parameter sliding control”. Control Engineering Practice, Vol. 11, No. 4, pp. 403-411

Milliken F.W. and L.D. Milliken, “Race car vehicle dynamics”, SAE International, 1995.

Moon S., I. Moon and K. Yi, (2009). “Design, tuning, and evaluation of a full-range adaptive cruise
control system with collision avoidance”. Control Engineering Practice, Vol. 17, No. 4, pp. 442-455.

Nouvelière L. and S. Mammar, (2007). “Experimental vehicle longitudinal control using a second

29
 order sliding mode”. Control Engineering Practice, Vol. 15, No. 8, pp. 943-954.

Pacejka H.B. and I.J.M. Besselink, (1997). “Magic Formula Tyre Model with Transient Properties”,
2nd International Tyre Colloquium on Tyre Models for Vehicle Dynamic Analysis, Berlin,
Germany.

Poursamad A. and M. Montazeri, (2008). “Design of genetic-fuzzy control strategy for parallel hybrid
electric vehicles”. Control Engineering Practice, Vol. 16, No. 7, pp. 861-873

Powers W.F. and P.R. Nicastri, (2000). “Automotive vehicle control challenges in the 21st century”.
Control Engineering Practice, Vol. 8, pp. 605-618.

Ren T.J., T.C. Chen and C.J. Chen, (2008). “Motion control for a two-wheeled vehicle using a self-
tuning PID controller”. Control Engineering Practice, Vol. 16, No. 3, pp. 365-375

Yamakawa J., A. Kojima and K. Watanabe (2007). “A method of torque control for independent wheel
drive vehicles on rough terrain”. Journal of Terramechanics, Vol. 44, pp. 371–381.

You S.H., J.O. Hahn and H. Lee, (2009). “New adaptive approaches to real-time estimation of vehicle
sideslip angle”. Control Engineering Practice, Vol. 17, No. 12, pp. 1367-1379.

30