You are on page 1of 6

A preliminary study to integrate LTV-MPC Lateral Vehicle Dynamics

Control with a Slip Control


Giovanni Palmieri, Osvaldo Barbarisi, Stefano Scala and Luigi Glielmo
AbstractIn this paper we present the integration of
a Linear-Time-Varying Model-Predictive-Control (LTV-MPC),
designed to stabilize a vehicle during sudden lane change
or excessive speed-entry in curve, with a slip controller that
converts the desired longitudinal tire force variation in pressure
variation in the brake system. The lateral controller is designed
using a three-degrees-of-freedom vehicle model taking into
account both yaw rate and side slip angle of vehicle while the
slip controller is a nonlinear gain scheduling P with feedforward
action. The performances are validated with SIL technique, in
particular, the authors use a proprietary simulator calibrated
on an oversteering sport commercial car. Simulation results
show the benets of the control methodology used.
I. INTRODUCTION
Vehicles are equipped with several electronic controller,
each of them begin designed to control a particular part of the
vehicle. These systems contribute to improve performances,
comfort and car safety in different working conditions. To
properly integrate all subsystems actions and and optimize
their cooperation a deep knowledge of vehicle dynamics is
needed. Obviously, proper modeling is essential to achieve
the above goal. In this work, we illustrate a model-based
lateral stability control with an integrated slip controller. The
goal of lateral controllers is to stabilize the vehicle during
sudden lane changes or braking. In these situations may be
unable to track the nominal trajectory and may follow a
larger radius trajectory with high values of the yaw rate speed
and the side slip angle [1]. Then, the goals of control system
are (i) the yaw velocity of the vehicle to track as much as
possible the nominal motion expected by the driver and (ii)
the reduction of side slip angle.
In order to achieve both goals, it is possible to manipulate
three groups of variables: the front (and rear) steering angle,
the braking force (through the so called differential braking)
and the engine torque, usually employed together with the
former two. In this paper, we underline the attention of
differential braking strategy. This approach is illustrated with
a hierarchical control scheme in [2], [3], where several
uncoupled SISO controllers with a supervisor strategy are
proposed for guaranteeing stable and robust behavior. In
[4] the lateral controller is a model-based MIMO system,
the model is the full vehicle model in body frame, the
control inputs are the variation of longitudinal forces on
each tires and the designed control strategy using LTV-
MPC techniques. The main contribution of this work is the
G. Palmieri, O. Barbarisi and L. Glielmo are with the Dipartimento
di Ingegneria - Universit` a del Sannio - P.za Roma, 21 - 82100 Benevento,
Italy. {barbarisi,glielmo,palmieri}@unisannio.it
Stefano Scala is with ELASIS S.C.p.A, Via Ex Aeroporto - 80038
Pomigliano DArco (NA), Italy stefano.scala@elasis.it
Fig. 1. General control scheme.
integration of the control strategy presented in [5] with a slip
controller, which acts on the pressure of the brake cylinder
in the brake system, as showed in Figure 1.
II. FULL VEHICLE MODEL
The double tracks model of the vehicle is depicted in
Figure 2; the dynamics of the vehicle are described in details
in [6] and reported in Equations (1)(4). The state variables
are the side slip angle of vehicle , the yaw rate and the
rotational speed
i j
of each wheel. (In the following we will
often omit the indices i, front/rear side, j, left/right to indicate
specic wheel quantities when this does not cause confusion
and simplies notation.) There are two different input types:
(i) the wheel turn angle is a disturbance input, i. e. it is
not a controlled variable, since the drive imposes it; (ii)the
contribution of engine torque on each wheel T
eng
and the
braking torque T
B
on each wheel, instead, are given by the
controller.
Fig. 2. Double track vehicle model.
1) Lateral and longitudinal forces: The interaction of
forces between the wheel and the ground are described by
the semi-empirical Pacejka model which we simplify by
Joint 48th IEEE Conference on Decision and Control and
28th Chinese Control Conference
Shanghai, P.R. China, December 16-18, 2009
ThB12.6
978-1-4244-3872-3/09/$25.00 2009 IEEE 4625
_
_



i j
_
_
=
_
_
f
1
_
v, , , , F
L
i j
, F
S
i j
_
f
2
_
v, , , , F
L
i j
, F
S
i j
_
f
3
_
F
L
i j
, T
eng
i j
, T
B
i j
_
_
_
, i ={F, R} j ={L, R} (1)
f
1
_
v, , , , F
L
i j
_
=
1
Mv
{(F
L
FL
+F
L
FR
)sin( +) (F
S
FL
+F
S
FR
)cos( )
+ (F
L
RL
+F
L
RR
)sin (F
S
FR
+F
L
FL
)cos} , (2)
f
2
_
v, , , , F
L
i j
_
=
1
J
z
{l
a
(F
S
FL
F
S
FR
)cos l
a
(F
L
FL
F
L
FR
)sin +l
b
(F
S
RL
+F
S
RR
)
+ l
c
(F
S
FL
F
S
FR
)cos +l
c
(F
L
FL
F
L
FR
)sin +l
c
(F
L
RR
F
L
RL
)}, (3)
f
3
_
F
L
i j
, T
eng
i j
, T
B
i j
_
=
1
J
w
_
rF
L
i j
T
B
i j
+T
eng
i j
_
. (4)
assuming that, for each wheel, the longitudinal force (resp.,
the lateral force) depends only on the longitudinal slip (resp.,
only on the side slip angle), i.e. [7]:
f
L
(s; , F
z
) =

f
L
(s, , , F
z
)

=0
, (5a)
f
S
(; , F
z
) =

f
S
(s, , , F
z
)

s=0
, (5b)
where

f
L
and

f
S
are the full (i.e., combined) Pacejkas
formulas for longitudinal and lateral force, is the wheel
side slip angle, i.e. the angle between the longitudinal wheel
axis and the velocity vector v
w
. In the following we will
often omit the dependence on and F
z
for the sake of
simplicity. The functions f
L
(pure driving/braking) and f
S
(pure cornering) are odd w.r.t. of s and respectively.
Finally, the tire stiffness coefcients are dened as [6]
c(; , F
z
) :=
f
S
(; , F
z
)

. (6)
III. LATERAL VEHICLE DYNAMIC CONTROL STRATEGY
The algorithm we propose here assumes the following
variables to be directly measured: , ,
i j
, a
y
and a
x
. The
variables and v, instead, have to be estimated on the basis
of the above mentioned measurements (details in [6], [1],
[8]). The control scheme is depicted in Figure 1. Given the
vehicle velocity v, a proper safety bound to guarantee tires
adherence for the side slip angle is computed; on the basis
of v and the steering wheel angle is computed a reference
yaw rate signal
ref
[8]. When either the difference between
the actual yaw rate and the reference yaw rate exceeds a
safety range or the vehicle side slip angle exceeds the bound,
the Supervisor block enables the Control block to compute
the braking torques to be applied independently to each
wheel.
A. Estimation of tire variables
1) Longitudinal slip ratio: The longitudinal slip ratio is
dened as in [7]
s
i j
:=
r
i j
v
w
i j
max{r
i j
; v
w
x
i j
}
, (7)
where r is the actual rolling radius of the tire, is the angular
speed of the tire (which is measured) and v
w
x
is the linear
speed of the tire center along the longitudinal axis of the
wheel (which is estimated, e.g. v
w
x
RR
= v
x
+l
c
).
2) Tire side slip angle: The tire side slip angle are
shown in Figure 2 and they result:

FL
= +atan
_
vsin

+l
a

vcos

l
c

_
+

+
l
a

v
, (8a)

FR
= +atan
_
vsin

+l
a

vcos

+l
c

_
+

+
l
a

v
, (8b)

RL
=atan
_
vsin

l
b

vcos

l
c

_



l
b

v
, (8c)

RR
=atan
_
vsin

l
b

vcos

+l
c

_



l
b

v
, (8d)
where we suppose cos

1, sin(

)

and v |l
c
|.
3) Estimation of tire forces: Once the slips s and and
the vertical tire forces are estimated as in [6], we use the
simplied Paceijkas formulas (5) to compute the tire force
F
L
and F
S
. We assume knowledge of .
4) Bounds on the control inputs: If we represent the
longitudinal and lateral forces that the road can exert on the
tire as in Figure 3, they belong to an area that depends on the
road conditions and the vertical force F
z
. We assume that
the area can be approximated by the ellipsoid, as shown in
Figure 3 (this can be justied on the basis of the combined
Pacejkas formulas)
F
2
L
+F
2
S
= r
2
, (9)
where r :=F
LIM
S
and :=
_
F
LIM
S
_
F
LIM
L
_
2
and F
LIM
L
and F
LIM
S
represent the maximum longitudinal force and lateral force
respectively, dened as
F
LIM
L
(, F
z
) :=max
s
f
L
(s; , F
z
), (10a)
F
LIM
S
(, F
z
) :=max

f
S
(; , F
z
). (10b)
Given the actual side force F
S
, the maximum achievable
longitudinal force that avoids the unstable slipping of the
wheel is given by
F
max
L
:=
_
1

_
r
2
(F
S
)
2
_
. (11)
ThB12.6
4626
Fig. 3. Lateral and longitudinal forces of the tire inscribed in the ellipsoid
of adherence. We draw negative longitudinal forces because we deal with
braking.
Fig. 4. Statechart of Supervisor. The default state is NoVDC where the
boolean enable signal is set to low and the controller is disabled. When one
of the two errors signals is outside the bounds, state YesVDC is activated.
The state YesVDC has two child-states: Normal and Wait. The rst is
the default state, while Wait is activated when both errors are inside the
thresholds; if this condition holds for a time t t
0
> T
del
, where t
0
is the
time when the controller enters in Wait state, then the NoVDC is activated.
Finally we point out that, since we are considering the
braking action of the tires, rather than the acceleration action,
the longitudinal forces F
L
will be negative (see Figure 3) and
the maximum braking forces corresponding to a side force
F
S
will be F
max
L
.
B. Supervisor
The lateral dynamics controller is activated when either
the error on yaw rate e

or the error on side slip angle
e

exceed respective activations thresholds. The controller is


deactivated when both e

and e

are within those thresholds


for a period T
del
. Figure 4 illustrates the statechart of the
corresponding automa.
The thresholds on yaw rate error depend on the vehicle
speed. Since the vehicle responds to the steering wheel angle
in different manners as the vehicle speed changes, we choose
to shape the yaw rate error activation threshold e
on

and the
deactivation threshold e
off

as the following
e
on

(v
x
) =
2v
x
/v
ch
_
1+v
2
x
/v
2
ch
_e
ON

, (12a)
e
off

(v
x
) =e
on

(v
x
), (12b)
where e
ON

> 0 is a calibration parameter and is a calibra-
tion parameter with (0, 1), typically = 0.75.
The side slip angle error activation threshold e
on

has been
chosen constant, since in any condition the side slip angle
should be small; the deactivation threshold e
off

= e
on

.
The delay T
del
on the deactivation was introduced after
observing that, without it, when the steering wheel angle
changes direction abruptly, the controller is disabled (for a
very short time) since also the errors on yaw rate and side
slip angle change their signs.
C. Model Predictive Control
The model presented in section II is nonlinear. To reduce
the computational load rather than working on the model (1),
we rst compute the linearization of the submodel described
by the rst two Equations of (1) around the current sub-state
(, ). We remind that f
S
( +)

= f
S
() +c().
Further, in our control model we consider as input the varia-
tion F
L
of the tire braking forces (i.e., negative) constrained
to - F
max
L
F
L
+F
L
0 so that
F
min
L
F
L
F
max
L
, (13a)


F
min
L


F
L


F
max
L
, (13b)
where F
min
L
:=F
max
L
F
L
and F
max
L
:=F
L
. Inequalities
(13a) and (13b) represent physical bounds: (13a) describes
the tire forces domain in a particular working point and (13b)
represents the slew-rate of the braking system.
The control goal is to maintain the errors e

and e

close to zero. The linearized model around the working point
( x, u, v,

) = (

, , F
L
, v, ) is
e = A
_
v,

, , F
L
_
e +B
_
v,

,
_
_

_
F
L
FL
F
L
FR
F
L
RL
F
L
RR
_

_
(14)
+d
_
v,

, , , F
L
i j
, F
S
i j
_
,
where
A
_
v,

, , F
L
_
=
f (x, u)
x

x,u
,
B
_
v,

, , F
L
_
=
f (x, u)
u

x,u
,
and
d
_
v,

, , , F
L
i j
, F
S
i j
_
=
_
_
f
1
_
v,

, , , F
L
i j
, F
S
i j
_
f
2
_
v,

, , , F
L
i j
, F
S
i j
_
_
_
.
The model and its constraints are then discretized by
using backward Eulers method with sampling time T
s
, thus
obtaining
x
k+1
= A
T
(k)x
k
+B
T
(k)u
k
+d
T
(k) (15)
where u
k
=
_
F
L
FL
F
L
FR
F
L
RL
F
L
RR

|
t=kT
s
is the
control input, A
T
(k) =I +A
_
v,

, , F
L
_
|
t=kT
s
T
s
is the system
ThB12.6
4627
matrix, B
T
(k) = B
_
v,

,
_
|
t=kT
s
T
s
is the input matrix and
d
T
(k) = d
_
v,

, , , F
L
i j
, F
S
i j
_
|
t=kT
s
T
s
.
We now consider a prediction horizon H
p
T
s
, for some
integer H
p
, in which the following approximations hold
A
T
= A
T
(k)

= A
T
(k +1)

= ...

= A
T
(k +H
p
1), (16a)
B
T
= B
T
(k)

= B
T
(k +1)

= ...

= B
T
(k +H
p
1), (16b)
d
T
= d
T
(k)

= d
T
(k +1)

= ...

= d
T
(k +H
p
1). (16c)
We also consider a control horizon H
c
T
s
, for some integer
H
c
H
p
, and dene U
k
=
_
u

k
... u

k+H
c
1

.
At each time k we solve the following optimization
problem:
U

k
= argmin
U
k
H
p
1

h=0
_
x
T
k+h+1
Qx
k+h+1
+u
T
k+h
Ru
k+h
_
(17a)
subject to
x
k+h+1
= A
T
x
k+h
+B
T
u
k+h
+d
T
, (17b)
x
k
=
_
e

(kT
s
)
e

(kT
s
)
_
, (17c)
u
k+h+1
= u
k+h
h H
c
1, (17d)
u
min
u
k+h
u
max
, (17e)
u
min
u
k+h
u
k+h1
u
max
, (17f)
where u
1
= 0. Inequality (17e) is related to (13a) and
inequality (17f) is related to (13b).
Since the cost function (17a) is quadratic and the con-
straints (17b)(17f) are linear, the optimization problem (17)
is convex and can be solved with an efcient quadratic
programming (QP) solver (see [8] for details). We denote
by U

k
= [u

k
, . . . , u

k+H
c
1
]

the sequence of optimal braking


torques computed at time k by solving problem (17) from
the current observed state x
k
. Then the rst element u

k
of
U

k
is actually applied to the system at time k.
Even though the MPC algorithm is developed and im-
plemented in discrete time, it is notationally convenient to
employ a continuous time description in the remaining part
of the work. To do so we denote:
u(t)
MPC
= [hold(t; F
k
, T
s
))/T
s
(18)
where, with some abuse of notation, hold(t; u
k
, T
s
)) = u
k
for
t [kT
s
, (k +1)T
s
).
IV. A REDUCED MODEL FOR SLIP CONTROL
The desired variation of longitudinal forces is achieved by
the braking system. The slip controller computes the desired
longitudinal slip corresponding to the desired longitudinal
force and the braking force that has to be applied. For
control design we will consider the third equation of (4)
and, thanks to the fast time response of the wheel, we can
neglect the other dynamics by considering as constant the
yaw rate, vehicle side sleep angle and the vehicle velocity
during the time interval of slip control intervention. With this
assumption the model of angular wheel velocity becomes
J
w
=r f
L
(s) +T
eng
T
B
, (19)
and the derivative of the slip ratio can be written as
s =
r v
max{v
2
, (r)
2
}
. (20)
We distinguish two cases:
Braking When v r, then s [1, 0] and Equation (20)
yields
s =
r
v
, (21)
so that Equation (19) becomes
s =
_
r
vJ
w
_
_
r f
L
(s) +T
eng
T
B
_
. (22)
Traction When r v then s [0, 1] and Equation (20)
can be written as
s =
r v
r
2

2
=
r
v
(1s)
2
(23)
so that (19) becomes
s =
_
r
vJ
w
_
_
r f
L
(s) +T
eng
T
B
_
(1s)
2
. (24)
Notice that the roots of
r f
L
(s) = T
eng
T
B
(25)
are the equilibrium points s of both (22) and (24) with s
[1, 0] and s [0, 1], respectively. Such equilibrium points
may or may not exist according to value of the RHS of
Equation 25, and there may be one or two equilibrium points.
If f
L
( s)/s is positive, s is asymptotically stable so that
we dene the stable interval as the interval of s where the
above derivative is positive.
One of the goals of vehicle dynamic control is to keep the
longitudinal slip s inside the stable interval where typically
the longitudinal slip is close to zero and (1s)
2
= 1.
Hence, in order to design the longitudinal slip controller,
we consider only the braking phase with accelerator pedal
released and a high gear, we neglect T
eng
and use the model
s =
_
r
vJ
w
_
(r f
L
(s) T
B
). (26)
Fig. 5. Slip control scheme. For the sake of simplicity we included into
the wheel block the algorithm related to the estimation of slip ratio s
ThB12.6
4628
V. A SLIP CONTROL STRATEGY
The goal of the slip controller is to obtain the desired
longitudinal force
F
ref
L
(t) := F
ref
L
(0) +
_
t
0
u
MPC
()d (27)
by applying a braking torque T
B
. The desired slip s
ref
is
obtained through inversion of Pacejkas formula, computed
for values of slip inside the stable interval:
s
ref
= f
1
L
(F
ref
L
). (28)
The control action on the wheel is given by

T
B
=
f f
+
f b
=

T
B
f f
+

T
B
f b
. (29)
The slip control action, then, is the sum of a feedforward
action
f f
and a feedback action
f b
. The complete control
scheme is depicted in Figure 5. The feedforward control law
is given by

T
B
f f
=
f f
=ru
MPC
,
T
B
f f
=rF
ref
L
+const =rF
ref
L
, (30)
where u
MPC
is the control input obtained from (18). In the
Equation 30 we ignore the const value because the wheel
slip and the longitudinal forces are continuously estimated
so that, using (28), we can write
r f
L
(s
ref
) +T
B
f f
= 0. (31)
A. Feedback action
Let us dene the slip error as e
s
=s
ref
s. Now, taking into
account (27), the dynamic equation of error can be written
as
e
s
= s
ref
s =
_
1/

s
f
L
_
s
ref
_
_
u
MPC
(32)

_
r
vJ
w
_
(r f
L
(s) T
B
).
The rst term of RHS of Equation 32 (of order of 10
5
in
our simulations), can be neglected w.r.t. to the second one
(of order 10
2
). Thus, Equation 32 becomes
e
s
=(a
1
a
2
)e
s
+a
1
T
B
f b
, (33)
where the term r f
L
_
s
ref
_
+T
B
f f
is zero in view of (31) and
a
1
=
r
vJ
w
, a
2
= r

s
f
L
(s
ref
),

T
B
f b
=
f b
, (34)
with a
2
> 0 if s
ref
is in the stable interval. The transfer
function V(p; s
ref
), between feedback control input
f b
and
slip ratio error e
s
, is given by
V(p; s
ref
) =
a
1
p(p+a
1
a
2
)
. (35)
We use a proportional regulator whose gain depends on a
1
and a
2

f b
=k
p
e
s
. (36)
The transfer function of the closed loop system V
o
(s) is given
by
V
o
(p) =
a
1
k
p
p
2
+a
1
a
2
p+a
1
k
p
. (37)
The bandwidth of (37) has to be kept at least one order of
magnitude less than the sampling frequency of the regulator
f
s
(here f
s
= 1000[Hz]). So we choose the parameters of
regulator as
k
p
=
1
a
1
_
2
10
f
s
_
2
. (38)
Equation 37 becomes
V
o
(p) =
_
2
10
f
s
_
2
p
2
+a
1
a
2
p+
_
2
10
f
s
_
2
. (39)
Simulations show that the above closed loop transfer function
depends mildly on the operating conditions: vehicle speed v,
vertical force F
z
and friction coefcient .
VI. SIMULATION RESULTS
To test our control strategy we have chosen a manoeuver
known as the ATI90 in dry asphalt condition ( = 1). In
this manoeuver the driver: (i) turns the steering wheel from
90deg to +90deg, (ii) decides the initial speed, (iii)releases
the accelerator pedal when the manoeuvre starts.
We tested our strategy on an oversteering sport car
simulated through an ELASIS-CRF (FIAT group) proprietary
simulator. The tuning parameters are: sampling time T
s
= 20
[ms]; the prediction horizon H
p
= 5; the control horizon
H
c
= 3; the control weight for the side slip angle :
q
1
=10; the control weight for yaw rate : q
2
=1; the control
weight matrix R for the variation of longitudinal forces:
diag
_
10
11
|maxF
z
ij
|
|F
z
FL
|
,
10
11
|maxF
z
ij
|
|F
z
FR
|
,
10
11
|maxF
z
ij
|
|F
z
RL
|
,
10
11
|maxF
z
ij
|
|F
z
RR
|
_
for the oversteering car, where the |F
z
max
| is the maximum
normal forces of four wheels; the deactivation time
T
del
= 0.12 [s], the side slip error activation threshold
e
on

= 0.5 [deg] and deactivation threshold is e


off

= 0.75e
on

.
The choice of the horizons length is a compromise between
computational load and necessary information for the
prediction model. We present results and performance at
140 [km/h] for the oversteering sport car and we omit the
comparison with open loop simulation because the when
performing this manoeuvre at 100 [km/ just lose stability.
In Figure 6 we show the value of : longitudinal speed v
x
in km/h; the lateral acceleration a
y
in g

s; the car yaw rate


(solid line) and reference yaw rate (dashed line) in deg/s;
the steering wheel angle SWA in deg; the side slip angle
in deg. It is clear that during the manoeuvre the longitudinal
speed decreases smoothly to about 90km/h, the lateral
acceleration reaches the maximum value when the steering
angle is constant and the yaw rate tracking is satisfactory.
There is no signicant high frequency chattering, thus
guaranteeing the drivers comfort. Notice too that the
maximum value of the sideslip angle is 4

. We wish to
point out that the peak of the yaw rate signal toward the
ThB12.6
4629
Nomenclature

i j
wheel side slip angle vehicle side slip angle

estimated wheel turn angle

i j
tire-road friction coefcient
i j
angular speed of the tire
yaw rate e

yaw rate error
e

side slip angle error


re f
reference yaw rate

re f
reference yaw rate a
x
vehicle longitudinal acceleration
a
y
vehicle lateral acceleration c
i j
tire cornering stiffness
l
a
longitudinal distance from CoG to the front axle l
b
longitudinal distance from CoG to the rear axle
2l
c
track width M mass of the vehicle
r
i j
rolling radius of the tire s
i j
longitudinal slip ratio
v vehicle velocity v
ch
characteristic speed
v
wi j
wheel velocity vector of the tire center v
x
velocity of the vehicle along the x direction
v
y
velocity of the vehicle along the y direction F
Li j
longitudinal force on the tire
F
Si j
lateral force on the tire F
Zi j
vertical force on the tire
J
z
yaw inertia moment of vehicle around the zaxis J
w
inertia of wheel around the yaxis
T
eng
i j
the contribution of engine torque on each wheel T
B
braking torque
end of the second steering change, which depends on the
sudden change of steering and subsequent loss of authority
of tires on one side, is not immediately followed by new
action on the other side.
In Figure 7 we show the control action. The second plot
depicts the variation of longitudinal forces. These have to be
inside their physical limits, which depend on load transfer
and normal force on each wheels, so that when the car is
curving on the right, the left forces increase and the right
forces are approximately zero, and viceversa on a left curve.
In other words tires with a larger instantaneous vertical load
have a greater authority. In the third plot the nal control
input (the pressure of brake cylinder) are depicted. The
slew rate of each signal depends on the limitation of input
variation (see 13b). These signals are computed only when
the controller is enabled; last plot shows the enabling signal.
0 2 4 6 8 10 12
100
120
140

V
x

[
k
m
/
h
]
0 2 4 6 8 10 12
1
0
1

a
y
[
G
s
]
0 2 4 6 8 10 12
20
0
20

d
<
/
d
t

[
d
e
g
/
s
]
0 2 4 6 8 10 12
50
0
50

S
W
A

[
d
e
g
]
0 2 4 6 8 10 12
2
0
2

E

[
d
e
g
]
Fig. 6. Simulation with control strategy @140 km/h, vehicle variables:
(a) longitudinal velocity, (b) lateral acceleration, (c) yaw rate and reference
yaw rate, (d) steering wheel angle, and (e) side slip angle.
VII. CONCLUSIONS
In this paper a novel integrated vehicle dynamics control
has been presented that utilizes differential braking. It is
based on an MPC force controller cascaded with a slip
controller. The simulations were performed on an oversteer-
ing sport car using an ELASIS-CRF proprietary simulator
and the MIMO controller yields encouraging results. The
strategy will be hardware-in-the-loop tested before being
implemented on test car. The next steps of our work will
be the introduction of on-line estimation of the friction
0 2 4 6 8 10 12
-100
0
100

S
W
A

[
d
e
g
]
0 2 4 6 8 10 12
-4000
-3000
-2000
-1000
0


F
L
ij
[
N
]
0 2 4 6 8 10 12
0
10
20
30

p
-
B
C

[
b
a
r
]


0 2 4 6 8 10 12
0
0.5
1
1.5

e
n
a
b
l
e
FL
FR
RL
RR
Fig. 7. Simulation with control strategy @140 km/h, control variables
from top: (a) steering wheel angle, (b) MPC control input (variation of
longitudinal tire force), (c) output of slip controller (oil pressure in the
braking system), and (d) activation signal of control.
coefcient and the use of engine torque in the control
strategy.
REFERENCES
[1] R. Rajamani, Vehicle Dynamics and Control. New York: Springer-
Verlag, 2005.
[2] K. K. Dongshin Kim, W. Lee, and I. Hwang, Development of mando
esp (electronic stability program), in 2003 SAE World Congress,
Proceeding of the, no. SAE 2003-10-0101, March 2004.
[3] A. T. van Zanten, R. Erthadt, and G. Pfaff, VDC, the vehicle dynamics
control of Bosch, in International Congress and Exposition, 1995.
Proceeding of the, ser. SAE950759, March 1995, pp. 926.
[4] P. Falcone, F. Borrelli, J. Asgari, H. E. Tseng, and D. Hrovat, Pre-
dictive active steering control for autonomous vehicle systems, IEEE
Transactions on Control Systems Technology, vol. 15, 2007.
[5] O. Barbarisi, G. Palmieri, S. Scala, and L. Glielmo, Dynamically
constrained differential braking for yaw rate control and side slip
control, Proc. European Control Conference ECC 09, August 2009.
[6] U. Kienke and L. Nielsen, Automotive Control Systems. Springer,
2000.
[7] H. Pacejka, Tire and Vehicle Dynamics. SAE International, 2006.
[8] O. Barbarisi, G. Palmieri, S. Scala, and L. Glielmo, LTV-MPC for
yaw rate control and side slip control with dynamically constrained
differential braking, European Journal of Control, August 2009.
ThB12.6
4630

You might also like