YAW DISTURBANCE ATTENUATION BY ROBUST

DECOUPLING OF CAR STEERING

J. Ackermann T. B
unte

Deutsche Forschungsanstalt
f
ur Luft- und Raumfahrt (DLR) Oberpfaffenhofen
Institut f
ur Robotik und Systemdynamik
82234 Wessling, Germany
e-mail: [email protected]

Abstract: Robust decoupling of the lateral and yaw motions of a car has been
achieved by feedback of the integrated yaw rate into front wheel steering. In the
present paper the yaw disturbance attenuation is analyzed for a generic single-track
vehicle model. The frequency limit, up to which yaw disturbances are attenuated,
is calculated. For specific vehicle data, it is shown that this control law significantly
reduces the influence of yaw disturbances on yaw rate and side-slip angle for low
frequencies. This safety advantage is experimentally verified for -split braking.
Keywords: Automotive control, decoupling problems, robust control, disturbance
rejection, sensitivity functions

1. INTRODUCTION

For the case of non ideal mass distribution

the reader is referred to (Ackermann, 1994).

Consider the vehicle of Fig. 1.

The input to the system is the front-wheel steering
angle f , and the output is the yaw rate r =
measured by a gyro. is the heading angle
between an inertial coordinate system (x0 , y0 ) and
the chassis coordinate system (x, y). Uncertain
parameters are
(1) the velocity v = |~v | > 0,
(2) the cornering stiffness of the tires,
(3) the masses mf and mr at the front and
rear axles; it is assumed, however, that the
location of the center of gravity (CG) at a
distance `f from the front axle and `r from
the rear axle is known, i.e. mf `f = mr `r .
The wheelbase is ` = `r + `f and the vehicle
mass m = mr + mf . The assumed ideal mass
distribution with mr = m`f /`, mf = m`r /`
implies the moment of inertia Jz w.r.t. a
vertical axis through the CG as
Jz = mr `2r + mf `2f = m`r `f .

(1)

For small sideslip angle and small steering

angle f the single-track model of car steering is
(Mitschke, 1990):
mv( + r) = Fy + Fd
Jz r = Mz + Md .

(2)

The lateral force Fy and the torque Mz around

a vertical axis through the CG are generated
by side forces of the tires. Also, a disturbance

Fig. 1. Vehicle with ideal mass distribution

force Fd and a disturbance torque Md act on

the vehicle. Examples are crosswind, lateral road
slope, flat tires, and asymmetric braking forces
on a slippery road. In particular, an unexpected
disturbance torque Md can lead to dangerous
driving situations because of the reaction time
of the driver, followed by overreactions. Timedelay and high gain may even cause instability of
the driver-vehicle feedback system. An automatic
control system with feedback of the yaw rate r
can react faster and more precisely than a human
driver (Ackermann et al., 1993; Ackermann et
al., 1996). By integration of r it also provides
a relative direction reference with respect to the
initial heading of the car before the disturbance.
The driver does not have to worry about the
automatically controlled yaw motion. He only has
to command a lateral acceleration to keep the
car, considered as a mass point, on his planned
path. He can easily compensate for the lateral
disturbance forces Fd , e.g. from a lateral slope of
the road; therefore Fd = 0 was set for simpler
notation.
The next sections show the derivation of the disturbance transfer function from Md to the sideslip
angle at the front axle and to the yaw rate, first in
Section 2 for a conventional car, then in Section
3 for a robustly decoupled car. In Section 4 the
steady-state effects and in Section 5 the ratio of
the transfer functions (sensitivity function) are
analyzed, and the frequency limit of disturbance
attenuation is calculated. Section 6 gives an example, and Section 7 shows experimental results.

2. YAW DISTURBANCE TRANSFER

FUNCTION OF THE CONVENTIONAL CAR
Equations (1) and (2) with Fd = 0 may be written

mv( + r)
m`r `f r

"

Fy
Mz

"

0
Md

(3)

Fy
Mz

"

`f `r

#"

Ff (f )
Fr (r )

(4)

The tire forces depend on the tire sideslip angles

f and r , as illustrated in Fig. 2 for the front
wheel. The local velocity vector ~vf forms the
(chassis) slip angle f with the car body and the
tire slip angle f with the tire direction; thus
f = f f
r = r r .

In this paper rear wheel steering is not used,

i.e. r 0. The front sideslip angle f will be
introduced as a state variable. The sideslip angles
at the CG () and at the rear axle (r ) are related
to f by the kinematic relations for small angles
`f
r
v
`
r = f r.
v
= f

(6)

The tire force characteristics are linearized as

Ff (f ) = cf f

(7)

Fr (r ) = cr r

where the cornering stiffnesses cf and cr are uncertain parameters that vary with the road tire
contact. It is assumed that cf = cf 0 , cv = cr0
where cf 0 and cr0 are nominal values for the dry
road and [ ; 1] is an uncertain parameter
with > 0. Further uncertain constant parameters are the vehicle mass m [m ; m+ ] and
velocity v [v ; v + ]. The model (3), (4) with
the above equations substituted becomes

`f

mv(f r + r)

=
v
(8)
m`r `f r

#
"
#
"
cf (f f )
1 1
0

` +
`f `r
Md
cr (f + r)
v

and, solving for f and r,

The steering force Fy and torque Mz are generated

by the lateral tire forces Ff (f ) and Fr (r ) via
"

Fig. 2. Variables of the tire model

(5)

1

f
mf v 0
=

1
1
r
mf ` mr `

cf (f f )

`
cr (f + r)
v

1

1
mf v`

r +
Md
mr + m f
0
mr mf ` 2

(9)

where the front and rear masses mf = m`r /`

and mr = m`f /` have been substituted to replace
m, `r and `f . The state equations of the system
are then

cf


1
f
f

m
v
f

=
+



1 cr
cf
cr r
r

` mr
mf
mr v
(10)
1

v
v
cf
f + 1
Md .
+

mf 1
mf ` mr + m f
mr `
`
Laplace-transformation of (10) yields

s+

cf
mf v

1
v

(11)

Md (s)

f (s) +

m + m mf ` .
r
f
mr `

B(s)
1

Md (s)
D(s)mf `
r(s)
R(s)
f (s)

1
B(s) =
v

R(s) =

cr
(mr + mf )v
s+

mr v
mr `

mr + m f
mr `



D(s) = s +

cf
mf v

s+

cr + c f
v(mr + mf )

s+

1
`

cr
mr v

cf
mf v

1+

cf
mf vs

1
v



1 cf
cr
cr
cf
s+

+
` mf
mr
mr v
mf `s

f (s)
r(s)

dec

(14)

Md (s)

.
mr + m f mf `
mr `

Bdec (s)
1

(15)
=
Ddec (s)mf `
Rdec (s)
rdec (s)
f dec (s)

Bdec (s) = sB(s)

cf
mr v`

Rdec (s) = sR(s)

(12)



cf
cr
Ddec (s) = sD(s) +
s+
mf `
mr v



cf
cr 
v
= s+
s2 +
s+
.
mf v
mr v
`
The control law (13) provides a reference for the
inertial front wheel orientation + f , where
Then
r = .

(t) = (t0 ) +

Zt

r( )d

t0

(t) = (t0 ) + f (t0 ) f (t)

(16)

(t) + f (t) = (t0 ) + f (t0 ).

cr
cf

mf
mr

In the decoupled car the inertial orientation of the

front wheels is not influenced by a disturbance
torque Md .

4. STEADY-STATE EFFECT OF YAW

DISTURBANCES

3. YAW DISTURBANCE TRANSFER

FUNCTION OF THE ROBUSTLY
DECOUPLED CAR
Robust decoupling (Ackermann et al., 1993) is
based on the requirement that the yaw rate r is
made unobservable from the lateral acceleration
af at the front axle. This goal is achieved by the
feedback control law
f = r + F L .

s+

Solve (11) for f (s) and r(s) and let f (s) 0.

Solve (14) for f dec (s) and rdec (s).

f (s)

r(s)



1 cf
cr
cr
s+

` mf
mr
mr v
1
v
cf

mf 1
`

Substitute f (s) = r(s)/s into (11):

(13)

It is assumed here that the steering wheel input

L of the driver is zero.

For a step disturbance input, Md (s) = 1/s, the

conventional car has the steady-state response
f st =

cr ` (mr + mf )v 2
B(0)
=
D(0)mf `
cf cr `2 v 2 `(cf mr cr mf )

(17)
(cr + cf )v
R(0)
=
.
rst =
D(0)mf `
cf cr `2 v 2 `(cf mr cr mf )
The robustly decoupled car has the steady-state
response

Bdec (0)
1
=
Ddec (0)mf `
cr `

Rdec (0)
= 0.
rst,dec =
Ddec (0)mf `

2.5
v/(km/h): 15, 29, 57, 112, 220
2

(18)

|rhor|

f st,dec =

1.5
v=15 km/h
1
0.5
v=220 km/h

The yaw rate goes to zero, i.e. the car has a

constant heading angle as a steady-state response
to a step yaw disturbance.

(s) =

Bdec (s)D(s)
Ddec (s)B(s)

Rdec (s)D(s)
sD(s)
=
Ddec (s)R(s)
Ddec (s)


cf
r (s)
(s) = 1
mr v`sB(s)
r (s) =

(19)

as disturbance attenuation ratios (sensitivity functions). The influence of the yaw disturbances on
f or r is reduced for all frequencies for which
| (j)| < 1 or |r (j)| < 1, respectively.
Let a = cf /mf b = cr /mr , then
r (s) =

[(s + a/v)(s + b/v) (a b)/`]s

(s + a/v)[s2 + (s + v/`)b/v]

(20)

1.6

0.2

0.4

0.6
0.8
1
omegal/(2*Pi) [Hz]

1.2

1.4

1.6

10

10

Fig. 3. Disturbance attenuation (sensitivity function) and frequency limit for varying velocity
p
`2 = c + c2 + ab2 /2`v 2
(25)
2b a
b2
c=
2.
4`
2v
The yaw disturbances are attenuated for frequencies
< ` , and amplified for > ` .
6. EXAMPLE

cf = 49400N/rad, cr = 103800N/rad

(21)

(22)

(23)

v ` (a + b) > 0.
which may be simplified to


ab2
b2
2b a
4
2 2
> 0. (24)

2`
v
2`v 2
The frequency limit is

1.4

Fig. 3 shows the disturbance attenuation at five

velocity values. The frequency limit in the lower
figure is between 0.6 and 1 Hertz.

(v 2 2 + a2 )[v 2 (` 2 + b)2 + `2 b2 2 ]
2 2

1.2

and uncertain velocity v [15 ; 220] km/h.

|r (j)|2 < 1 if and only if

[ab` v 2 ` 2 (a b)v 2 ]2 2

0.8
1
omega/(2*Pi) [Hz]

`f = 1.514m, `r = 1.323m, m = 1916kg,

[ab` v 2 ` 2 (a b)v 2 ]2 + v 2 `2 (a + b)2 2

=
(v 2 + (a/)2 )[(v` 2 + vb)2 + `2 b2 2 ]

r (0) = 0

0.6

Consider a car with the following parameters

[v 2 `s2 + v`(a + b)s + ab` (a b)v 2 ]s

(vs + a)[v`s2 + (`s + v)b]

|r (j)|2 = r (j)r (j) =

0.4

Define

0.2

10

v/(km/h)

5. DISTURBANCE ATTENUATION RATIOS;

FREQUENCY LIMIT

0
0

7. EXPERIMENTAL RESULTS
A periodic disturbance torque Md is not typical
for car driving. A much harder test is a step
disturbance Md (s) = M/s. This is generated experimentally in the split test shown in Fig. 4.
The left-hand column of pictures, taken at 1second intervals, shows the experiment with a conventional car. The car drives with its right-hand
wheels (left-hand side of the figure) on waterflooded tiles with a friction coefficient 0.1,
and with the left-hand wheels on wet asphalt
with a friction coefficient 0.9. At an initial
velocity of 80 km/h the driver brakes and keeps
his steering wheel straight. In the second picture
the beginning of skidding is recognizable; only one
second later the car has all four tires on asphalt in
a safe situation. In normal driving on an icy road it
takes the driver just this second to react, and that
may be too late. The right-hand column shows the

Fig. 4. -split-braking at 80 km/h, left: conventional vehicle, right: robustly decoupled vehicle

v/[km/h]

conventional car

decoupled car

60

60

40

40

20

20

f[], L[] ()

6
10

r[/sec]

10

ayCG[m/sec2]

10

10

1
0

4
t/sec

and for the decoupled car (right). The velocity v

shows the occurrence of the Md -step by braking
and the speed reduction. The angle L (handwheel
angle steering gear ratio) is kept close to zero
in both experiments, in the decoupled car a front
wheel steering angle f from the controller (13)
occurs in response to the initial peak in the yaw
rate r. In contrast to the conventional vehicle,
feedback quickly reduces r and changes its sign
such that (t) = (t0 ) according
to (17). The



lateral acceleration ayCG = v + r at the center of gravity shows a significant reduction by the
decoupling controller.

8. CONCLUSIONS
0

The robustly decoupling control law for cars attenuates yaw disturbances, e.g. from crosswinds,
flat tires, or asymmetric braking forces for low
frequencies. The frequency limit is calculated for
generic vehicle parameters. The effect for a specific car is shown by the calculated disturbance
attenuation ratio and by -split braking tests.

t/sec

Fig. 5. Experimental data for -split braking with

inital velocity v 50km/h
same experiment with the robustly decoupled vehicle. In the second picture the situation is similar
to that of the conventional car. But now the yaw
rate is measured and the controller steers the car
quickly back to the original heading angle. The
car remains straight with yaw rate back to zero,
see (18), until the complete stop. This spectacular safety improvement is not only obtained at
nominal speed, load and friction coefficients but
with perfect robustness against changes in these
parameters.
Fig. 5 shows a comparison of experimental data
for -split braking for the conventional car (left)

9. REFERENCES
Ackermann, J. (1994). Robust decoupling of car
steering dynamics with arbitrary mass distribution. In: Proc. American Control Conference. Vol. 2. Baltimore, USA. pp. 19641968.
Ackermann, J., A. Bartlett, D. Kaesbauer,
W. Sienel and R. Steinhauser (1993). Robust
control: Systems with uncertain physical parameters. Springer. London.
Ackermann, J., T. B
unte, W. Sienel, H. Jeebe
and K. Naab (1996). Driving safety by robust
steering control. In: Proc. Int. Symposium on
Advanced Vehicle Control. Aachen, Germany.
Mitschke, M. (1990). Dynamik der Kraftfahrzeuge.
Vol. C. Springer. Berlin.