Copyright © 2002 IFAC

15th Triennial World Congress, Barcelona, Spain

DETECTION OF LATERAL VEHICLE DRIVING CONDITIONS BASED ON THE

CHARACTERISTIC VELOCITY

Marcus Börner*, Luc Andréani*, Pedro Albertos#, Rolf Isermann*

*
Institute of Automatic Control, Darmstadt University of Technology,
Landgraf-Georg-Str. 4, D-64283 Darmstadt, Germany; Phone: +49 6151 16-7412
{mboerner, landreani, risermann}@iat.tu-darmstadt.de
#
Departement of System Enineering and Control, Universidad Politécnica de
Valencia, PO Box 22012, E-46071 Spain; Phone: +34 963879570
[email protected]

Abstract: Using real-time and online models in vehicle control and fault diagnosis
necessitates knowledge about time variant physical parameters and current driving
situation. Detailed information about vehicle conditions, in form of a Characteristic
Velocity Stability Indicator, is used to select the corresponding vehicle model structure
for adaptive drive dynamic control. After a short introduction to a lateral vehicle model,
a new approach for the online calculation of different driving conditions (i.e. stability,
understeering, oversteering, and neutralsteering) is given. Copyright © 2002 IFAC

Keywords: Vehicle dynamics, stability, driving situation, characteristic velocity.

1 INTRODUCTION 2 VEHICLE MODEL

Modern vehicles are equipped with an increasing For deriving the lateral dynamics, a coordinate
number of driver assistance systems. These systems system is fixed to the center of gravity (C.G.) and
are designed to aid the driver by preventing any Newton’s laws are applied. Roll, pitch, bounce, and
unstable or unpredictable vehicle behavior. deceleration dynamics are neglected to reduce the
Although they are using real-time and online model to two degrees of freedom: the lateral position
models, fault detection and diagnosis is mostly and yaw angle states. Further simplifications assume
performed by relatively simple algorithms, which do that each axle shares the same steering angles and
not work correctly in certain driving situations. One that each wheel produces the same steering forces
approach to cope with the problems encountered by (Fig. 1).
these simple algorithms is to calculate the driving
condition and react with a reconfiguration or y o ,ϕ o
FxF
adaptation of models to this new situation. ICM
lF δ st
x o ,κ o
z0 ,ψ 0
Former work was based on fuzzy logic systems (e.g. l FyF
ρ
Albertos and Börner, 1999; Börner and Albertos, v
1999). This approach was able to detect lR β
y ,ϕ
understeering, oversteering, and braking in the curve ψ
x,κ
situations and used five measured input signals z,ψ
(steering wheel input δst, lateral acceleration ÿ, yaw C.G.
m
v2
ρ
rate ψ& , longitudinal acceleration &x& , and velocity v). J zψ&&
R mv&

In this work, a new approach for the online FxR FyR

discrimination of different driving situations based

on the characteristic velocity is presented. The
advantage is the usage of only little sensor
information. Results are then checked against real Fig. 1: Second order non-linear one-track model
measurements of a test vehicle. Isermann (2001) explains the resulting non-linear
dynamic model, known as the bicycle model, in
detail.
x& (t ) = A(t ) ⋅ x(t ) + b(t ) ⋅ u (t ) parameter variations a linear investigation based on
y (t ) = C(t ) ⋅ x(t ) the transfer functions
where 11
ψ& ( s ) b0 +11b1 s
G11 ( s ) = =
 y& (t )   y& (t )  δ st ( s ) 11
a 0 + a1 s + 11a 2 s 2
11
x(t ) =   y (t ) =  
ψ (t )
& ψ& (t )  11 ψ& 1 cα' F cαR vl
 b0 = = '
 cα′ F (t ) + cαR (t ) + mv&(t ) cαR (t )lR − cα′ F (t )lF   δ st stat
2 2 '
cαF cαR l + mv (l R cαR − l F cαF ) i st
−  
mv(t ) mv(t )  11b1 = mvl F 11b0
A(t ) =  2 2
  cαR l
 c αR (t )l R − cα ′ F (t )lF cαR (t )lR + cα′ F (t )lF   11
 −  where  a 0 = 1
 J z v(t ) J z v(t )   ' 2 ' 2
 11 a = 2 D = J z v(cαF + cαR ) + mv(l F cαF + l R cαR )
 cα′ F (t )   1 ω0 ' 2 2 '
cαF cαR l + mv (l R cαR − l F cαF )
 mi  1 0 
b(t ) =  st  C(t ) =    11 1 J z mv 2
 cα′ F (t )lF  0 1  a2 = 2 = '
 ω 0 cαF cαR l + mv 2 (l R cαR − l F cα' F )
2
 J z ist 
u (t ) = δ st (t ) (2)
and
21
(1) &y&( s ) b0 + 21b1 s + 21b2 s 2
G 21 ( s ) = =
δ st ( s ) 21
a 0 + 21 a1 s + 21a 2 s 2
Although the bicycle model is relatively simple, it
 21 lR J z 21
has been proven to be a good approximation for  b0 =
&y&
= v11 b 0 21
b1 = 21
b0 21
b2 = b0
 δ st v cαR l
vehicle dynamics when lateral acceleration is limited where  stat

 21 a = 1 2D 1
to 0.4 g on normal dry asphalt roads (Milliken and  0
21
a1 = =11 a1 21
a2 = =11 a 2
 ω0 ω 02
Milliken, 1995). Note that the velocity v and the
cornering stiffness c'αF and cαR are assumed to be (3)
time-variant. and
31
β (s) b0 + 31b1s
G31 ( s ) = =
Table 1: Vehicle Parameters δ st ( s ) 31
a0 + 31a1s + 31a 2 s 2
 31 β l R  ml R v 2 11 J zv
Sym Description Value Unit  b0 = = 1− b0 31
b1 = 31
b0
 δ st v  cαR l R l  cαR l R l − l F mv 2
bol where  stat

x longitudinal position error - [m]  31 a = 1 31 2D 1

 0 a1 = =11a1 31
a2 = =11a 2
ω0 ω 02
y lateral position error - [m] 
z vertical position error - [m] (4)
κ roll angle - [rad]
ϕ pitch angle - [rad] is carried out. To derive the transfer function during
ψ yaw angle - [rad] normal driving situations, all parameters have to be
δ steering angle - [rad]
constant. The transfer functions resulting from
δst steering wheel angle - [rad]
Eq. (1) are considered with s as the Laplace
β side slip angle - [rad]
transform variable.
m vehicle mass 1720 [kg]
Jz mom. of inertia, z-axis 2275 [kgm2]
v longitudinal velocity - [m/s] In the following section the poles and zeros of the
c'αF effective front wheel cornering 50000 [N/rad] transfer function G1(s) are firstly studied for varying
stiffness velocity v and then for varying front and rear wheel
cαR rear wheel cornering stiffness 60000 [N/rad] cornering stiffness c’αF and cαR.
lF,lR length of front, rear axle from 1.3/ [m]
C.G. 1.43 The results of these investigations are important to
l length between front and rear 2.73 [m] get a first insight into the stability and the transient
axle
behavior of the one-track model (Isermann and
ist steering system gear ratio 13.5 [-]
ρ radius - [m] Börner, 2002). Hereby, cαR = 1.2 c’αF was assumed,
FxR longitudinal force acting on - [N] the remaining vehicle parameters were set to the
rear tire values according to Table 1.
FxF longitudinal force acting on - [N]
front tire 3.1 Velocity variation
FyR side force acting on rear tire - [N]
FyF side force acting on front tire - [N] While the velocity v of G1(s) increases from 7.2
R radius - [m] km/h to 180 km/h, the poles of the systems remain in
ICM instantaneous center of motion - [-]
the left hand s-plane. Note that Fig. 2 (poles as x and
roots as o), is not a standard root locus because the
3 PARAMETER VARIATION parameter v enters nonlinearly. For all velocities v ≥
0 km/h the vehicle has stable dynamics. However,
Changes of parameters affect the different lateral for v>20 km/h, the system becomes less damped, i.e.
vehicle behaviors. To discuss the influence of the pair of real eigenvalues meets at the real axis and
branches into a pair of complex eigenvalues.
 The cornering stiffness may become small for icy
roads but never negative. The zeros of the system
are always on the left real axis. With decreasing
front and rear cornering stiffness c’αF,R the gain also
decreases and the dynamics become less damped.

Fig. 2: Parameter root locus G1(jω) for changing

velocity v ∈ [7.2 km/h;180 km/h] and constant
cornering stiffness cαR=1.2c’αF=60000 [N/rad]
The corresponding step responses of transfer
function G1(s) are graphed in Fig. 3.
Fig. 5: Step responses for step inputs of the
steering angle δst and different cornering stiffness
cαF,R..
The vehicle is unstable, if the cornering stiffness cαR
is 0 N/rad. Until now, vehicle stability of vehicles
was investigated decoupled from other parameters.
However, the stability is a function of many vehicle
parameters:

stability = f (m(t ), J z (t ), lF (t ), lR (t ), cα′ F (t ), cαR (t ), v(t ) ) (5)

In the next sections, an approach is presented to

calculate the stability with the characteristic
Fig. 3: Step responses for step input of the steering velocity, which is a function of
angle δst and different velocity v
vch (t ) = f (m(t ), lF (t ), lR (t ), cα′ F (t ), cαR (t ) ) . (6)
3.2 Cornering stiffness variation
4 DERIVATION OF THE CHARACTERISTIC
While the front and rear wheel cornering stiffness VELOCITY
c’αF,R are increased in the range between 0 to 100000
A stationary circular drive is assumed with
N/rad, the vehicle behaves stable for positive c’αF,R
and becomes unstable with a double pole at the
coordinate-origin for c’αF,R = 0 N/rad.
δ st = const. , ψ& = const. (7)

and

β& = 0 , v& = 0 . (8)

The dynamic equation of motion leads to the

algebraic relationships (Mitschke, 1990)

ψ& (t ) 1 v(t )
=
δ st (t ) ist l  v(t ) 
2
(9)
1+  
 v (t ) 
 ch 
and
&y&(t ) 1 v 2 (t )
Fig. 4: Parameter root locus of G1(jω) for different = .
δ st (t ) ist l  v(t ) 
2
(10)
cornering stiffness cαR=1.2c’αF ∈ [0 N/rad;100000
1+  
N/rad] and constant velocity v = 14 [m/s] ≈ 50  v (t ) 
 ch 
[km/h]
The term vch(t) is termed the characteristic velocity. 5 STABILITY OF A VEHICLE

cα′ F (t )cαR (t )l 2 Based on the state equation (1) the characteristic

2
v ch (t ) = (11) equation det(sI-A)=0 of the lateral vehicle dynamics
m(cαR (t )l R − cα′ F (t )l F ) becomes

(The characteristic velocity becomes complex for a1

644444 47 444444 8
cαR(t)lR<c’αF(t)lF) and v2ch(t)<0). In modern vehicles 2 ' 2
( J + ml F )cαF + ( J z + ml R )cαR
the characteristic velocity ranges between 15-35 m/s s2 + z s
and is time-variant. J Z mv
With the measured steering wheel angle δst(t) as the 644444447 a0
44444448
input, the velocity v(t) and the yaw rate ψ& (t) as the cα' F cαR (l F + l R ) 2 + mv 2 (cαR l R − cα' F l F )
output, the characteristic velocity vch(t) follows from + =0
J Z mv 2
Eq. (9)
(14)
v 2 (t ) According to the Hurwitz stability criterion, stability
2
vch (t ) = − requires that
δ (t )v(t ) (12)
1 − st
ψ& (t )ist l
a1 > 0 and a0 > 0
or from Eq. (10) As a1>0 is always satisfied, because no negative
values arise, only a0 has to be considered. For
2 v(t ) understeering vehicles, a0>0 always holds true since
vch (t ) = −
δ st (t )v(t ) (13)
1− cαR l R > cα' F l F .
&y&(t )ist l
However, for oversteering vehicles with
Hence, it can be determined by measuring either cαR l R < cα' F l F
ψ& / δ st or &y& / δ st . The characteristic velocity can be the sign of a0 depends on the velocity v. With the
used to examine the stability. characteristic velocity (11), the following stability
condition results:
The driving situation detection is realized via the
a0
calculation of the characteristic velocity vch and a 6444444 47 4444444 8
driving situation decision logic, Fig. 6. cαF cαR (lF + lR ) + mv 2 (cαRlR − cα' F lF ) > 0
' 2

cα' F cαR (lF + lR )2 (15)

⇒ v2 > −
v, δst ψ& ∨ &y& m(cαRlR − cα' F lF )
Process
⇒ v 2 > −1 ⋅ vch
2

Calculation of The quadratic characteristic velocity can be positive

characteristic or negative in dependency of cαR l R and cα' F l F :
velocity
vch
< 0 ∀ cαR l R < cα' F l F
CVSI 2  '
Driving situation v ch = 0 ∀ cαR l R = cαF l F (16)
decision logic > 0 ∀ c l > c ' l
 αR R αF F

Fig. 6: Driving situation detection with output CVSI This means that in the case of an oversteering
(Characteristic Velocity Stability Indicator) vehicle, the vehicle becomes unstable if the
quadratic velocity is smaller than the negative
The input of the model is the steering wheel angle quadratic characteristic velocity.
signal δst. As output signal the yaw rate sensor ψ& or
Further more for small steering angles δ st < δ stthres ,
lateral accelerometer ÿ can be used to calculate the
it is assumed that the driving condition is mainly a
characteristic velocity vch. With help of the on-line
straight run, just compensating for disturbances. If
calculated characteristic velocity vch, the over ground
velocity v, and the steering wheel angle δst the δ st > δ stthres cornering can be assumed. Then a
current driving situation can be concluded. If the classification of different driving situations can be
vehicle is in unstable conditions, the situation can be made as shown in Table 2. This driving condition
classified while cornering into understeering, detection uses only a few sensor signals: δst ,v, ψ& .
neutralsteering and oversteering. The same classification can be defined for lateral
acceleration measurements. Table 2 introduces a
Characteristic Velocity Stability Indicator (CVSI) to
distinguish between different driving and stability The test vehicle is equipped with special sensors for
conditions. measuring the following signals: The steering wheel
angle δst, lateral acceleration ÿ, yaw rate ψ& , and
Table 2: Classification of different driving
conditions ABS velocity v1..4.
Now a driving experiment is considered for a dry
road containing two different types of lateral
Signal processing Driving condition Stability
excitation. First, a slalom test drive was performed
followed by a longitudinal acceleration and another
^ ( ψ& < ψ& ) - Stable slalom course. The lateral acceleration of the slalom

Straight run
threshold
(δst < δst,Lim)
^ test drive was partially higher than 6 m/s2.
(v2 > 0) Braking / In the upper plot of Fig. 9 the steering wheel angle
^ ( ψ& ≥ ψ& ) accelerating Unstable
threshold
under µ-split
δst is shown. The lower plot shows the ABS wheel
speed velocities.
^ - Understeering Stable
(vch2≥
0) ^
(vch2>>v2)
Neutralsteering Stable
^
Cornering

(δst ≥ δst,Lim) (|vch2|>>v2)

^
(v2 > 0) ^
^ Oversteering Stable
(|vch2| > v2)
(vch2<
0) ^ High In-
(|vch2| = v2) oversteering different
^
(|vch2| < v2) Break away Unstable

Fig. 7 demonstrates the whole driving condition

detection.

Classification number
δst ψ& CVSI Fig. 9: Slalom experiments:
process
a) Steering wheel angle signal δst
calculation of vch
b) Related ABS speed sensor signals for a
^ ((dψ/dt) < (dψ/dt)thres) 0
slalom test drive (vFR=Front right, vFL=front
v2
v
v 2ch,opt = −
δ st v δL < δL,Lim ^ ((dψ/dt) ≥ (dψ/dt)thres) -1
left, vRL=rear left, vRR=rear right, v=Correvit
1−
ψ&ist l
- 1
sensor)
2
2 ^ (vch ≥0)
v ch ^ (vch2>>v2) CVSI
(v2>0)

Opt

2 2
2
^ (|vch |>>v )
δL ≥ δL,Lim
2 2
^ (|vch | > v ) 3
^ (vch2<0)
^ (|vch2| = v2) 4
^ (|vch2| < v2) 5

Fig. 7: Driving condition detection system

6 EXPERIMENTAL RESULTS Zoom see Fig. 11

The following results are based on experimental

data, which have been obtained using an Opel
Omega vehicle on an airfield runway, (Fig. 8).

Fig. 10: Slalom experiments as Fig. 9

a) Yaw rate ψ&
b) Output of the driving condition detection
(Characteristic Velocity Stability Indicator
CVSI), which was calculated using the yaw
rate and the steering wheel angle δst
After the slalom test-drive the vehicle accelerates to
a velocity of 15 m/s and is getting excited with a
steering wheel frequency in the range of the natural
Fig. 8: Opel Omega test-vehicle at an airfield near frequency of the yaw dynamics (0.5 Hz). The test
Darmstadt.
vehicle reacts with an unstable behavior at t=18 s, as on the oversteering situation by reducing the steering
can be seen from the wheel velocities. Fig. 10 wheel input (CSVI=-1 unstable straight run) and by
illustrates the corresponding results of the driving counter-steering (CSVI=5 unstable cornering
condition detection based on the yaw rate and the condition). With counter-steering, it is possible to
steering wheel input. The upper plot displays the stabilize the vehicle (CSVI=1).
lateral acceleration ÿ. The sideslip angle β in the
second slalom test reaches 0.12 to 0.18 radiants for 7 CONCLUSIONS
18s < t < 22s which indicates a critical driving
situation. Here the driver stabilized the vehicle by A new approach to detect critical situations was
counter-steering, such that it was stable for t > 22 s. presented and validated by test-drives. This new
method is based on measurement of the steering
This application shows that the derived critical
driving condition indicator (CVSI) can be applied wheel angle, velocity, and either yaw rate or lateral
acceleration. The experiments show that the
for detecting the driving situation by on-board
calculations. classification of critical driving situations via the
Characteristic Velocity Stability Indicator (CVSI)
was done correctly and can be physically explained.
Table 3 defines the Characteristic Velocity Stability
Indicator (CVSI) for different driving and stability
conditions. REFERENCES

Albertos, P. and Börner, M. (1999). Neuro-fuzzy

Table 3: Characteristic Velocity Stability Indicator approach in measurement: Road vehicle
(CVSI) velocity estimation. IEEE International
Workshop on Intelligent Signal Processing, 4.-
Driving Condition Stability CVSI 7. September 1999, Budapest, Hungary.
- Stable 0 Börner, M. and Albertos, P. (1999). Intelligent Data
braking / unstable -1 Fusion: Vehicle Driving Monitoring. Interkama
straight run
accelerating under ISA Tech Conference, 18. -20. October 1999,
µ-split Düsseldorf, Germany.
understeering stable 1 Börner, M.; Semmler, S.J.; Isermann, R. (2001).
neutralsteering stable 2 Estimation of Vehicle Velocity based on
cornering oversteering stable 3 Suspension Signals. European Control
high oversteering indifferent 4 Conference 2001 (ECC`01), 3. -7. September
break away unstable 5 2001, Porto, Portugal.
Isermann, R. (2001). Diagnosis Methods for
The driving condition algorithm detects the unstable Electronic Controlled Vehicles. IEEE
situations as expected. In addition, the classification Transactions on Vehicular Technology, USA
of normal situations can be physically explained. By Isermann, R. and Börner, M. (2002). Supervision,
starting the critical driving situation, the driving Fault Detection and Diagnosis for Mechatronic
condition algorithm detects an understeering vehicle Automobiles. 3. Braunschweiger Symposium
(CVSI=1, see also Fig. 11). Automatisierungs- und Assistenzsysteme für
Transportmittel, 27. and 28. February 2002,
Braunschweig, Germany.
Milliken, W.F. and Milliken, D.L. (1995). Race Car
Vehicle Dynamics. USA.
Mitschke, M. (1990). Dynamik der Kraftfahrzeuge.
Bd. C. Fahrverhalten, 2. Aufl., Springer Verlag,
Berlin, Heidelberg, Germany.
Sienel, W. (1997). Estimation of the Tire Cornering
Stiffness and its Application to Active Car
Steering. Proceedings of the 36th Conference on
Decision and Control, San Diego, California,
USA.

Fig. 11: Steering wheel angle and Characteristic

Velocity Stability Indicator (CVSI),
CVSI=0,-1,1,2,3,4,5

Next, an oversteering condition (CVSI=3) is

detected due to the inertia of the vehicle and
especially the steering wheel input. The driver reacts