EVALUATION OF VEHICLE DYNAMIC CONTROL FOR ROLLOVER

PREVENTION
Ali.Y. Ungoren and Huei Peng

University of Michigan
(Received date 12/31/2003; revised date )
ABSTRACT Evaluation of active safety control systems usually relies heavily on field testing and is time-
consuming and expensive. Advances in computer simulations make it possible to perform exhaustive design trials
and evaluations before field testing, and promise to dramatically reduce development cost and cycle time. In this
paper, a comprehensive simulation-based evaluation procedure is proposed, which combines standard evaluation
maneuvers, worst-case techniques, and a driver model for closed-loop path following evaluations. A vehicle dynamic
controller (VDC) for a popular Sport Utility Vehicle is evaluated using the proposed procedure. Simulation results
show that the proposed procedure can be used to assess the performance of the VDC under various conditions and
provides valuable information for the re-design of the VDC.

KEY WORDS: VDC, Rollover, Stability Control, Active Safety

Corresponding author, Associate Professor, Department of Mechanical Engineering, University of Michigan, Ann
Arbor, MI 48109-2133, Tel: (+)1-734-936-0352, hpeng@umich.edu
1. INTRODUCTION
The growing popularity of Sport Utility Vehicles
(SUV) over the last decade together with their higher
rollover tendency necessitates a closer look at
regulations aimed at reducing rollover fatalities.
National Highway Traffic Safety Administration
(NHTSA) reported 10,647 (19.5%) rollover deaths in
cars, vans and trucks in 2001 (NHTSA 2002), up from
10,013 (18.3%) in 1997 (NHTSA 1998). The increase in
rollover deaths seems to be closely related to the
increased population of SUVs and light trucks. The
rollover propensity of SUVs thus has caught the
attention of consumers, car companies, major suppliers,
and government agencies in recent years.
Effort by the National Highway Traffic Safety
Administration (NHTSA) on safety standard for rollover
resistance goes back to the 1970's (NHTSA 1973).
Tremendous work has been done to establish standard
procedures and regulations for vehicle roll performance
assessment and minimum performance requirements.
Susceptibility to rollover is a complex dynamic
phenomenon involving many factors such as vehicle
parameters, road conditions, and driver characteristics.
The complexity and dynamic nature of this problem
make it difficult to define a stability testing procedure
for vehicle testing. The problem becomes more
complicated when the vehicle is equipped with
advanced control devices, such as the vehicle stability
control systems (Zanten 1995, Matsumoto 1992) or anti-
rollover systems (Wielenga 1999, Hac 2002).
To assess the on-road, un-tripped rollover
propensities of vehicles, two types of metrics have been
proposed in the literature. The NHTSA dynamic testing
program proposes to use a set of rollover evaluation
maneuvers (J-turn and three Fishhook maneuvers) to
examine the frequency of vehicle Two-Wheel-Lift
(TWL) (Garrick 2002). The Consumer Unions double
lane change driving test also falls into this dynamic
testing category. On the other hand, the static type
of rollover metrics are usually based on simple
measurement of vehicle parameters related to its roll
behavior. For example, Static Stability Factor (SSF),
Tilt Table Angle, Tilt Table Ratio, Critical Sliding
Velocity, and Side Pull Ratio all fall into this category.
One of the major findings from the NHTSA Phase
IV study (Garrick 2002) is that all path-following
(driver in loop) maneuvers they tested (ISO3888,
Consumer Union Short Course, Pseudo Double Lane
Change) failed either in repeatability or discriminatory
capability. This experience provides a valuable lesson
for car companies that are designing and evaluating
advanced control systems with human driver
interactions: repeatable and discriminatory human-in-
loop tests are very hard to conduct because of human
variations. It is desirable to seek a more reliable
alternative.
A driver model was developed in (Ungoren 2004)
specifically targeting VDC evaluations. This model was
developed using the adaptive predictive control (APC)
framework. Three key features are included in the APC
framework: use of preview information, internal model
identification and weight adjustment to simulate
different driving style. The driver uses predicted
vehicle information in a future window to determine the
optimal steering action. A tunable parameter is defined
to assign relative importance of lateral displacement and
yaw angle error in the cost function to be optimized.
The model is tuned to fit three representative drivers
(average, aggressive and smooth) obtained from driving
simulator data taken from 22 human drivers. Including
this driver model in the evaluation process enables us to
evaluate the performance of VDC using virtual drivers
with different characteristics.
The main contribution of this paper is the
development of a three-stage approach for computer
evaluations of a vehicle dynamic control (VDC) system.
The procedure iterates through three stages: standard
open-loop test matrix, worst-case evaluation, and
human-in-loop simulations. The open-loop test matrix
can include regulatory (from NHTSA) as well as
company accustomed test maneuvers. The worst-case
maneuvers are computed based on the Iterative
Dynamic Programming technique. This method does
not require system gradient, and achieves fast
convergence through the use of coarse-grid-search and
approximating the identified cost function through curve
fitting. The human-in-loop simulations use three virtual
drivers based on the model described above (Ungoren
2004). The basic idea of this evaluation procedure is
illustrated by studying the rollover performance of a
VDC system.
2. VEHICLE MODEL
The vehicle studied in this paper is a popular mid-
size SUV. The vehicle model was developed by TRW
Automotive using CARSIM (2003) and the full-car
performance has been verified against test data. The
nonlinear mathematical model has 14 degrees-of-
freedom (6 DOF for the sprung mass, 2DOF for each of
the axles, and 1 DOF for each of the wheels) and is
quite suitable for simulating vehicle response under
significant (+- 10 degrees) roll motions. The vehicle
simulation model also includes a Vehicle Dynamic
Control (VDC) algorithm designed by the TRW
Automotive. This version of VDC was modified based
an early production-intent design and was meant to be
used solely for this research. It has most of the
fundamental VDC functions but is not identical to any
TRW products. The basic concept of this VDC is to
enhance vehicles yaw rate response (yaw rate
following) while maintaining small vehicle side slip
angle. The vehicle yaw motion is controlled by
applying differential braking force either to reduce the
difference between the interpreted drivers desired yaw
rate and the actual yaw rate, or to reduce vehicle side
slip angle. The drivers steering, throttle and braking
commands are disturbance inputs that influence vehicle
motions. In the meantime, they also generate reference
signal to be followed by the VDC. The evaluation
process needs to ensure that the VSC works well under
a wide array of disturbance inputs.
The vehicle roll motion was not explicitly
considered in the original VDC design. However, since
the vehicle side slip motion is regulated, the vehicle
with VDC usually has improved tripped and un-tripped
rollover resistance, because the likelihood of building
up a large lateral speed is greatly reduced. Due to the
yaw rate following functionality, it is possible the
vehicle with VDC will have slightly higher roll angle
when the driver demands a large yaw rate on high
friction roads. Due to the relative matureness of VDC
technology, it is natural to extend it for rollover
prevention purposes, hopefully without compromising
its original design goals.
A VDC designed for its original (yaw and side slip
only) goals is put through the proposed evaluation
process, to be explained in the next section. If the
vehicle performance was found to be unsatisfactory, an
add-on anti-roll control (ARC) can be designed so that
differential braking will be applied when excessive
vehicle roll motion occurs. Because we do not have
access to the source code of the original VDC, the add-
on ARC was not designed in coordination with the yaw-
lateral design part. It was only tuned based on a couple
of selected maneuvers. Therefore, the revised VDC
does not represent a polished design but is adequate to
demonstrate the overall evaluation procedure. In this
paper, we focus on the evaluation of the roll
performance of VDC. In particular, the vehicle outputs
we watched closely include roll angle and tire normal
forces. When a rollover (roll angle > 10 degrees and
growing) does not occur, we examine tire normal forces.
Zero tire normal force on any tire indicates the
occurrence of wheel-lift-off, which is assumed to be an
unacceptable roll event.
3. EVALUAITON PROCEDURE
Figure 1 shows the flowchart of the proposed
iterative VDC evaluation procedure. When a VDC is
designed for a particular vehicle, it will first be tested
with open loop maneuvers in a standard test matrix. For
example, since NHTSA suggested J-Turn and Fishhook
maneuvers for dynamic rollover test, it is natural VDC
designers will include the NHTSA identified maneuvers
in their standard test matrix. In this paper, to simplify
the results, our standard test matrix only includes two J-
turns and two Fishhook maneuvers. A real test matrix
can be a lot more comprehensive, depending on the
preference of the company engineers. If the VDC
design performs satisfactorily, it can be allowed to
continue for the worst-case evaluation in Stage 2.
Vehicle
Design
Field Tests Open-loop Simulations
Standard
Test Matrix
Worst-case
Test Matrix
VDC
Design
HIL simulations
Closed-loop Simulations
Design modifications
Worst-case
Analysis
Driver Model

Figure 1 Iterative vehicle and VDC evaluation process

Worst-case maneuvers (Ungoren 2001) are
computed to identify the worst possible scenario for the
vehicle. Since the driver generated steering and braking
inputs are modeled as disturbance signals, the worst-
case maneuvers are obtained by solving an optimization
problem, to identify disturbances that maximize vehicle
roll motion in a pre-defined input range (steering angle
and brake force). This evaluation phase aims to ensure
the VDC works even when atypical inputs were
generated by the driver (e.g., panic or evasive
maneuvers). In general, this optimization problem is a
concave problem for a vehicle described in a numerical
format. Therefore, we have to solve the optimization
problem numerically. The mathematical core is based
on the Iterative Dynamic Programming (IDP) technique
(Luus 2000). The derivative free method with
convergence characteristics within the searched grid
points provides an excellent balance between
computation time and convergence.
The IDP technique used in this paper consists of two
phases: a dynamic programming phase and a strategy
phase. In the dynamic programming phase a standard
dynamic programming problem is solved over a sub-
domain of input and/or state space with sparse grid
points. Sparse grids have to be used because of the high
dimensionality (14DOF). In the strategy phase, the
solutions at the sparse grid points are approximated by
smooth curves, and a local optimum point is estimated.
The center, direction and the size of the new search-
domain for the next dynamic programming phase are
then selected by using the new local optimum point and
the most recent old optimum point (see Figure 2, where
the red points denote estimated optimum points). The
size of the search-domain reduces with the number of
iterations, until the convergence rate decreases to a
certain level. The domain size can then be increased to
avoid trapping at local optimum. Alternating between
strategy phases and dynamic programming phases
usually leads to improved solutions (Pierre 1986).


Figure 2 Curve-fitting and determination of search
domain (2-dimension example)

The next stage of the evaluation procedure is the
(virtual) driver-in-loop simulations. The VDC is tested
with steering profiles generated by a driver model,
tuned to represent three different driver characteristics.
This stage uses closed-loop simulations because the
road path to be followed is specified, and the driver
model closes the loop by using the sensed path and yaw
motion errors. This stage more closely resembles on-
road driving compared with the open-loop simulations
in the first two stages. This stage complements the first
two stages by including human-VDC interactions,
which is unfortunately lacking in the NHTSA proposed
dynamic tests.

After each set of simulations, the designer could
decide to go back to the drawing board and come back
with a better design, or continue with the next stage.
Once all simulation results are satisfactorily, the VDC
can be sent to the final stage--field testing with the
prototype controllers implemented on an actual vehicle.
The field test stage is beyond the scope of this paper
4. EVALUAITON RESULTS
A simulation exercise of the proposed procedure is
presented in this section. Two iterations are reported. In
the first iteration, the original VDC design is tested.
Based on the finding from this iteration, an anti-rollover
patch is designed to improve the performance of the
original VDC design. The new design is then tested in
the second iteration.

4.1 Iteration 1, Stage 1: Standard test matrix

For this stage we use four open loop maneuvers: J-
Turn, J-Turn with braking, Fishhook#1, and
Fishhook#2. These maneuvers are recommended by
NHTSA for dynamic rollover testing (Garrick 2002).
The vehicle is evaluated at three different initial
vehicle speeds (80, 100, and 120 kph). Hand-wheel
steering angle range and maximum braking force for the
simulations are set to [-80, 80] degrees and 25 lbf
respectively. These levels are somewhat smaller than
the NHTSA recommendations because we want to test
at higher vehicle speeds. Figure 3 shows the steering
angle, braking force and roll angle response both with
and without VDC. The roll angle is reduced slightly by
VDC, mainly because of the reduction in vehicle speed.
Figure 4 shows the steering angle and roll angle of the
Fishhook #1 maneuver. The vehicle without VDC rolls
over, while the vehicle with VDC has a maximum roll
angle of 5.83 degrees. The maximum roll angle values
obtained from these maneuvers for both VDC-on and
VDC-off cases are summarized in Table 1.
0 1 2 3 4 5
0
20
40
60
80
100
Time (sec)
H
a
n
d
w
h
e
e
l

S
t
e
e
r
i
n
g

A
n
g
l
e

(
d
e
g
)
0 1 2 3 4 5
0
5
10
15
20
25
Time (sec)
B
r
a
k
in
g

F
o
r
c
e

(
lb
f
)
0 1 2 3 4 5
0
1
2
3
4
5
Time (sec)
R
o
ll

A
n
g
l
e

(
d
e
g
)
VDC-off
VDC-on
100 kph

Figure 3 J-Turn with braking at 100 kphinputs and
roll response

As can be seen in Table 1, for all four maneuvers at
all speeds, the VDC helps to reduce the maximum roll
angle. We determine the performance of the controller
to be acceptable and continue on to the second step, the
worst-case analysis.
0 1 2 3 4 5
-40
-20
0
20
40
60
80
100
Time (sec)
H
a
n
d
w
h
e
e
l

S
t
e
e
r
in
g

A
n
g
le

(
d
e
g
)
Fishhook #1
0 1 2 3 4 5
0
5
10
15
20
Time (sec)
R
o
ll
A
n
g
le

(
d
e
g
)
VDC-off
VDC-on
120 kph

Figure 4 Fishhook #1inputs and roll angle (120 kph)

Table 1: Vehicle speed vs. maximum roll angle

Speed (km/hr) 80 100 120
VDC-off 3.04 4.85 rollover
J-Turn
VDC-on 2.99 4.57 5.97
VDC-off 3.18 4.95 rollover
J-Turn with
braking
VDC-on 3.05 4.59 5.89
VDC-off 2.95 4.70 rollover
Fishhook#1
VDC-on 2.95 4.53 5.83
VDC-off 2.92 4.60 rollover
Fishhook#2
VDC-on
max
roll
angle
(deg)
2.92 4.55 5.84

4.2 Iteration 1, Stage 2: Worst-case analysis

In this stage, the VDC is turned on all the time and
the worst-case performance for the vehicle plus VDC is
examined carefully. The purpose of a worst-case study
is to ensure that the active safety system works well in
the field, when panicked drivers, under a wide variety of
conditions, could produce unpredictable steering and
braking patterns. The worst-case analysis is thus
valuable for all active safety products of ground
vehicles. The worst-case analysis searches for the
worst-case inputs (steering angle and braking force)
within the same range identified in stage 1 (hand-wheel
steering angle [ 80,80] degrees and brake force
[0, 25] lbf). The initial speed can be set at the same
three levels as stage one (80,100 and 120 kph).
However, we decide to only focus on the highest speed
(120kph) because higher forward speed always
produces higher worst-case roll angle. The worst-case
study is extremely time-consuming and we need to
focus the limited resource on the most important case.
It should be noted that the IDP technique used in this
paper tries to perform search through a range that
shrinks and grows, and thus is able to escape from some
local optimum. However, due to the extremely large
state space (14 DOF, 28 states), we start from many
different initial conditions for the search. The results
are multiple worst-case scenarios, rather than a single
global optimum.
More than 120 input profiles, including the standard
maneuvers in stage 1 tests are used as initial guesses for
the worst-case maneuvers search. Figure 5 shows the
steering and braking inputs of one of the identified
worst-case scenarios on high friction surface, which
results in a rollover (Figure 6). For comparison
purposes, we use the same steering and braking inputs
for the vehicle with the VDC turned off. The vehicle
without VDC was found to have a maximum roll angle
of 4.27 degrees (Figure 6). This is quite alarming
because the maneuver does not pose a real threat for the
VDC-off case. But there is a good explanation for this:
this scenario resembles an evasive obstacle avoidance
steering (lane change) with heavy braking, which is not
that uncommon. It seems the VDC was trying to
achieve yaw rate following and in the process helps the
vehicle to become more oversteer. A good patch for this
problem is to add additional logic to ensure that the goal
of yaw-rate following will be overridden when the
vehicle is in danger of rollover. At this moment, we can
iterate back to the design stage. However, we choose to
continue to show a complete comparison of the two
designs.
0 1 2 3 4
-100
-50
0
50
100
Time (sec)
H
a
n
d
w
h
e
e
l
S
t
e
e
r
in
g

A
n
g
le

(
d
e
g
)

0 1 2 3 4
0
5
10
15
20
25
Time (sec)
B
r
a
k
i
n
g

F
o
r
c
e

(
lb
f
)

Figure 5 Worst-case hand-wheel steering and braking
force for VDC

0 1 2 3 4
-5
0
5
10
15
20
Time (sec)
R
o
ll
A
n
g
le

(
d
e
g
)
VDC-on
VDC-off
4.27 deg
rollover
0 1 2 3 4
0
2000
4000
6000
8000
10000
Time (sec)
N
o
r
m
a
l
F
o
r
c
e

(
N
)

VDC-off
VDC-on
Left Rear Tire

Figure 6 Roll angle and left rear tire normal force under
a worst-case scenario

4.3 Iteration 1, Stage 3: Human in loop

Designing an active safety system without
considering how drivers interact with it may result in
unsatisfactory performance. How the control authority
is shared between the driver and the controller is a
question that designers need to consider seriously.
Analyzing the interaction among vehicle-controller-
environment through human-in-the-loop (HIL)
simulations is critical to identify possible problems in
the design stage. In this section the driver model
proposed in (Ungoren 2004) is used to evaluate the
VDC system on a path following test. We use Toyotas
peak-to-peak yaw rate measure (Toyota 2001) to
investigate the driver and controller interactions. In the
Toyotas method, a star rating system is suggested,
which uses maximum entry speed to a Moose Test track
without hitting the obstacles, together with the peak-to-
peak yaw rate to determine a vehicles performance.
The obstacle avoidance, double-lane change test path is
shown in Figure 7.
The proposed driver model is tuned to fit three
different drivers: average, aggressive, and smooth
drivers. These drivers were obtained by tuning the
driver model parameters to fit the average, and average
plus/minus one standard deviation of hand-wheel
steering angle responses from 22 human drivers. The
path performance of driver 1 (aggressive: mean+),
driver 2 (average: mean) and driver 3 (smooth: mean-)
over the Moose test track for the VDC-off case are
shown in Figure 8. When the vehicle hits any orange
cone that delineates the un-allowed region, we declare
that the vehicle failed the test (see Figure 9). A driver
will be put through the HIL test starting from lower
initial speed. If it passed, the entrance speed will be
increased by 2kph and repeat the test, until it fails. We
found that driver 3 (smooth driver) fails at very low
speed. Obviously, to pass this demanding test, a driver
needs to respond and steer quickly.

0 10 20 30 40 50 60 70 80
-2
-1
0
1
2
3
4
5
x (m)
y
(
m
)

Figure 7 Moose Test track used in HIL tests

0 10 20 30 40 50 60 70 80 90 100
-2
-1
0
1
2
3
4
5
6
x (m)
y

(
m
)
y
d

y @ driver 1 (mean + )
y @ driver 2 (mean)
y @ driver 3 (mean - )
60 kph

Figure 8 Trajectory of the three drivers over the Moose
Test Track


Figure 9 Animation of test results: VDC-off vs. VDC-
on

Figure 10 shows the performance of the three drivers,
with and without VDC, on Toyotas suggested star-
ranking system. In Figure 10, moving in the direction
toward the lower right corner of the figure denotes
improvementeither the entry speed is increased, or the
peak-to-peak yaw rate is reduced. VDC is able to
improve the performance of both driver 1 (aggressive)
and driver 2 (average). It could not help driver 3
(smooth) because the driver is too slow and failed at
extremely low speeds. We are not too concerned about
this because driver 3 does not represent the nimble
response of a professional driver. The trajectories of the
vehicle with and without VDC at 63 kph are shown in
Figure 11. It can be seen that VDC helps driver 1 by
reducing the tail swing at the exit of the test track.

30 35 40 45 50 55 60 65 70
20
25
30
35
40
45
50
55
60
65
70
Max Entry speed at ISO 3888-2 course (kph)
Y
a
w

R
a
t
e

p
e
a
k
-
t
o
-
p
e
a
k

(
r
a
d
/
s
e
c
)
Star Rating Based on LAR and peak-to-peak yaw rate

VDC-off
VDC-on
VDC-off
VDC-on
1 Star
2 Star
3 Star
4 Star
5 Star
VDC-off
VDC-on
driver 1
driver 3
driver 2

Figure 10 Star rating based on Toyotas proposed
dynamic testing procedure.

0 20 40 60 80 100
-2
-1
0
1
2
3
4
5
6
x (m)
y

(
m
)
y
d

y @ VDC-off
y @ VDC-on
driver 1 - 63kph

Figure 11 Trajectory profiles for driver 1 on the test
track.

Even though the VDC helps to improve vehicle
performance in this HIL test, the VDC needs to be re-
designed due to its worst-case performancewhich
suggests that more can be done to improve its resistance
to rollover. This conclusion is not surprising since this
particular VDC was originally designed only for lateral
stability.


4.4 VDC Design Iteration

In this section an add-on anti-rollover patch is
designed to improve the vehicles roll stability under
worst-case scenarios. The patch plus the original VDC
system is then re-evaluated through another iteration. A
predictive approach is used to design the anti-rollover
patch. Future vehicle roll angle profile is predicted,
based on which proper control actions may be taken
before the vehicle reaches unsafe situations. Roll angles
are predicted using an adaptive discrete-time vehicle
model. If the model is accurate, it will be able to predict
the occurrence of an unacceptable event. For example,
a threshold vehicle roll angle can be chosen
corresponding to the occurrence of wheel liftoff. In
such a case at time t, a time-to-wheel lift-off (TTW) can
be calculated. If the input is held constant and the
system is predominantly linear, TTW counts down to 0
(i.e. wheel lift-off) with a slope of -1 with time (Chen
2001), thus is an excellent threat index.


time
TTW
Rollover threshold
Predicted response
Real time response
t t + TTW
Roll
angle
Current roll angle
1
2
3

Figure 12 Roll angle prediction and time-to-wheel life-
off (TTW)

The anti-rollover controller is designed to operate
on the difference between the predicted roll angle and
the rollover threshold. The prediction window (
p
t ) used
for the simulations is 0.2 seconds. A 4
th
order Recursive
Auto Regressive with eXtra input (RARX) model is
used for roll angle predictions, which can be represented
in the discrete time transfer function form as
(t) =
1
1
( )
( )
B z
A z

z
-d
(t-1) +
1
1
( ) A z

e(t)
(1)
where A(z
-1
) = 1 + a
1
z
-1
+ + a
na
z
-na
and B(z
-1
)= b
1
z
-d
+
+ b
nb
z
-nb-d
. is the predicted roll angle, is the
hand-wheel steering angle, d is the dead time of the
system, and e(t) is the exogenous input which is
assumed to be white. Parameters used are: d=1, nb=3,
na=4, and A(z
-1
) and B(z
-1
) are constructed using the
bicycle model. The sampling time of the RARX model
is 25ms, which achives reliable roll angle prediction.
The RARX model is trained on-line with steering and
roll signals from CARSIM.
Based on the predicted roll angle, the differential
braking torque (B
t
) may be applied either to the left
front ( <0) or right front ( >0) tire according to the
following logic:
| |
( ) ,
1 1
1 sgn( )
2 0.3 1
0
p
t o p d
t
If t t
B K K K
s
else B

+ >

( = + + +
`

+
)
=
(2)
where is the roll angle threshold, which is selected to
be high enough to prevent degradation of handling
performance but low enough to be effective in rollover
prevention. A first order lag with 0.3 sec time constant is
included to approximate the brake dynamics. The
controller gains are chosen to be
o
K = 300 Nm,
p
K =
1250 Nm/deg,
d
K = 4 Nm.sec/deg, and = 4 deg.
The structure of the anti-rollover patch and the original
VDC is shown in Figure 13.
This add-on unit plus the original VDC is assumed
to be the new VDC design, which will be put through
the evaluation process again in the second iteration. The
combined controller will be referred to as VDC-AR
(VDC with anti-roll) for the remainder of this paper. In
Figure 14 the roll responses under the VDC-on worst-
case maneuver (Figure 5) are given for the VDC-on and
VDC-AR-on cases. The improved roll performance of
VDC-AR is mainly because of its prediction capability.

+
+
Vehicle
Model
Anti-rollover controller
Differential
Braking
VDC
+
Predictor
(t+tp)
< 0
Roll Angle
Threshold

Controller
(Eq.(2))
(t)

|u|
Figure 13 Anti-rollover controller patch with the
original VDC

0 1 2 3 4
-5
0
5
10
15
20
Time (sec)
R
o
l
l

A
n
g
l
e

(
d
e
g
)
VDC-on
VDC-AR-on

Figure 14 Roll angle profiles: VDC vs. VDC-AR under
inputs shown in Figure 5

4.5 Iteration 2, Stage 1: Standard test matrix

Simulation results with the VDC-AR controller
under the standard test matrix are shown in Table 2. By
comparing Table 1 and Table 2, we can see that the
VDC-AR system improves vehicles maximum roll
angle under all standard tests.

Table 2 VDC-AR results under the standard test matrix

Initial speed (kph) 80 100 120

J-Turn
VDC-AR-on 2.99 4.21 4.26

J-Turn w/b
VDC-AR-on 3.05 4.03 4.49

Fishhook#1
VDC-AR-on 2.95 4.19 4.18

Fishhook#2
VDC-AR-on
max
roll
angle
(deg)
2.92 4.20 4.28

4.6 Iteration 2, Stage 2: Worst-case simulations

The steering angle and brake force range is the same
at [-80, 80] deg and [0,25] lbf, and the initial speed is
again set at 120 kph. Multiple initial guesses are chosen,
including the VDC-on worst-case maneuver. Out of all
the identified local optimum results under different
initial guesses, the one that results in highest roll angle
is shown below. The steering and braking time
trajectories look quite similar to a standard fishhook
maneuver with heavy braking coincides with the roll-
reversal point. Figure 16 shows comparison of no-
control, VDC-on and VDC-AR-on cases under the
fishhook worst-case maneuver identified for the
VDC-AR-on case. Due to the early activation of
differential braking, VDC-AR achieves lowest roll and
side slip and in the meantime achieves smallest turning
radius. Table 3 shows the maximum roll angle obtained
through simulations with the two worst-case scenarios:
for the VDC-on case and the VDC-AR-on case,
respectively. As can be seen, the worst-case scenario for
the VDC-on case results in a rollover, in contrast, worst-
case scenario for the VDC-AR-on case results in a
maximum roll angle of 4.72 degrees. Overall, the anti-
rollover patch seems to be working really well to
without compromising other vehicle performance.
0 1 2 3 4 5
-100
-50
0
50
100
Time (sec)
H
a
n
d
w
h
e
e
l

S
t
e
e
r
i
n
g

A
n
g
le

(
d
e
g
)

0 1 2 3 4 5
0
5
10
15
20
25
Time (sec)
B
r
a
k
i
n
g

F
o
r
c
e

(
l
b
f
)

Figure 15 Worst-case hand-wheel steering and braking
force for VDC-AR

0 1 2 3 4 5
-6
-4
-2
0
2
4
6
Time (sec)
R
o
ll
A
n
g
le

(
d
e
g
)
VDC-off
VDC-on
VDC-AR-on

0 1 2 3 4 5
-10
-5
0
5
10
15
Time (sec)
S
lip

A
n
g
le

(
d
e
g
)
VDC-off
VDC-on
VDC-AR-on

0 50 100 150
0
5
10
15
20
x (m)
y

(
m
)
VDC-off
VDC-on
VDC-AR-on

Figure 16 Vehicle response under worst-case VDC-AR-
on maneuver

Table 3 Maximum roll angle under worst-case
evaluation

Speed (kph)

optimized

120
VDC-off 4.27
VDC-on rollover VDC-on
VDC-AR-on 4.26
VDC-off 5.16
VDC-on 5.65 VDC-AR-on
VDC-AR-on
max
roll
angle
(deg)
4.72

4.7 Iteration 2, Stage 3: Human in loop

The Toyota peak-to-peak yaw rate based star rating
is again used. It is observed that for all three drivers the
anti-rollover patch was not activated. The reason is that
the predicted roll angle on this test course never exceeds
the treshhold value of 4 degrees. In other words, the
VDC-AR performs exactly like the original VDC in this
test. Since the VDC-AR works satisfactorily in all three
simulation tests, it can be forwarded to the field test
stage. That, of course, is beyond the scope of this paper.
5. CONCLUSION
In this paper a computer-based evaluation process is
outlined. As presented in Figure 1, the procedure
consists of three stages: standard test matrix, worst-case
test, and human (model)-in-loop Moose test. A Vehicle
Dynamic Control (VDC) system was evaluated through
this process. Both the standard test matrix and the
driver-in-loop simulations show that in general VDC
improves vehicles performance. However, we
identified a worst-case scenario with steering input that
looks like a lane change maneuver, which results in
rollover for the vehice with VDC. This needs to be
corrected because the same maneuver does not pose a
threat for a vehicle without VDC. An anti-rollover
patch using predicted vehicle roll motion and a switch
based on time-to-wheel-lift-off is designed to improve
the original VDC systems performace. It was shown that
for the new controller, VDC-AR successfully passed the
evaluation with improved roll performance in all tests.
The proposed evaluation procedure provide a
comprehensive evaluation for the VDC, and could
significantly reduce the field testing time by providing
valuable feedbacks through computer simulations .

ACKNOWLEDGEMENT This paper is based upon work
supported by the National Science Foundation under Grant No.
9734022, as well as a grant from TRW Automotive.
REFERENCES
National Highway Traffic Safety Administration (2002).
Traffic Safety Facts 2001.
National Highway Traffic Safety Administration (1998).
Traffic Safety Facts 1997.
National Highway Traffic Safety Administration (1973).
Advance Notice of Proposed Rulemaking (ANPRM)
on resistance to rollover, NHTSA 38 FR 959.
Zanten, A.V., Erhardt, R., Pfaff, G. (1995). VDC, The
Vehicle Dynamics Control System of Bosch, SAE
paper #950759, SAE Congress and Exposition.
Matsumoto, S., Yamaguchi, H., Inoue, H., Yasuno, Y.
(1992). Improvement of Vehicle Dynamics Through
Braking Force Distribution Control, SAE paper
#920645, SAE Congress and Exposition.
Wielenga, T.J. (1999). A Method for Reducing On-
Road Rollovers Anti-Rollover Braking. SAE Paper
#1999-01-0123, SAE Congress and Exposition.
Hac, A. (2002). Influence of Active Chassis Systems on
Vehicle Propensity to Maneuver-Induced Rollovers,
SAE Paper#2002-01-0967, SAE Congress and
Exposition.
Garrick, J. et al. (2002). A Comprehensive
Experimental Examination of Test Maneuvers That
May Induce On-Road, Untripped, Light Vehicle
Rollover-Phase IV of NHTSAs Light Vehicle
Rollover Research Program, DOT HS 809 513.
Ungoren, A.Y. and Peng, H. (2004) An Adaptive
Lateral Preview Driver Model, to appear in Vehicle
System Dynamics.
CARSIM User manual (2003). http://www.carsim.com/
Uogoren, A.Y., Peng, H. and Milot, D. (2001). Rollover
Propensity Evaluation of an SUV Equipped with a
TRW VSC System, SAE paper#2001-01-0128, , SAE
Congress and Exposition.
Luus, R. (2000). Iterative Dynamic Programming,
Chapman & Hall/Crc, Monographs and Surveys in
Pure and Applied Mathematics.
Pierre, D.A. (1986). Optimization theory with
applications, Dover Publications, New York.
Toyota (2001). Comments on Part 575 Consumer
Information Regulations, Rollover resistance
Dynamic rollover testing. Docket No. NHTSA-2001-
9663.
Chen, B. Peng, H. (2001). Differential-braking-based
rollover prevention for sport utility vehicles with
human-in-the-loop evaluations, Vehicle System
Dynamics, Vol.36, No.4-5, pp.359-389.