Linear State Space Modeling and Control Homework

École Supérieure d’Ingénieurs de Poitiers

Maı̂trise de l’Energie Electrique
Second Year

Guillaume Mercère
Email : [email protected]
1 Motivations 1

1 Motivations
The goal of this project is to test different state space based solutions for the lateral dy-
namics control of standard motor vehicles. Controlling the lateral vehicle motion is indeed
the keystone of recent smart software solutions when lane departure warning systems, lane
keeping systems or yaw stability control systems come into play [2]. Lane departures are
indeed “the number one cause of fatal accidents in the United States, and account for more
than 39% of crash-related fatalities” [4]. Developing effective and reliable feedback based
solutions for lane departure control is thus of prime interest for car manufacturers when
new smart and safe transportation answers are required by car users as well as decision
makers.
Before developing innovative information and communication technology tools for such
an objective, it is essential for you to understand how current controller based solutions
work at least in simulation, i.e., under ideal conditions. Off the shelf algorithms used for
trajectory control, braking assistance or avoidance maneuvers are all based on the use of
• specific state space representations for describing the vehicle motion,

• observers which enable to estimate non measurable vehicle dynamical states as well
as tire efforts and, by extension, tire/road friction coefficients online,

• dedicated state or output feedback ideas when the former two steps have been carried
out efficiently.
Implementing the state and output feedback solutions introduced in your lecture entitled
Linear State Space Modeling and Control for lateral dynamics control of passenger
cars seems thus somehow commonsensical.

2 Lateral vehicle dynamics description

In the literature [4, 2], several models have been suggested to describe the lateral
dynamics of passenger cars. In this project, in order to make simulations easier without
loosing dynamics modeling capabilities, we use a single track model, i.e., a model for which
• the left and right steering angles are assumed to be equal,

• the rear steering angle is assumed to be null,

• the pitch and roll effects are neglected.

As shown in Figure 1, such a vehicle model only has one front and one rear wheel, thus is
often called the bicycle model. In Figure 1,
• vx and vy are the longitudinalqand lateral vehicle speeds at the center of gravity
(CoG), respectively, while v = vx2 + vy2 stands for the total vehicle speed,

• ψ is the yaw angle of the vehicle in the global frame,

2 2 Lateral vehicle dynamics description

ψ
X
αf
fr vy v ff
β δ
αr CoG vx

ℓr ℓf

Figure 1: Single track model for lateral dynamics description.

• δ is the front steering angle,

• αr and αf are the rear and front wheel slip angles, respectively,
• β is the slip angle of the vehicle at the center of gravity, i.e., the direction of the
global car velocity,
• ℓr and ℓf are the longitudinal distances of the center of gravity to the rear and front
tires, respectively,
• fr and ff are the lateral tire forces on rear and front tires, respectively.
S By assuming that we focus herein on the lateral dynamics only, we can assume that the
longitudinal vehicle speed is constant, i.e.,
v̇x (t) = 0, ∀ t ≥ 0. (1)
Thanks to Newton’s law, we have
may (t) = fr (t) + ff (t) cos(δ(t)), (2)
where m stands for the vehicle mass while ay is the lateral vehicle acceleration measured at
the vehicle center of gravity. Clearly, because of the lateral car motion, the lateral vehicle
acceleration has a centrifugal component, i.e.,
ay (t) = v̇y (t) + vx (t)ψ̇(t), (3)
where v̇y is the acceleration due to the motion along the y axis while vx ψ̇ stands for the
centripetal acceleration. Similarly, moment balance around the z axis yields the equation
for the yaw dynamics, i.e.,
Iz ψ̈(t) = ℓf ff (t) − ℓr fr (t), (4)
2 Lateral vehicle dynamics description 3

where Iz is the yaw moment of inertia of the vehicle. As far as the slip angles are concerned,
standard trigonometric formulas lead to

vy (t)
tan(β(t)) = , (5)
vx (t)
!
vy (t) + ℓf ψ̇(t)
αf (t) = δ(t) − arctan , (6)
vx (t)
!
ℓr ψ̇(t) − vy (t)
αr (t) = arctan . (7)
vx (t)

Figure 2: Lateral tire forces fi vs. slip angles αi for different loads.

The last step of this analysis consists in modeling the lateral tire forces ff and fr which
act on the vehicle. As explained, e.g., in [3], many tire models have been introduced in the
literature. Most of them try to describe the link between the lateral tire forces fi and the
slip angles αi , i ∈ {f, r}, via curve fitting techniques, i.e., by suggesting physical models
or black box equations mimicking experimental data sets as those available in Figure 2-lhs.
As shown in Figure 2-rhs, fr (resp. ff ) can be approximated by a linear function of the
slip angle αr (resp. αf ) for small angle values. Thus, in this project, we could thus assume
that, for small slip angle values,

f f = c yf α f , (8)
f r = c yr α r , (9)

where cyi stands for the cornering stiffness, i ∈ {f, r}. These lateral force models are easy
to use, should match with real measurements when small angle values are considered but
are, by construction, static only. In order to take into account the fact that tire forces
are not developed instantaneously at maneuvering actions, but require a certain rolling
distance of the tire to build up due to the flexible structure of a tire, models involving
transient tire behaviors should be introduced, at least when moderate or fast vehicle speeds
4 3 Problem formulation and list of specifications

are considered. In [1], a first order model is suggested, i.e.,

τf f˙f (t) + ff (t) = cyf αf (t), (10)
τr f˙r (t) + fr (t) = cy αr (t),
r (11)
where τi stands for a relaxation time constant, i ∈ {f, r}. This time constant can be
approximated as follows
c yi σα
τi = = i , i ∈ {f, r}, (12)
ki v x vx
where ki , i ∈ {f, r}, stands for an equivalent tire lateral stiffness while σαi , i ∈ {f, r}, is a
relaxation length which approximates the distance needed to build up tire forces.

3 Problem formulation and list of specifications

Symbol Definition Units Nominal Values

vx vehicle longitunal speed m/s 30
m vehicle mass Kg 1500
c yf front axle cornering stiffness N/rad 1 × 105
c yr rear axle cornering stiffness N/rad 1.5 × 105
ℓf front axle distance to the CoG m 1.3
ℓr rear axle distance to the CoG m 1.7
Iz inertia about the z axis Kg/m2 3000
n steering ratio - 15
σ αf front axle relaxation length m 1.2
σ αr rear axle relaxation length m 0.3

Table 1: Bicycle model parameters.

In this project, the main goal is to control the yaw velocity ψ̇(t) with a state and/or
output feedback controller based approach. In order to reach this goal, it is essential
• to determine a reliable state space representation of the vehicle dynamics,
• to introduce a list of specifications for yaw velocity control,
• to determine to controller gains and/or observer gains tuned to satisfy the aforemen-
tioned list of specifications.
These solutions can be tested by using dedicated software like MATLAB or OCTAVE.

4 Modeling, state and output feedback

Question 1: By using the bicycle model, i.e., the equations gathered in the former
section, and by assuming that the angles β, δ, αr and αf are small, determine the linear
4 Modeling, state and output feedback 5

time invariant state space representation describing the dynamical behavior of a passenger
car when the state vector is selected as follows
 
β(t)
 ψ̇(t) 
x(t) = ff (t) ,
 (13)
fr (t)

the output of the model is chosen equal to the yaw velocity ψ̇(t) while the input sig-
nal is δ(t). More specifically, show that the following linear time invariant state space
representation

ẋ(t) = Ax(t) + Bu(t), (14a)

y(t) = Cx(t), (14b)

with
1 1
 
0 −1 mvx mvx
ℓf
0 0 − Iℓzz
r
 
 Izz 
A= −cy vx ℓ , (15a)
f σα
−cyf σαf − σvαx 0 
 f f f 
−cyr σvαx cyr σℓαr 0 − σvαx
 r r r

0
 0 
B =  vx cyf  , (15b)
 
 nσαf 
0
 
C= 0 1 0 0 , (15c)

can be generated from the equations gathered in the former Section. Notice the presence
of a parameter n in B standing for a reduction ratio due to the direction gear.

Question 2: Based on the file LatDynSimulation.m, generate either with MATLAB-

SIMULINK or directly with lsim on MATLAB-OCTAVE a simulator of this state space
representation by using the physical parameter values gathered in Table 1. Test your sim-
ulator by feeding it with a chirp steering wheel angle, i.e., a signal in which the frequency
increases (down-chirp) with time. The experiment duration should be equal to 40 s. The
sampling period is chosen equal to 1 ms. The minimum and maximum frequencies of this
chirp signal are equal to 0 Hz and 5 Hz, respectively. The steering wheel angle should
vary between 10 and −10 degrees (do not forget to convert these angles into rad/s). As
far as the initial state vector is concerned, you are asked to use
 
0.01
0.03
 200  .
x(0) =   (16)
300
6 4 Modeling, state and output feedback

Once a model of the system to control is available and reliable, the next step of the
project consists in controlling the yaw velocity by assuming first that
• the state variables are measurable,

• the desired yaw rate ψ̇des (t) is constant and determined by road curvature and vehicle
speed.
More specifically, we assume that
vx
ψ̇des = , (17)
R
where R stands for the curve radius.

Question 3: Based on the file LatDynStateFeedback.m, by using the Bass-Gura pro-

cedure, determine the state feedback controller gain K ∈ R1×4 as well as the tracking
controller gain kr ∈ R, respectively, which guarantee a good tracking of the desired yaw
rate ψ̇des (t). As far as the list of specifications is concerned, you must
• determine the poles of the open loop system,

• choose closed loop eigenvalues ensuring a fast closed loop dynamics,

• observe the steering angle and place the closed loop eigenvalues so that the steering
angle is kept small enough (lower than or equal to 1 rad) to guarantee that the
approximations made to generate a linear time invariant state space representation
of the vehicle dynamics are still valid,

• plot the time evolution of the output y and compare it with the desired yaw rate.
As far as the numerical values are concerned, we assume that vx = 10 m/s while R =
50 m. The desired yaw rate shape should be zero during 20 s then vRx for the rest of the
simulation.

Question 4: Test the efficiency of the tracking controller gain solution by assuming, e.g.,
that the real car speed is not 10 m/s but 10.1 m/s. Explain the reason why the tracking
control gain solution is not efficient for this specific case.

In order to solve the problem encountered with the tracking controller gain solution,
introducing a integral action into the controller law is suggested. More specifically, the
new control law structure is selected as follows
Z
u(t) = −Kx(t) − ki (r(t) − y(t))dt. (18)

Question 5: Based again on the file LatDynStateFeedback.m, by adapting the Bass-Gura

procedure, determine the state feedback controller gain K ∈ R1×4 as well as the integrator
REFERENCES 7

gain ki ∈ R, respectively, which guarantee a good tracking of the desired yaw rate ψ̇des (t).
Test the performance of this controller by considering the former list of specifications.

The former solutions all assume the availability of the state variables. On most of the
standard passenger cars, this assumption is not satisfied because sensors to measure all the
components of x are not available in practice. It is thus necessary to resort to an observer
in order to reconstruct the unknown state variable dynamics from available measurements.

Question 6: Based on the file LatDynObserver.m, by using the Bass-Gura procedure,

design an observer able to reconstruct the four signals composing the state vector x. Test
your solution by reconstructing the components of x knowing y by using as the input
signal the chirp signal generated in Question 2. Plot, in the same figures, each component
of x and x̂ in order to illustrate the efficiency of your procedure.

The last step of the procedure consists in combining the state observer designed pre-
viously with the state feedback controller used in Question 5.

Question 7: After explaining with your own words the principle of the output feedback
control, generate a file called LatDynOutputFeedback.m which implements this output
feedback control for tracking the desired yaw velocity when the state variables are not
measured but reconstructed with an observer.

References
[1] M. Doumiati, A. Charara, A. Victorino, and D. Lechner. Vehicle dynamics estimation
using Kalman Filtering. Wiley, 2013.

[2] M. Guiggiani. The science of vehicle dynamics. Springer, 2014.

[3] S. Khaleghian, A. Emami, and S. Taheri. A technical survey on tire-road friction

estimation. Friction, 5:123–146, 2017.

[4] R. Rajamani. Vehicle dynamics and control. Springer, 2012.