Vehicle System Dynamics

ISSN: 0042-3114 (Print) 1744-5159 (Online) Journal homepage: http://www.tandfonline.com/loi/nvsd20

MPC-based yaw and lateral stabilisation via active

front steering and braking

Paolo Falcone , H. Eric Tseng , Francesco Borrelli , Jahan Asgari & Davor
Hrovat

To cite this article: Paolo Falcone , H. Eric Tseng , Francesco Borrelli , Jahan Asgari & Davor
Hrovat (2008) MPC-based yaw and lateral stabilisation via active front steering and braking, Vehicle
System Dynamics, 46:S1, 611-628, DOI: 10.1080/00423110802018297

To link to this article: https://doi.org/10.1080/00423110802018297

Published online: 28 Jan 2009.

Submit your article to this journal

Article views: 1879

Citing articles: 93 View citing articles

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=nvsd20
Vehicle System Dynamics
Vol. 46, Supplement, 2008, 611–628

MPC-based yaw and lateral stabilisation via active front

steering and braking
Paolo Falconea *, H. Eric Tsengb , Francesco Borrellic , Jahan Asgarib and Davor Hrovatb
a Department of Signals and Systems, Chalmers University of Technology, Göteborg, Sweden; b Ford
Research Laboratories, Dearborn, MI, USA; c Department of Mechanical Engineering, University of
California, Berkeley, USA

(Received 1 August 2007; final version received 27 February 2008 )

In this paper, we propose a path following Model Predictive Control-based (MPC) scheme utilising
steering and braking. The control objective is to track a desired path for obstacle avoidance manoeuvre,
by a combined use of braking and steering. The proposed control scheme relies on the Nonlinear
MPC (NMPC) formulation we used in [F. Borrelli, et al., MPC-based approach to active steering
for autonomous vehicle systems, Int. J. Veh. Autonomous Syst. 3(2/3/4) (2005), pp. 265–291.] and
[P. Falcone, et al., Predictive active steering control for autonomous vehicle systems, IEEE Trans.
Control Syst. Technol. 15(3) (2007), pp. 566–580.]. In this work, the NMPC formulation will be used
in order to derive two different approaches. The first relies on a full tenth-order vehicle model and has
high computational burden. The second approach is based on a simplified bicycle model and has a
lower computational complexity compared to the first. The effectiveness of the proposed approaches
is demonstrated through simulations and experiments.

Keywords: vehicle dynamics control; model predictive control; autonomous vehicles

Introduction

Dynamic control systems have been introduced to the market for improved vehicle handling
and safety. Among them, yaw stability controller using the four wheel braking, known as
Electronic Stability Control, has demonstrated its effectiveness [1]. In addition to yaw stability
control, braking and traction systems, Active Front Steering (AFS) systems make use of front
steering to further enhance lateral and yaw vehicle stability [2–4]. Moreover, the steering
command can be used to reject external destabilising forces arising from mu-split, asymmetric
braking or wind [5]. For instance, in [6] the authors present a decoupling strategy between
the path following and external disturbances rejection in a four wheels steering setup. The
automatic car steering is split into the path following and the yaw stabilisation tasks; the first
is achieved through the front steering angle, the latter through the rear steering angle.
Current research efforts in the field of vehicle dynamics control aim to integrate the braking
and the steering systems to improve the yaw and the lateral stability. In this paper, we focus

*Corresponding author. Email: [email protected]

ISSN 0042-3114 print/ISSN 1744-5159 online

© 2008 Taylor & Francis
DOI: 10.1080/00423110802018297
http://www.informaworld.com
612 P. Falcone et al.

Figure 1. Model predictive control scheme including the braking logic.

on an obstacle-avoidance, double-lane-change manoeuvre [7] and propose MPC-based con-

trollers to optimise the combination of the steering and braking in performing the specified
manoeuvre. The proposed control schemes relies on the MPC formulation we used in [8].
Two different MPC approaches, computing the optimal steering/braking combinations, are
investigated in this paper and compared. The first utilises a full four wheels vehicle model
where individual wheel dynamics are included and the front steering and the individual wheel
brake torques are the control inputs. The second approach relies on a simplified bicycle model
[9] where the control inputs are the braking yaw moment, i.e. yaw moment generated from
differential braking and the front steering angle. The braking yaw moment is generated for yaw
stabilisation while the front steering is for controlling the lateral dynamics. To generate yaw
moment by braking, with minimum effect on the longitudinal or the lateral dynamics, single
wheel braking logic is applied. Starting from the desired braking yaw moment the braking
torques at each wheel is then generated through a braking logic as shown in Figure 1.
In the first MPC approach, the interaction between longitudinal braking/slip ratio and lateral
force/slip angle is fully captured by the vehicle model. However, as a result of model com-
plexity, the tuning process could be quite involved. In the second approach, the interaction
between braking and steering are only optimised at the vehicle level but not at the tyre level.
For example, the effect of future longitudinal slip on lateral force is neglected. By formulating
an easier MPC problem a more intuitive tuning process and an easier interpretation of derived
control actions are facilitated.
The contributions of this paper are (i) the formulation of two combined steering and braking
MPC approaches, (ii) the comparison of the proposed approaches through simulations and (iii)
the presentation of path following via combined steering and braking experimental results. In
particular, we present simulation results of the two MPC approaches showing the effects of
simplifications, introduced in the vehicle modelling, on the achievable performances. More-
over, a first-cut of experimental testing is also showed to demonstrate the real-time feasibility
of this study.
The paper is organised as follows: in Section 1 we present the vehicle models used in the
two MPC approaches formulated in Section 2. In Section 3 we present simulation results of
the two approaches and experimental results obtained with a combined braking and steering
MPC controller implemented in real-time. Finally, Section 4 closes the paper with remarks
and ideas for future works.

1. Modelling

This section describes the vehicle and tyre models we used for simulations and control design.
The nomenclature refers to the four wheels and the bicycle models depicted in Figures 2 and 4,
 Vehicle System Dynamics 613

Figure 2. The simplified vehicle dynamical model.

Figure 3. Tyre force characteristics.

respectively. We denote by Fl , Fc the longitudinal (or ‘tractive’) and lateral (or ‘cornering’)
tyre forces, respectively, Fx , Fy the components of the tyre forces along the longitudinal and
lateral vehicle axes, respectively, Fz the normal tyre load, M the braking yaw moment, X, Y the
absolute car position inertial coordinates, a, b (distance of front and rear wheels from centre
of gravity) and c (distance of the vehicle longitudinal axis from the wheels) the car geometry,
g the gravitational constant, m the car mass, I the car inertia, Id the driveline inertia, Bd the
damping coefficient of the driveline, r the wheel radius, s the slip ratio, μ the road friction
coefficient, vl , vc the longitudinal and lateral wheel velocities and vx , vy their resultants along
the longitudinal and lateral vehicle axes respectively, ψ the heading angle, ẋ and ẏ the vehicle
longitudinal and lateral speeds, respectively, α the tyre slip angle, δ the wheel steering angle,
Tb the brake torque at the brake pad.
614 P. Falcone et al.

Figure 4. The simplified vehicle ‘bicycle model’.

In the four wheels model, we will use two subscript symbols to denote variables related to
the four wheels. In particular the first subscript denotes the front and rear axles, the second
denotes the left and right sides of the vehicle. As example, the variable (·)f,l is referred to the
front left wheel.
In the bicycle model, we will use a subscript symbol to denote variables related to the two
wheels. In particular, the subscripts (·)f and (·)r particularise a variable at the front wheels
and the rear wheels, respectively, e.g. Flf is the front wheel longitudinal force.

1.1. Four wheels full vehicle model

Figure 2 depicts a diagram of the four wheels vehicle model, which has the following
longitudinal, lateral and turning or yaw degrees of freedom (DOF)

mÿ = −mẋ ψ̇ + Fyf,l + Fyf,r + Fyr,l + Fyr,r , (1a)

mẍ = mẏ ψ̇ + Fxf,l + Fxf,r + Fxr,l + Fxr,r , (1b)
I ψ̈ = a(Fyf,l + Fyf,r ) − b(Fyr,l + Fyr,r )
+ c(−Fxf,l + Fxf,r − Fxr,l + Fxr,r ). (1c)

The vehicle’s equations of motion in an absolute inertial frame are

Ẏ = ẋ sin ψ + ẏ cos ψ, (2a)

Ẋ = ẋ cos ψ − ẏ sin ψ. (2b)

The longitudinal and lateral forces generated by the four tyres lead to the following components
along the lateral and longitudinal vehicle axes:

Fy = Fl sin δ + Fc cos δ, (3a)

Fx = Fl cos δ − Fc sin δ. (3b)
 Vehicle System Dynamics 615

Tyre forces for each tyre are given by

Fl = fl (α, s, μ, Fz ), (4a)
Fc = fc (α, s, μ, Fz ), (4b)

where fl and fc are complex nonlinear functions described in details in Section 1.2.
In Equation (4), α is the slip angle of the tyre and s is the slip ratio defined as
⎧ rw
⎨ − 1 if νl > rw, ν  = 0 for braking
s = vl ν l (5)
⎩1 − if νl < rw, w  = 0 for driving
rw
where vl is the longitudinal wheel velocity, r and w are the wheel radius and angular speed
respectively. By assuming that the vehicle is travelling at zero engine torque, the wheels
angular speeds are obtained, for the four wheels, by integrating the following set of differential
equations

Id,f ẇf l = −Flf,l r − Tbf,l − Bd wf,l , (6a)

Id,f ẇf r = −Flf,r r − Tbf,r − Bd wf,r , (6b)
Id,r ẇrl = −Flr,l r − Tbr,l − Bd wr,l , (6c)
Id,r ẇrr = −Flr,r r − Tbr,r − Bd wr,r , (6d)

The slip angle represents the angle between the wheel velocity and the direction of the wheel
itself:
νc
α = tan−1 . (7)
νl
In Equations (5) and (7), vc and vl are the lateral (or cornering) and longitudinal wheel
velocities, respectively, which are expressed as

νl = νy sin δ + νx cos δ, (8a)

νc = νy cos δ − νx sin δ, (8b)

and

νyf,l = ẏ + a ψ̇ νxf,l = ẋ − cψ̇, (9a)

νyf,r = ẏ + a ψ̇ νxf,r = ẋ + cψ̇, (9b)
νyr,l = ẏ − bψ̇ νxr,l = ẋ − cψ̇, (9c)
νyr,r = ẏ − bψ̇ νxr,r = ẋ + cψ̇, (9d)

The parameter μ in Equation (4) represents the road friction coefficient and Fz is the vertical
load of the vehicle. Next we make use of the following:

ASSUMPTION 1 We assume constant normal tyre load, i.e., Fzf,l , Fzf,r , Fzr,l , Fzr,r = constant.
This is distributed between the front and rear wheels based on the geometry of the car model,
described by the parameters a and b:
bmg amg
Fzf,l = Fzf,r = , Fzr,l = Fzr,r = . (10)
2(a + b) 2(a + b)
616 P. Falcone et al.

Using the Equations (1)–(10), the nonlinear vehicle dynamics can be described by the
following compact differential equation:

ξ̇ (t) = fμ(t)
4w
(ξ(t), u(t)), (11)

where the state and input vectors are ξ = [ẏ, ẋ, ψ, ψ̇, Y, X, wf,l , wf,r , wr,l , wr,r ] and u =
[δf , Tbf,l , Tbf,r , Tbr,l , Tbr,r ], respectively, and μ(t) = [μf,l (t), μf,r (t), μf,r (t), μr,r (t)].

1.2. Tyre Model

The model for tyre tractive and cornering forces (4) used in this paper is described by a
Pacejka model [10]. This is a complex, semi-empirical model that takes into consideration
the interaction between the tractive force and the cornering force in combined braking and
steering. The longitudinal and cornering forces are assumed to depend on the normal force,
slip angle, surface friction, and longitudinal slip. Sample plots of predicted longitudinal and
lateral force versus longitudinal slip and slip angle, in pure braking/driving and cornering, are
shown in Figure 3a. Lateral tyre force characteristics, in combined driving and cornering, are
shown in Figure 3b. These plots are shown for a single tyre.

1.3. Simplified bicyle model

In this section we derive a reduced order model, for the MPC optimisation process, from the
four wheels model (11). It is called single track or ‘bicycle model’ [9] and it is depicted in
Figure 4. It is based on the following set of simplifications:
Simplification 1: At front and rear axles, the left and right wheels are lumped in a single wheel.
Simplification 2: Pure yaw moment can be introduced through braking application (i.e. the
additional impact on longitudinal and lateral forces is negligible).
Simplification 3: The lateral force is calculated based on zero tyre slip ratio, i.e., Fc =
fc (α, 0, μ, Fz ).
Simplification 4: The steering and braking effects on vehicle speed are negligible.

Remark 1 Based on typical driving preference and evaluation, braking is generally used for
maintaining yaw stability purpose only. Since a practical cost function (for MPC optimisa-
tion) should reflect that, minimum and single-sided usage of brakes is expected. Therefore,
Simplifications 1–4 are deemed reasonable.

With the Simplifications 1–4, we have:

ẍ  0, (12a)
2Fyf  2Fcf |s=0 cos δf , (12b)
2Fyr  2Fcr |s=0 . (12c)

According to (12), the Equation (1) for the simplified bicycle model can be re-written as
follows

mÿ = −mẋ ψ̇ + 2Fyf + 2Fyr , (13a)

ẍ = 0, (13b)
I ψ̈ = 2aFyf − 2bFyr + M, (13c)
 Vehicle System Dynamics 617

where M is the braking yaw moment that, in the four wheels model (1), is computed as follows
M = c(−Fxf,l + Fxf,r − Fxr,l + Fxr,r ). (14)
The forces in Equation (13) can be computed through the Equations (3)–(5) and (7)–(10),
particularised for the front and rear axles. In particular the Equations (9) can be rewritten as
follows:
νyf = ẏ + a ψ̇ νyr = ẏ − bψ̇, (15a)
νxf = ẋ νxr = ẋ. (15b)

Remark 2 We remark that the forces Fyf and Fyr in Equation (13) and in Figure 4 represent
the lateral components of the cornering tyre forces Fc generated by the contact of a single
wheel with the ground.

The nonlinear vehicle dynamics described by the Equations (2)–(5), (7) and (8), (10),
(13)–(15) can be rewritten in the following compact from:
ξ̇ (t) = fμ(t)
2w
(ξ(t), u(t)), (16)

where μ(t) = [μf (t), μr (t)]. The state and input vectors are ξ = [ẏ, ẋ, ψ, ψ̇, Y, X] and
u = [δf , M] respectively.

2. Model predictive control problem

Next we design two MPC controllers computing the front steering angle and the braking
torques at the four wheels, such that a desired path is followed as close as possible at a
given longitudinal speed. In Sections 2.1 and 2.2 two controllers, based on the vehicle models
presented in Sections 1.1 and 1.3, respectively, are presented.

2.1. MPC controller based on the full vehicle model (11)

In order to obtain a finite dimensional optimal control problem, we discretise the model (11)
with the Euler method, to obtain
4w,dt
ξ(t + 1) = fμ(t) (ξ(t), u(t)), (17a)
u(t) = u(t − 1) + u(t), (17b)
η(t) = h4w (ξ(t)), (17c)
where u(t) = [δf , Tbf,l , Tbf,r , Tbr,l , Tbr,r ], u(k) = [δf (k), Tbf,l , Tbf,r , Tbr,l , Tbr,r ] and
⎛ ⎞
0 1 0 0 0 0 0 0 0 0
⎜0 0 1 0 0 0 0 0 0 0 ⎟
h4w (ξ(t)) = ⎜ ⎝0 0 0 1 0 0 0 0 0 0 ⎠ .
⎟ (18)
0 0 0 0 1 0 0 0 0 0
We consider the following cost function:


Hp
 c −1
  H

J (ξ(t), Ut ) = η̂t+i,t − ηref 2 + ut+i,t 2R + ut+i,t 2S , (19)
t+i,t Q
i=1 i=0
618 P. Falcone et al.

where, as in standard MPC notation ([11]), ut = [ut,t , . . . , ut+Hc −1,t ] is the optimisation
vector at time t and η̂t+i,t denotes the output vector predicted at time t + i obtained by starting
from the state ξt,t = ξ(t) and applying to system (17) the input sequence ut,t , . . . , ut+i−1,t .
Hp and Hc denote the output prediction horizon and the control horizon, respectively. As in
standard MPC schemes, we use Hp > Hc and the control signal is assumed constant for all
Hc ≤ i ≤ Hp , i.e., ut+i,t = 0 ∀i ≥ Hc . The reference signal ηref = [ẋref , ψref , ψ̇ref , Yref ]
represents the desired outputs. Q, R and S are weighting matrices of appropriate dimensions.
In (19) the first summand reflects the desired performance on target tracking, the second and
the third summands weight the steering and braking effort.
At each time step we solve the following optimisation problem:

min J (ξt , Ut ) (20a)

Ut

subj. to
ξk+1,t = fμ4w,dt
k,t
(ξk,t , uk,t ), k = t, . . . , t + Hp − 1 (20b)
uk,t = uk−1,t + uk,t , k = t, . . . , t + Hc − 1 (20c)
ηk,t = h4w (ξk,t ), k = t + 1, . . . , t + Hp (20d)
ηt,t = ξ(t), (20e)
ut−1,t = u(t − 1), (20f)
μt,t = μ(t), (20g)
uf,min ≤ uk,t ≤ uf,max , k = t, . . . , t + Hp − 1 (20h)
uf,min ≤ uk,t ≤ uf,max . k = t, . . . , t + Hc − 1 (20i)

where the Equations (20b)–(20g) are the vehicle dynamics, the constraints (20h) limit the
steering angle and the braking torques and the constraints (20i) limits the steering angle
and the braking torques variations between two consecutive time steps. We denote by
Ut∗  [u∗t,t , . . . , ∗t+Hc −1,t ] the sequence of optimal steering and braking torques incre-
ments computed at time t by solving the problem (20) for the current observed states ξ(t)
and the previous input u(t − 1). Then, the first sample of Ut∗ is used to compute the optimal
steering and braking torques and the resulting state feedback control law is

u(t) = u(t − 1) + u∗t,t , (21)

where u∗t,t is the vector of the first five elements of Ut∗ .

Remark 3 The optimisation problem (20) has 5Hc optimisation variables, 10Hp nonlinear
constraints and 5(Hc + Hp ) linear constraints.

2.2. MPC controller based on the simplified bicycle model (16)

We consider a discrete time version the model (16)

2w,dt
ξ(t + 1) = fμ(t) , (ξ(t), u(t)), (22a)
u(t) = u(t − 1) + u(t), (22b)
η(t) = h2w (ξ(t)), (22c)
 Vehicle System Dynamics 619

where u(t) = [δf , M], u(t) = [δf (t), M] and

⎛ ⎞
0 0 1 0 0 0
h2w (ξ(t)) = ⎝0 0 0 1 0 0⎠ . (23)
0 0 0 0 1 0

Moreover, we consider the cost function (19), with ηref = [ψref , ψ̇ref , Yref ] , and at each time
step we solve in receding horizon the following optimisation problem:

min J (ξt , Ut ) (24a)

Ut

subj. to
ξk+1,t = fμ2w,dt
k,t
(ξk,t , uk,t ), k = t, . . . , t + Hp − 1 (24b)
uk,t = uk−1,t + uk,t , k = t, . . . , t + Hc − 1 (24c)
ηk,t = h2w (ξk,t ), k = t + 1, . . . , t + Hp (24d)
ξt,t = ξ(t), (24e)
ut−1,t = u(t − 1), (24f)
μt,t = μ(t), (24g)
uf,min ≤ uk,t ≤ uf,max , k = t, . . . , t + Hp − 1 (24h)
uf,min ≤ uk,t ≤ uf,max . k = t, . . . , t + Hc − 1 (24i)

Remark 4 The optimisation problem (24) has 2Hc optimisation variables, 6Hp nonlinear
constraints and 2(Hc + Hp ) linear constraints.

Remark 5 We recall that, due to the Simplifications 2 and 4, the simplified bicycle model (22)
does not model the longitudinal dynamics of the vehicle. Therefore, compared to controller
(20) and (21), the longitudinal vehicle velocity cannot be a tracking variable in controller
(21)–(24).

Once the optimal value of the braking yaw moment M has been computed through the
(21)–(24), braking torques at the four wheels are computed through Algorithm 1. In particular,
in order to induce the desired braking yaw moment with minimum longitudinal dynamics
effect, Algorithm 1 implements a single wheel braking logic. The braking logic in Algorithm 1
is based on the following well-known results:
• Outside wheel braking induces understeer while inside wheel braking induces oversteer.
• Left/right brake distribution is more effective in steering the vehicle than front/rear
distribution [12].
• Braking at the rear inside corner is most effective in inducing an oversteer yaw moment, and
braking at the front outside corner is most effective to induce an understeer yaw moment
[13–15].

3. Simulation and experimental results

We considered a scenario where the objective is to follow a desired path as close as possible on
a snow covered road (μ = 0.3) at a given desired speed. The control inputs are the front tyre
steering angle and the brake torques at the four wheels and the goal is to follow the trajectory
620 P. Falcone et al.

as close as possible by minimising the vehicle deviation from the target path. The experiment
is repeated with increasing entry speeds until the vehicle loses control.
Next the MPC controllers presented in Sections 2.1 and 2.2 are referred to as Controller A
and Controller B, respectively, and are defined as follows:

• Controller A. Controller (20), (21) presented in Section 2.1 with the following parameters:
sampling time: T = 0 : 05 s.
horizons: Hp = 20; Hc = 2.
bounds:
∗ δf,min = −10 deg. δf,max = 10 deg, δf,min = −0.85 deg, δf,max = 0.85 deg.
∗ Tb∗,•min = 0 Nm, Tb∗,•max = 600 Nm, Tb∗,•min = −58.33 Nm, Tb∗,•max = 58.33 Nm, with
∗ ∈ {f, r} and • ∈ {l, r}.
friction coefficient: μ = 0.3.
weighting matrices:
∗ Q ∈ R4×4 with Q11 = 10, Q22 = 12, Q33 = 1, Q44 = 57 and Qij = 0 for i  = j .
∗ R ∈ R5×5 for R11 = 30, R22 = 102 , R33 = 102 , R44 = 102 , R55 = 102 and Rij = 0 for
i = j .
∗ S ∈ R5×5 with S11 = 10−3 , S22 = 20, S33 = 20, S44 = 10, S55 = 20 and Sij = 0
for i  = j .
• Controller B. Controller (24), (21) with braking logic in Algorithm 1, presented in
Section 2.2 with the following parameters:
sampling time: T = 0.05s.
horizons: Hp = 15; Hc = 2.
bounds:
∗ δf,min = −10 deg, δf,max = 10 deg δf,min = −0.85 deg, δf,max = 0.85 deg.
∗ Mmin = −103 Nm, Mmax = 103 Nm, Mmin = −50 Nm, Mmax = 50 Nm.
friction coefficient: μ = 0.3.
weighting matrices:
∗ Q ∈ R3×3 with Q11 = 10, Q22 = 1, Q33 = 100 and Qij = 0 for i  = j .
∗ R ∈ R2×2 for R11 = 1, R22 = 102 and Rij = 0 for i  = j .
∗ S ∈ R2×2 with S11 = 1, S22 = 1 and Sij = 0 for i  = j .

In Sections 3.1 and 3.2, we present the simulation results of Controller A and Controller B.
In particular, we compare Controller A and Controller B and show the performance limitations
in Controller B due to the Simplifications 1–4 introduced in the vehicle modelling.
In Section 3.3 we show experimental results of a low complexity path following MPC
algorithm coordinating steering and braking. This is obtained from Controller A by approx-
imating the nonlinear model (17) with linear models computed through successive online
linearisations along the desired trajectory. Further details can be found in [16].

ALGORITHM 1 Braking torques calculation

Input: M, αf , αr .
Output: Tbf,l , Tbf,r , Tbr,r , Tbr,r .
begin
if M = 0 then
Tbf,l = 0, Tbf,r = 0, Tbr,l = 0, Tbr,r = 0;
else if M > 0 then /* braking the left side */
if αf − αr > 0 then /* understeering */
Tbf,l = 0, Tbf,r = 0, Tbr,l = sin(arctan|M|r √
c/b) b2 +c2
, Tb r,r
= 0;
else /* oversteering */
 Vehicle System Dynamics 621

|M|r √
Tbf,l = sin(arctan c/a−δf ) a 2 +c2
, Tbr,l = 0, Tbr,r = 0, Tbr,r = 0;
end
else /* braking the right side */
if αf − αr > 0 then /* understeering */
Tbf,l = 0, Tbf,r = 0, Tbr,l = 0, Tbr,r = sin(arctan|M|r √
c/b) b2 +c2
;
else /* oversteering */
|M|r √
Tbf,l = 0, Tbf,r = sin arctan c/a−δ , Tb = 0, Tb = 0;
( f ) a +c
2 2 r,l r,r

end
end

3.1. Controller A. Simulation results

Next we present the simulation results of Controller A at 14 and 19.5 m/s (50 and 70 Kph,
respectively). In Figures 5–7 the simulation results at 50 Kph are reported, while Figures 8–10
show the simulation results at 70 Kph.
RMS and maximum tracking errors are summarised in Table 1. Notice that only the max-
imum deviation of the longitudinal vehicle speed from the reference value is reported while
the RMS error is not. We chose to report the first only because it is a more straightforward
and useful measure of the vehicle deceleration during the manoeuvre.
In Figures 5a and 8a the tracking variables at 50 and 70 Kph, respectively, are shown.
In particular, the longitudinal speed (upper left), the yaw angle (upper right), the yaw rate
(lower right) and the lateral vehicle position in the inertial frame (lower right) are reported.
In Figures 5b and 8b the braking torques at 50 and 70 Kph are shown. Figures 6 and 9 show
the steering angle, in Figures 7a and 10a are the slip ratios while Figures 7b and 10b show the
tyre slip angles at the four wheels at 50 and 70 Kph, respectively.
The simulation results in Figures 5–7 show very good tracking at 50 Kph. Nevertheless,
we notice excessive oscillation in braking due to imperfect tuning. Braking torques could be
smoothed by increasing the weights in the matrix R. This has to be properly traded off with
the need of fast braking torque variation.
We point out that the tuning of such controller is a very time-consuming task, since each
simulation takes almost 15 min. Therefore, this controller has not been finely tuned. By looking
at the slip ratios in Figures 7a we observe that (i) the controller does not brake much and (ii)

Figure 5. Controller A. Simulation results at 14 m/s entry speed.

622 P. Falcone et al.

Figure 6. Controller A. Simulation at 14 m/s entry speed. Steering angle.

Figure 7. Controller A. Simulation results at 14 m/s entry speed.

Table 1. Simulation results. ControllerA. RMS and maximum tracking errors as function
of vehicle longitudinal speed.

ẋref ψrms ψ̇rms Yrms ψmax ψ̇max Ymax ẋmax

[m/s] [deg] [deg/s] [m] [deg] [deg/s] [m] [m/s]

14 0.7749 1.2110 0.1277 2.5236 3.8832 0.2916 7.0407

19.4 2.817 7.4100 0.6641 10.2552 28.0773 2.3827 9.0314

the slip ratios are within the linear region of the longitudinal tyre force characteristic that, for
this particular scenario is ±2% (see Figure 3a).
In Figures 8–10 the simulation results at 70 Kph are presented. We notice that the RMS
and maximum tracking errors (except the yaw rate maximum tracking errors) are significantly
larger than the errors at 50 Kph. We also observe that at 70 Kph the vehicle operates at the
limit of handling capabilities, as shown by the large values of the tyre slip angles reported in
Figure 10b.
 Vehicle System Dynamics 623

Figure 8. Controller A. Simulation results at 19.4 m/s entry speed.

Figure 9. Controller A. Simulation results at 19.4 m/s entry speed. Steering angle.

Figure 10. Controller A. Simulation results at 19.4 m/s entry speed.

624 P. Falcone et al.

Consider the braking and the steering in Figures 8b and 9, respectively. We observe that
the steering and braking well cooperates in order to achieve the yaw moment necessary for
path following. At the beginning of the first lane change, approximatively 2 s, the controller
starts a right steering. In the same time interval the controller brakes at the right side in order
to generate a negative yaw moment. After 4 s the controller starts a left steering, brakes the
wheels at the left side and releases the braking at the right side in order to achieve a positive
yaw moment. This cooperation between steering and braking is probably not observed in
the simulations at 50 Kph because of the excessive oscillations. A better tuning would make
clearer the braking action at the two vehicle sides. As final remark, in Figure 10a we observe
that the controller limits the slip ratios at the four wheels such that the vehicle always operates
within the region of the longitudinal tyre force characteristic where the maximum force is
achieved.

3.2. Controller B. Simulation results

Next we present the simulation results of Controller B at 14 m/s only. We do not show sim-
ulation results at 19.4 m/s, because Controller B is not able to stabilise the vehicle along the
desired path when the entry speed exceeds 15.3 m/s (55 Kph). Figures 11a and 11b show the
tracking variables and the braking torques, respectively, in Figures 12a and 12b the steering
angle and the desired braking yaw moment are reported while the slip ratios and the tyre slip
angles of the four wheels are shown in Figures 13a and 13b, respectively.
RMS and maximum tracking errors are summarised in Table 2. By comparing the RMS and
maximum tracking errors at 14 m/s in Tables 1 and 2, we observe that Controller A performs
better than Controller B. On the other hand, as shown in Figure 12, the braking intervention
in Controller B is more intuitive than Controller A. We observe, in fact, that between 4 and 6 s
Controller B requires a negative yaw moment and, in the same time interval, steers to the right.
After 6 s the desired braking yaw moment is positive and the steering angle starts to increase.
Nevertheless, we recall that Controller A is not finely tuned, while extensive simulations have
been performed in order to tune the Controller B. Moreover, we observe that the main limitation
of Controller B is that it is not able to stabilise the vehicle at high entry speed.
As final remark, we highlight that in Controller A the selection of the bounds on the braking
torques, Tb∗,• , and the variations of the braking torques, Tb∗,• , is straightforward, since they
can be directly derived from the specifications of the braking system. In Controller B, to

Figure 11. Controller B. Simulation results at 14 m/s entry speed.

 Vehicle System Dynamics 625

Figure 12. Controller B. Simulation results at 14 m/s entry speed.

Figure 13. Controller B. Simulation results at 14 m/s entry speed.

Table 2. Simulation results. Controller B. RMS and maximum tracking

errors as function of vehicle longitudinal speed.

ẋref ψrms ψ̇rms Yrms ψmax ψ̇max Ymax

[m/s] [deg] [deg/s] [m] [deg] [deg/s] [m]

14 1.506 2.637 0.2189 4.2318 8.8536 0.6676

the best of author’s knowledge, there is not a systematic method to compute the bounds on
the braking yaw moment and the variation of braking yaw moment. These bounds could be
estimated through extensive simulations.

3.3. Experimental results

Next we show the experimental results obtained with a low-complexity MPC controller. This
is a sub-optimal MPC controller, derived from Controller A, and is based on successive on-line
linearisations of the non-linear vehicle model. The method stems from an accurate analysis
of the vehicle nonlinearities, the constraints and the performance index in the optimal control
626 P. Falcone et al.

problem (20). The complexity of this controller greatly reduces compared to Controller A,
allowing real-time implementation on standard rapid prototyping hardware with a sampling
time of 50 ms. Further details on the problem formulation and the experimental setup can be
found in [16].
Figures 14 and 15 show the experimental results when the vehicle enters the double lane
change with a longitudinal speed of 50 Kph. In Figures 14a and 14b the tracking variables and
the braking torques at the four wheels are reported, respectively. In particular, the solid lines
represent the desired torques computed by the controller, while the torques delivered by the
braking system are plotted with dashed lines. In Figures 15 the steering angle is reported. The
solid line represents the desired steering angle computed by the controller, while the actual
steering angle is plotted with a dashed line. We also observe the steering angle from the driver
plotted with a dash-dotted line that, during the autonomous path following test, should be as
small as possible.

Figure 14. Controller C. Experimental results at 14 m/s entry speed.

Figure 15. Experimental results at 14 m/s entry speed. Demanded (solid line), delivered (dashed line) and driver’s
(dash-dotted) steering angle.
 Vehicle System Dynamics 627

The experimental results show an overall good tracking of the desired path, although the per-
formance should be further improved (probably through a better tuning) in order to eliminate
oscillations in the braking torques and decrease the braking effort.

4. Conclusions and future works

In this paper, we presented two combined steering and braking MPC controllers based on
vehicle models with different complexities. The first controller (Controller A) is based on a
tenth order four wheels vehicle model with five control inputs. This controller computes the
steering angle and the braking torques at the four wheel in order to best follow a desired path.
The second MPC controller (Controller B) relies on a sixth-order simplified bicycle model
with two control inputs. The controller computes the steering angle and a braking yaw moment.
An external braking logic computes the braking torques at the four wheels such that the desired
braking yaw moment is achieved.
We finally showed experimental results of a combined steering and braking MPC controller
derived from Controller A, relying on a low-complexity MPC formulation presented in [16].
We showed through simulations that Controller A has good tracking performance at both low
and high entry speed. Nevertheless, because of high computational burden, simulations are very
time consuming and the tuning is difficult. We have also shown that, due to a simpler vehicle
model, Controller B is not able to stabilise the vehicle at high entry speeds. Nevertheless, in
spite of performance limitations, the Controller B represents an attractive solution toward real-
time implementation of a low complexity combined steering and braking MPC algorithms.
Future research activities will aim to reduce the performance gap between Controller A and
B and exploit the low-complexity MPC formulations presented in [16] in order to obtain high
performance and low-complexity combined steering and braking MPC approaches.

References

[1] J.N. Dang, Preliminary results analyzing the effectiveness of electronic stability control (esc) systems. Tech.
Rep. DOT-HS-809-790, U.S. Department of Transportation, Washington, DC, 2004.
[2] J. Ackermann, Advantages of active steering for vehicle dynamics control. 29th Conference on Decision and
Control, Honolulu, Hawaii, 1990.
[3] J. Ackermann and W. Sienel, Robust yaw damping of cars with front and rear wheel steering, IEEE Trans.
Control Syst. Technol. 1(1) (1993), pp. 15–20.
[4] J. Ackermann, D. Odenthal, and T. Bunte, Advantages of active steering for vehicle dynamics control. 32nd
International Symposium on Automotive Technology and Automation, Vienna, Austria, 1999.
[5] T. Keviczky, P. Falcone, F. Borrelli, J. Asgari, and D. Hrovat, Predictive control approach to autonomous vehicle
steering, Proceedings of American Control Conference, Minneapolis, Minnesota, 2006.
[6] J. Ackermann, W. Walter, and T. Bunte, Automatic car steering using robust unilateral decoupling. International
Conference on Advances in Vehicle Control and Safety, Genoa, Italy, 2004.
[7] H.E. Tseng, J. Asgari, D. Hrovat, P. Van Der Jagt, A. Cherry, and S. Neads, Evasive maneuvers with a steering
robot, Veh. Syst. Dyn. 43(3) (2005), pp. 197–214.
[8] F. Borrelli, P. Falcone, T. Keviczky, J. Asgari, and D. Hrovat, MPC-based approach to active steering for
autonomous vehicle systems, Int. J. Veh. Autonomous Syst. 3(2/3/4) (2005), pp. 265–291.
[9] D.L. Margolis and J. Asgari, Multipurpose models of vehicle dynamics for controller design. SAE Technical
Papers, 1991.
[10] E. Bakker, L. Nyborg, and H.B. Pacejka, Tyre modeling for use in vehicle dynamics studies. SAE paper No.
870421, 1987.
[11] D.Q. Mayne, J.B. Rawlings, C.V. Rao, and P.O.M. Scokaert, Constrained model predictive control: Stability
and optimality, Automatica 36(6) (2000), pp. 789–814.
[12] S. Motoyama, H. Uki, K. Isoda, and H. Yuasa, Effect of traction force distribution control on vehicle dynamics.
Proceedings of 1992 International Symposium on Advanced Vehicle Control (AVEC92), Yokohama, Japan,
1992, pp. 447–451.
628 P. Falcone et al.

[13] H.E. Tseng, B. Ashrafi, D. Madau, T. A. Brown, and D. Recker, The development of vehicle stability control at
ford, IEEE-ASME Trans. Mechatronics 4(3) (1999), pp. 223–234.
[14] G. Bahouth, Real world crash evaluation of vehicle stability control (vsc) technology. 49th Annual Proceedings,
Association for the Advancement of Automotive Medicine, 2005, pp. 19–34.
[15] E. Bedner, D. Fulk, and A. Hac, Exploring the trade-off of handling stability and responsiveness with advanced
control systems. SAE, SAE 2007-01-0812, 2007.
[16] P. Falcone, Nonlinear model predictive control for autonomous vehicles. PhD thesis, Università del Sannio –
Dipartimento di Ingegneria, June 2007. Available at http://www.grace.ing.unisannio.it/pfalcone.
[17] P. Falcone, F. Borrelli, J. Asgari, H. E. Tseng, and D. Hrovat, Predictive active steering control for autonomous
vehicle systems, IEEE Trans. Control Syst. Technol. 15(3) (2007), pp. 566–580.