Você está na página 1de 86

Modeling and Control of a Motorbike

Marco De Vittori

Dissertação para a obtenção de Grau de Mestre em


Engenharia Electrotécnica e de Computadores

Júri
Presidente: Prof. Antonio Rodrigues
Orientador: Prof. João Silva Sequeira
Vogais: Prof. Eduardo Morgado
Prof. Rodrigo Ventura

Julho 2011
ii
Acknowledgments

Foremost, I would like to express my sincere gratitude to my advisor Prof. João Sequeira for the contin-
uous support of my study and research, for his patience, motivation, enthusiasm, and knowledge. His
guidance helped me in all the time of research and writing of this thesis.
Besides my advisor, I would like to thank the rest of IST institution, who gave me the opportunity
to work in an enjoyable environment. I am grateful to all the ETSEIB professors, for forming me as a
future engineer during four years of hard study in Barcelona and for transmitting me the indispensable
knowledges for the development of my thesis.
I thank all my friends in Lisbon, who made it a convivial place to live by day or during the sleepless
nights we were working together and for all the fun we have had. Lastly, and most importantly, I wish to
thank my parents, Chiara e Carlo. They bore me, raised me, supported me, taught me, and loved me.
To them I dedicate this thesis.

iii
iv
Resumo

O trabalho desenvolvido nesta tese aborda numa primeira fase a construção dos modelos cinemático
e dinâmico para uma motocicleta e, numa segunda fase, o desenho dum controlador capaz de manter
o veı́culo numa trajectória pré-definida. Todos os resultados, finais e intermédios, foram obtidos em
variadas condições através de um simulador construı́do em Matlab.
O modelo cinemático incorpora informação adicional relativamente a outros modelos existentes na
literatura. As caracterı́sticas incorporadas no modelo são por exemplo a inclinação da forquilha, o movi-
mento das suspensões, as consequências do perfil arredondado dos pneumáticos durante a inclinação
da mota e a possibilidade do veı́culo ter duplo braço oscilante.
O modelo dinâmico foi obtido através do método de Lagrange, partindo da energia cinética. Numa
primeira fase o modelo foi analisado através do Lagrangeano livre e numa segunda fase incluı́ram-se
as forças exteriores. Os resultados obtidos sugerem a validade dos modelos.
O sistema de controlo considerado é do tipo PID. Esta escolha tem como motivação o facto do
conjunto de termos (e de parâmetros) desta classe de controladores poder ser naturalmente identificada
com ações tı́picas utilizadas por condutores humanos. A comparação dos resultados obtidos com dados
reais de uma moto num circuito fechado permite concluir que esta estratégia de controlo se adapta aos
objectivos de autonomia, implı́citos nos objectivos da tese.

Palavras-chave: Motocicleta, Controlo, Modelo Dinámico, Modelo Cinemático.

v
vi
Abstract

The work developed in this thesis explains the construction of both kinematic and dynamic models for
a motorcycle, and the design of a control strategy that can command the bike autonomously, under
normal driving conditions. All results have been tested in several diverse conditions through a Matlab
implementation, in order to validate the model and to obtain information about the motorcycle behavior.
The kinematic model builds upon existing models, which do not include some of the aspects con-
tained in this work yet. For instance, the features incorporated in the model are the front fork inclination,
the movement of suspensions along with the damping system, the consequences of the pneumatic
shapes when the motorbike leans, and the analysis of the kinematics of the double swingarm vehicles.
A dynamics model was built through the Lagrangian method. Firstly, the motorcycle has been ana-
lyzed with the help of the free Lagrangian equations and secondly all external forces have been added.
The results obtained suggest that the model is indeed valid.
The control system considered is a simple PID. This choice is motivated by the fact that the structure
and parameters of a PID can be easily identified with the actions used by a typical human rider. The
comparison between the simulated results with those from a real motorbike in a racetrack validates the
choice of this simple control strategy in this particular application.

Keywords: Motorbike, Control, Model, Simulation, Automation, Dynamics, Kinematics.

vii
viii
Contents

Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Resumo . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix

1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 State-of-the-art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Kinematics 5
2.1 Starting model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 forward velocity fork angle γ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Variable forward velocity fork’s length (l) . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Linearization of trigonometric terms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5 Moving swingarm, β angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Linearization of the additional trigonometric terms . . . . . . . . . . . . . . . . . . . . . . 15
2.7 Suspension model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.7.1 Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.7.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.8 Pneumatic shape effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.9 Double swingarm model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.10 Model assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.11 Model validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Dynamics 29
3.1 Kinetic energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3 Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4 Constrained Lagrangian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4.1 Differential-algebraic solution for constrained dynamics . . . . . . . . . . . . . . . 34

ix
3.4.2 Simplified constrained dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5.1 Model building function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.2 Simulation function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6 Qualitative model validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.6.1 No forward velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.6.2 Straight line driving . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.3 Steering angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.6.4 Driver action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.6.5 Further tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.6.6 Simulink model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4 Control 47
4.1 Theoretical basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.2 PID implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5 Results 55
5.1 Simulation background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.2 Empirical control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

6 Conclusions 61
6.1 Achievements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.2 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Bibliography 66

x
List of Tables

5.1 Proprieties of path’s sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

xi
xii
List of Figures

2.1 Lateral view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5


2.2 Rear view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3 Main coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Schematic top view of the motorbike. Relations between the velocities and the wheelbase 7
2.5 γ - δ relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.6 l - γ and l - w relations 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7 l - γ and l - w relations 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.8 l - γ and l - w relations 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.9 Trigonometric functions between 0.36rad and 0.51rad . . . . . . . . . . . . . . . . . . . . 11
2.10 Tangent for −30◦ ≤ δ cos γ ≤ 30◦ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.11 Swingarm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.12 New l value . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.13 Suspension travel - β relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.14 Swingarm and main frame geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.15 Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.16 Geometrical configuration of the wheel while leaning . . . . . . . . . . . . . . . . . . . . . 19
2.17 Double swingarm motorbikes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.18 Double swingarm geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.19 ϕ function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.20 ψ function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.21 ϕ function for double swingarm model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.22 ψ function for double swingarm model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.23 Braking effect on ϕ and ψ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.24 Braking effect on ϕ and ψ for double swingarm model . . . . . . . . . . . . . . . . . . . . 28

3.1 Mass center position . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31


3.2 Real mass center position . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3 Blocs diagram of the Matlab implementation . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.4 No initial leaning angle and no forward velocity. Leaning angle [◦ ] (red) . . . . . . . . . . . 38

3.5 Initial lean angle without forward velocity. Leaning angle [ ] (red) . . . . . . . . . . . . . . 39

xiii
3.6 Additional analysis with initial lean angle and without forward velocity. Leaning angle [◦ ]
(red); rotation speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.7 Straight line driving with and without initial lean angle. Leaning angle [◦ ] (red); forward
velocity [Km/h] (green) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.8 Additional analysis of straight line driving. Leaning angle [◦ ] (red); forward velocity [Km/h]
(green) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

3.9 Steering angle effects. Leaning angle [ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.10 Leaning angle variation. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.11 Steering angle variation. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.12 Forward velocity variation. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rota-
tion speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.13 Torque applied with variation of driver’s mass center position. Leaning angle [ ] (red);
forward velocity [Km/h] (green); rotation speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . 44
3.14 Torque applied through the action of the driver’s feet. Leaning angle [◦ ] (red); forward
velocity [Km/h] (green); rotation speed [◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . 44
3.15 Further analysis. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation speed
[◦ /s] (blue) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.16 Simulink model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4.1 Autonomous driving during a 100m-radius curve. Leaning angle [◦ ] (red); rotation speed
[◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg] (black) . . 49
4.2 Autonomous driving during a 50m-radius curve. Leaning angle [◦ ] (red); forward velocity
[Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent
mass for ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.3 Controlled model with acceleration while turning. Leaning angle [◦ ] (red); forward velocity
[Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent
mass for ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.4 System behavior when modifying Pψ . Leaning angle [◦ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for
ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.5 System behavior when modifying Iψ . Leaning angle [◦ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for
ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

4.6 System behavior when modifying Dψ . Leaning angle [ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for
ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

xiv
4.7 System behavior when modifying Pϕ . Leaning angle [◦ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for
ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

4.8 System behavior when modifying Iϕ . Leaning angle [ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for
ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.9 System behavior when modifying Dϕ . Leaning angle [◦ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for
ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.1 Montmeló circuit sections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55


5.2 Speed evolution during first three curves of Montmeló circuit (telemetry data) . . . . . . . 56
5.3 Speed evolution (km/h) through time (s) in Matlab simulation . . . . . . . . . . . . . . . . 56
5.4 Montmeló curves and trajectory followed by the simulated model [m] (blue) and real mo-
torbike [m] (red) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.5 Evolution of the main variables through time [s]. Leaning angle [◦ ] (red); forward velocity
[Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent
mass for ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.6 Variables evolution through time [s] without Γϕ torque limitation. Leaning angle [◦ ] (red);
forward velocity [Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (ma-
genta); equivalent mass for ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . . . 59
5.7 Variables evolution through time [s] with 60◦ ϕ leaning angle limitation. Leaning angle
[◦ ] (red); forward velocity [Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ]
(magenta); equivalent mass for ϕ-torque [Kg] (black) . . . . . . . . . . . . . . . . . . . . . 60

xv
xvi
Nomenclature

Greek symbols

α Angle between d and x axis.

β Swingarm angle.

β2 Front swingarm angle.

∆t Time step.

δ Steering angle.

 Error.

Γ Torque.

γ Front fork inclination angle.

λ Lagrange multipliers vector.

Ω Generic angular speed.

ψ Vertical rotation angle.

τ Forces vector.

θ Engine rotation angle.

ϕ Leaning inclination angle.

Roman symbols

A Inertia matrix.

B Dissipative forces vector.

b x-coordinate the mass center.

C Contact forces vector.

c Damping factor.

D Control derivative parameter.

xvii
d Distance between the mass center and the coordinates origin.

F Force.

gr Gear ratio.

h z-coordinate the mass center.

I Control Integral parameter.

J Constraints matrix.

K Kinetic energy.

k Spring constant.

l Front fork length.

M Mass and inertia matrix.

m Mass.

N Vertical external force.

O Origin of Cartesian coordinates.

P Control proportional parameter.

Q Actuation variable vector.

q Joint configuration vector.

r Generic radius.

r1 Distance between the rear wheel center and the instant center of vertical rotation of the motor-
bike.

r2 Distance between the front wheel center and the instant center of vertical rotation of the motor-
bike.

Rr Real rear wheel radius.

rt Tyre shape radius.

rw Distance between the wheel center and the tyre shape center.

T Horizontal external force.

t Distance between the swingarm rotation center and the front fork.

V Velocity terms vector.

v Front velocity.

xviii
w Wheelbase.

x, y, z Cartesian components.

Subscripts

0 Rest conditions.

β Corresponding to the swingarm.

ψ̇ Corresponding to the vertical rotation movement.

θ̇ Corresponding to the spin movement.

ϕ̇ Corresponding to the leaning movement.

ẋ Corresponding to the movement along x axis.

ẏ Corresponding to the movement along y axis.

ψ Corresponding to the vertical rotation axis.

θ Corresponding to the spin axis.

ϕ Corresponding to the leaning axis.

ext Corresponding to external constraints.

f Corresponding to the front wheel.

goal Objective reference value.

l Corresponding to the front fork.

m Corresponding to the engine.

r Corresponding to the rear wheel.

Superscripts

T Transpose.

xix
xx
Chapter 1

Introduction

The study of bicycle modeling and control dates back to the last years of 19th century, namely through
the work of F.J. Whipple (see for instance Limbeer and Sharp [2006]). The research on bicycle kinemat-
ics and dynamics produced a number of scientifically interesting examples, e.g., from the viewpoint of
controllability (see for instance Aström et al. [2005]).
During most of the 20th century, the development of the motorbike industry kept the focus of the
research in the engines but it was not until the 80s, when the performance started to boost and new riding
techniques were developed, that a renewed interest in modeling emerged. In recent times there has
been a renewed interest in this type of vehicles from an academic viewpoint, with multiple researchers
discussing different kinematics and addressing also dynamics and control issues.
The main purpose of this work is to achieve a realistic motorbike model, and develop an automatic
control for an autonomous motorbike. Starting from the basic bike structure, already extensively studied
in the literature, the thesis explores also alternative structures, not often used in commercial motorbikes.
The dynamic model is built using the classic free Lagrangian method based on total kinetic energy, and
the external forces that constrain the free Lagrangian. This model is then used to test the automatic
control using a PID based approach.

1.1 Motivation

The development of a motorbike automatic control is clearly a topic of interest for future transportation
progress. The innumerable advantages of an autonomous driving reside in transports optimization,
user comfort, and driving security, among others. For instance, if vehicles were able to completely
govern themselves after having received the destination and the path to be followed (for example with
driving control combined with GPS technology), the driver task would be drastically reduced to a mere
supervision. In a shorter time horizon, as happened with many other driving technologies, different
features of the autonomous control could be integrated gradually in commercial vehicles increasing
security and comfort in diverse driving aspects, without removing human main role.
Although autonomous four-wheel vehicles already exist and the research in this field has widely

1
evolved in the last decades. In what concerns motorbike control this is not the case despite the increas-
ing attention the subject is attracting.
The different physical equilibrium intrinsic to the two-wheels vehicles makes them somewhat more
complex to control, even from the mere theoretical perspective. The search for a dynamic equilibrium
is indeed a great obstacle to the development of a self controlled motorbike is. As opposed to cars,
whose equilibrium is based on the four-wheels contact that guarantees an almost constant geometry
configuration, two-wheel vehicles have to reach the right balance of forces and torques in every time
instant. Moreover, the required geometry configuration depends on several variables, such as front
velocity, steering angle, leaning angle, forces/torques applied by the driver, and forces/torques imposed
by the contact between the motorbike and the ground.
As it will be explained in section 1.2, several studies about motorbikes and their modeling are avail-
able in the literature. The first contribution of this study is the creation of a realistic physical model which
includes diverse kinematic and dynamic effects that are often not included in other models. Among these
are the wheel radius variation due to pneumatic shape during inclination, a double swingarm kinematic
model, and the effect of the rotation of the engine itself. The underlying idea is that the more effects are
included, the more information the automatic control will have available, and, eventually, the easier will
be to find an autonomous controller capable of high performance riding such us that found in motorbike
racing.

1.2 State-of-the-art

The motorcycle modeling has been extensively studied by a number of researchers over many years.
The study of bicycle modeling and control dates back to the last years of 19th century, namely through
the work of F.J. Whipple (see for instance Limbeer and Sharp [2006]). The research on bicycle kinemat-
ics and dynamics produced a number of scientifically interesting examples, e.g., from the viewpoint of
controllability (see for instance Aström et al. [2005]). During most of the 20th century, the development
of the motorbike industry kept the focus of the research in the engines but it was not until the 80s, when
the performance started to boost and new riding techniques were developed, that a renewed interest
in modeling emerged. In recent times there has been a renewed interest in this type of vehicles from
an academic viewpoint, with multiple researchers discussing different kinematics and addressing also
dynamics and control issues.
R.S. Sharp, S. Evangelou and D.J.N. Limbeer studied the effects of an advanced tyre modeling on
the motorcycle dynamic simulation, Sharp et al. [2004]. In their work, many improvements are described,
such as the tyre-road contact geometry, the tyre shear force and moment descriptions, as functions of
load, slip and camber, the tyre relaxation properties, a new analytic treatment of the monoshock rear
suspension mechanism with sample results, the parameter values describing a contemporary high per-
formance machine and rider, the steady-state equilibrium and power checking and the steering control.
The results of the simulations show that the reached model behaves as the field experiments suggested,
leading to some interesting conclusions; for instance, the good results of a more accurately constructed

2
model; the fact that the frame flexibility is an important issue in design and analysis form the point of
view of motorbike oscillations; the stabilizing action of the driver mass.
The linearized equations of motion for an uncontrolled bicycle were presented in Schwab et al. [2004].
In this case, the system is built with many simplifications which permits to obtain a linearizable model.
For example, the bike is considered as a structure formed by rigid bodies, the wheels have no width, the
rider is fixed to the vehicle. Validated results permit to consider the analysis as a benchmark for future
advances based on this simple model, such as the improvements suggested by the authors: the finite
width of the tyres, the control torque at the handle bar, the relative motion between the rider and the rear
frame and tyre models that include wheel slips and compliance.
General dynamics from the perspective of control have been analyzed by Karl J. Åström, Richard
E. Klein, and Anders Lennartsson, Aström et al. [2005]. In their paper, a wide range of different bike
architectures are studied along with the consequences of the particular aspects of each one of them.
A part of the illustrated steps of the dynamic model construction, this work shows the usefulness of
the feedback control in the bike stable behavior, especially as a demonstration that a driver is able to
control some non-conventional bikes, which may be used in some particular applications. An interesting
observation arisen in the analysis is the importance of the automatic control study as a premise for a
deeper understanding of human driving.
Gabriele Virzı̀ Mariotti and Francesco Vitale published an on-line article, Vitale and Mariotti [2007],
that explains the global issue of the rear wheel damping system. Firstly the kinematics of the suspen-
sions is studied, secondly the results are validated, thirdly the dynamics behavior is simulated and finally
the mechanical solicitations of the different pieces are analyzed. Some aspects of the work become ex-
tremely useful for this thesis when approaching the damping system modeling, since the suspensions
structure considered in the article is indeed the same used for the reference motorbike of the thesis1 .
Popov et al, in Popov et al. [2010], explains the aspects related to the steering angle control in
human driving. The stability dependence on control parameters is analyzed starting from the human
automatisms processes. The optimal control and the model predictive control are then implemented for
this aspect of the motorbike driving. The conclusions of the work lead to some open questions about the
obtainment of a more complex model for the driver action and the global strategy frame to be followed
in order to reach a future complete automatic control.

1 Honda CBR600RR

3
4
Chapter 2

Kinematics

The simplest physical model of a motorbike can be obtained from the kinematic analysis. Kinematic
models encode how different speed and position variables are connected with each other and how they
behave when changing system geometry.

2.1 Starting model

Figure 2.1: Lateral view

As shown in Limbeer and Sharp [2006]1 , the kinematics equations for a simple motorbike model are,

ẋ = v cos ψ (2.1a)
1 page 40

5
Figure 2.2: Rear view

ẏ = v sin ψ (2.1b)

v tan δ
ψ̇ = (2.1c)
w cos ϕ

Where v is the forward velocity, x direction is longitudinal with respect to the motorbike’s frame and y is
perpendicular to this. x and y coordinates indicate the motorbike position in the world frame. The other
variables and parameters are indicated in picture 2.3.

Figure 2.3: Main coordinates

The first two equations are obtained straight from basic geometric analysis whereas the third one can

6
be explained with the help of picture 2.4.

Figure 2.4: Schematic top view of the motorbike. Relations between the velocities and the wheelbase

v = r1 ψ̇ cos ϕ (2.2a)

w
= tan δ (2.2b)
r1

Combining these two equations we obtain what we wanted to demonstrate,

v tan δ
ψ̇ = (2.3)
w cos ϕ

In the case of Honda CBR600RR, w parameter is 1369.0 mm.

2.2 forward velocity fork angle γ

As a γ angle is added to the model, the rotation of the forward velocity wheel with respect to z axis is
not equal to the steering angle of the forward velocity fork with respect to the frame anymore. In fact,
forward velocity wheel vertical rotation (δ cos λ) is now only a vectorial portion of δ forward velocity fork
rotation, as explained in the diagram 2.5 2 :

As a result, the kinematics equations must be modified to,

ẋ = v cos ψ (2.4a)

ẏ = v sin ψ (2.4b)

v tan(δ cos γ)
ψ̇ = (2.4c)
w cos ϕ

In resting conditions, the Honda CBR600RR has a γ angle of 23.5◦ , so that cos γ = 0.91706.

2 See Schwab et al. [2004], page 2

7
Figure 2.5: γ - δ relation

2.3 Variable forward velocity fork’s length (l)

An important variable in the geometry of a motorbike is the forward velocity fork length. As experienced
riders know, changes in the order of 1 mm can affect significantly the behavior of the motorbike. There-
fore, it is important to assess how this geometric variable influences the kinematics. In particular the γ
and w parameters (Figure 2.6) are affected by changes in the forward velocity fork length.

Figure 2.6: l - γ and l - w relations 1

In the case of the Honda CBR600RR, the steering axis can be assumed to pass through the forward
velocity wheel center and since both motorbike wheels have approximately equal radius (difference is
indeed 15 mm), the red right triangle in diagram 2.6 can be simplified. Labels are assigned to its sides:
l would be the length of the forward velocity fork, w the horizontal wheelbase and w cos γ the distance
between the center of the rear wheel and the forward velocity fork (Figure 2.7).

8
Figure 2.7: l - γ and l - w relations 2

When l changes the geometry of the triangle changes, and the motorbike model as well, whereas the
w cos γ side remains constant. This measure is constant because it represents a fixed property of the
main frame and swingarm, which are still fastened to each other in the current model. That is why
w cos γ = w0 cos γ0 remains constant when the triangle changes its properties into generic states (l, γ, w)
from an initial configuration (l0 , w0 , γ0 ) as shown in diagram 2.7.

Figure 2.8: l - γ and l - w relations 3

This diagram helps to obtain some useful equations, which will be included in the kinematics, namely,

l
sin γ = (2.5a)
w
cos γ0 w0
cos γ = (2.5b)
w
l
tan γ = (2.5c)
w0 cos γ0
l
w= (2.5d)
l
sin arctan
w0 cos γ0

The kinematics equations after having added the l variable are,

ẋ = v cos ψ (2.6a)

ẏ = v sin ψ (2.6b)
 
l
v tan δ cos arctan
w0 cos γ0
ψ̇ = (2.6c)
l
cos ϕ
l
sin arctan
w0 cos γ0

For the Honda CBR600RR, the forward velocity suspension trail is at most 119.0 mm, whereas l0 =

9
w0 sin γ0 = 545.9 mm. Therefore, variation gap of γ and w parameters can be calculated,

119.0
lmax = lo + = 695.4mm
2

119.0
lmin = lo − = 486.4mm
2

lmax
wmax = = 1435.2mm
lmax
sin arctan
w0 cos γ0
lmin
wmin = = 1346.4mm
lmin
sin arctan
w0 cos γ0

∆w = 47.4mm

w0 cos γ0
γmax = arccos = 28.98◦
wmax

w0 cos γ0
γmin = arccos = 21.17◦
wmin

∆γ = 4.57◦

2.4 Linearization of trigonometric terms

The behavior of the trigonometric functions can be studied for 21◦ ≤ γ ≤ 29◦ . In fact, this is the
approximate range of γ values seen in section 2.3.
As shown in Figure 2.9, these functions can be studied as linear functions within the gap of real move-
ment. In fact, the error that appears after this simplifications can be considered negligible. The equations
of section 2.3 can then be simplified around the point of rest condition,

sin γ ' 0.01580γ + 0.02673 (2.7a)

cos γ ' −0.00739γ + 1.08898 (2.7b)

tan γ ' 0.02133γ − 0.06427 (2.7c)

l
γ= + 3.01305 (2.7d)
0.02133w0 cos γ0

w0 cos γ0 (w0 cos γ0 )2


w= = (2.7e)
−0.00739γ + 1.08898 1.06671w0 cos γ0 − 0.346524l

After these simplifications the kinematics equations become,

ẋ = v cos ψ (2.8a)

10
(a) Sine (b) Cosine

(c) Tangent

Figure 2.9: Trigonometric functions between 0.36rad and 0.51rad

ẏ = v sin ψ (2.8b)
0.346524l
v tan(δ(1.06671 − ))
w0 cos γ0
ψ̇ = (2.8c)
(w0 cos γ0 )2
cos ϕ
1.06671w0 cos γ0 − 0.346524l

The graph 2.10 shows that the tangent of the rotation angle of the forward velocity wheel can be also
linearized, because its value will always be small enough to consider tangent function a f (x) = x function
(Figure 2.10).

The final equations for the model are then,

ẋ = v cos ψ (2.9a)

ẏ = v sin ψ (2.9b)

vδ(1.06671w0 cos γ0 − 0.34652l)2 π/180


ψ̇ = (2.9c)
(w0 cos γ0 )3 cos ϕ

11
Figure 2.10: Tangent for −30◦ ≤ δ cos γ ≤ 30◦

2.5 Moving swingarm, β angle

A moving swingarm is the final step for the completion of the standard kinematic model. This movement
can be reduced to the value of β angle between the swingarm and the horizontal x axis (Figure 2.11).

Figure 2.11: Swingarm

From now on, the w0 and γ0 parameters can not be considered fixed values anymore. In particular
the w0 cos γ0 distance is not constant, as the main frame and the swingarm are now considered as two
separate solid bodies. Therefore we convert these parameters into generic γ and w variables. Moreover
value of l variable has to be modified, as explained in diagram 2.12.
The l value will now vary between 356.0 and 475.0 mm, whereas 415.0 mm is the rest position value.
The rear suspension has a travel of 130 mm (the rear wheel’s axis vertical movement), so that the
range of β values can be calculated with the help of the diagram 2.133 ,
The range of β values is approximately [0.4◦ ; 13.4◦ ], being 10.5◦ the value for rest position.
It is now possible to describe γ and w as β- and l-dependent functions (Figure 2.14).
The w distance can be calculate as,
3 Bikervoodoo [2010]

12
Figure 2.12: New l value

Figure 2.13: Suspension travel - β relation

Figure 2.14: Swingarm and main frame geometry

w = s cos β + t cos γ + l sin γ (2.10)

where s is 578.5 mm and t is 682.5 mm, so that,

w = 578.5 cos β + 682.5 cos γ + l sin γ (2.11)

13
To determine the γ value we can follow these steps,

l
l t sin(arctan − γ)
cos arctan = t (2.12a)
t s sin β

l s l
γ = arctan − arcsin( sin β cos arctan ) (2.12b)
t t t

l l
γ = arctan − arcsin(0.8422 sin β cos arctan ) (2.12c)
682.5 682.5

The equations of the kinematics model that includes the swingarm are detailed below. Given their
algebraic complexity, they will be simplified in the following section.

ẋ = v cos ψ (2.13a)

ẏ = v sin ψ (2.13b)
π l l
ψ̇ =(vδ (1.06671(578.5 cos β + 682.5 cos(arctan − arcsin(0.8422 sin β cos arctan ))
180 682.5 682.5
l l
+ l sin(arctan − arcsin(0.8422 sin β cos arctan )))
682.5 682.5
l l
cos(arctan − arcsin(0.8422 sin β cos arctan )) − 0.346524(l + 130.9))2 )
682.5 682.5 (2.13c)
l l
(((578.5 cos β + 682.5 cos(arctan − arcsin(0.8422 sin β cos arctan ))
682.5 682.5
l l
+ l sin(arctan − arcsin(0.8422 sin β cos arctan )))
682.5 682.5
l l
cos(arctan − arcsin(0.8422 sin β cos arctan )))3 cos ϕ)−1
682.5 682.5
The ranges of w and γ values can be calculated since we know l and β ones,

356mm ≤ l ≤ 475mm

0.4◦ ≤ β ≤ 13.4◦

lmin lmin
γmin = arctan − arcsin(0.8422 sin βmax cos arctan ) = 17.58◦
682.5 682.5

lmax lmax
γmax = arctan − arcsin(0.8422 sin βmin cos arctan ) = 34.56◦
682.5 682.5

∆γ = 16.98◦

14
wmax = 578.5 cos βmin + 682.5 cos γmax + lmax sin γmax = 1410.0mm

wmin = 578.5 cos βmax + 682.5 cos γmin + lmin sin γmin = 1320.9mm

∆w = 89.0948mm

2.6 Linearization of the additional trigonometric terms

The trigonometric functions contained in the kinematic ones are now studied aiming at simplifying the
whole model. These functions are,

l l
arctan ; sin β ; cos arctan ;
682.5 682.5

l
sin(arctan − γ) ; cos β ; cos γ ; sin γ
682.5

The graphs for the real range of values of these variables are plotted(Figure 2.15) and show that, within
those ranges the linearizations are acceptable.
The linearizations considered are then,

l
arctan ' 0.061258l + 5.73919 (2.14a)
682.5

sin β ' 0.01729β + 0.000065 (2.14b)

l
cos arctan ' −0.000553l + 1.08362 (2.14c)
682.5
l
sin(arctan − γ) ' 0.001064l − 0.017363γ + 0.099674 (2.14d)
682.5

cos β ' −0.002092β + 1.00386 (2.14e)

cos γ ' −0.007642γ + 1.08765 (2.14f)

sin γ ' 0.01562γ + 0.027434 (2.14g)

and the γ and w expressions can now be reformulated as,

γ = 0.000464(l(β + 132.087) − 1959.53(β − 6.31298)) (2.15a)

w = 7.24428 × 10−6 l2 (β + 132.087) + l(−0.016614β − 0.202461) + 3.52975(β + 366.351) (2.15b)

yielding the kinematic equations,

ẋ = v cos ψ (2.16a)

ẏ = v sin ψ (2.16b)

15
l
(a) arctan (b) sin β
682.5

l l
(c) cos arctan (d) sin(arctan − γ)
682.5 682.5

(e) cos β (f) cos γ

(g) sin γ

Figure 2.15: Functions

16
vδ 3
ψ̇ = − 7.7 × 108 [l (β + 132.1)2 − 4253l2 (β + 75.8)(β + 132.1)+
cos ϕ
5 × 106 l(β 2 + 195.3β + 8925.6) − 9.6 × 108 (β 2 + 516.7β + 53326.4)]2

[l(β + 132.1) − 1959.5(β + 150.3)]−3 [l2 (β + 132.1) − 2293.4l(β + 12.2) + 487247(β + 366.4)]−3
(2.16c)

Equation 2.16 is not simple to understand intuitively. However, it can be easily used in a simulation of
the kinematic behavior of the motorbike. It is indeed almost completely polynomial and simpler than
equation 2.9c.

2.7 Suspension model

2.7.1 Equations

In order to model the behavior of the suspensions we can introduce some simple dynamic equations
representing the damping system. This dynamic model will be able to provide some useful data for
the kinematic study of the motorbike. Namely, two input variables of the kinematic model (the front
fork length and the swingarm angle) come from these dynamic equations. Though its simplicity, this
suspension model has a dynamic behavior that reasonably emulates that of a real suspension. By
coupling it with the kinematic model one can understand how the bike geometry influences its behavior.

A standard second order model will be used for the forward velocity and rear suspensions,

¨lml + lc
˙ l + lkl = Fl (2.17a)

l¨r mβ + l˙r cβ + lr kβ = Fβ (2.17b)

where l is the forward velocity fork length, m the suspended mass, c the dynamic coefficient, k the
static coefficient, F the force applied on the suspension and lr the height of the rear wheel center, that
depends on β swingarm angle,

lr = 578.5 sin β (2.18)

so that damping equations are,

¨lml + lc
˙ l + lkl = Fl (2.19a)

d2 d
(578.5 sin β)mβ + (578.5 sin β)cβ + 578.5 sin βkβ = Fβ (2.19b)
dt2 dt

This new element integrated in the kinematic model permits us to express the forward velocity fork
length and the swingarm angle as functions of the respective loads.

17
2.7.2 Parameters

The F , l and β in equations 2.19a and 2.19a are variables, whereas m, c and k are constant parame-
ters. In order to be able to simulate the behavior with different F , l and β input variables the following
parameters are used4 .

ml = 99kg
cl = 52N s/m
kl = 4800N/m
mβ = 81kg
cβ = 78N s/m
kβ = 50000N/m

l0 = 0.415m
β0 = 0.18326rad

2.8 Pneumatic shape effect

The pneumatic shape has an effect on the wheels radius, because as the motorbike leans (ϕ angle), the
distance between the wheel center and the point of contact with the floor decreases (see diagram 2.16,
where rw is the wheel rim radius and rt the tyre tube radius.

To include this variation in the model the motorbike speed must be defined as Rr θ˙r and specify Rr as a
function depending on ϕ,

rt
Rr = rt cos ϕ + rw cos arcsin( sin ϕ) (2.20)
rw

where Rr is the effective wheel radius, rw = 227.7mm and rt = 87.2mm for the Honda CBR600RR
(figures 2.1 and 2.2). The v forward velocity in equations 2.16a, 2.16b and 2.16 can now be expressed
as a function of θ̇ engine speed, gr gear ratio and Rr rear wheel radius,

v = θ̇grRr (2.21)

The kinematic model can be summarized in the following equations,

ẋ = θ̇grRr cos ψ (2.22a)

ẏ = θ̇grRr sin ψ (2.22b)

4 mass distribution: 55 percent forward velocity, 45 percent rear.

18
Figure 2.16: Geometrical configuration of the wheel while leaning

θ̇grRr δ 3
ψ̇ = − 7.7 × 108 [l (β + 132.1)2 − 4253l2 (β + 75.8)(β + 132.1)+
cos ϕ
5 × 106 l(β 2 + 195.3β + 8925.6) − 9.6 × 108 (β 2 + 516.7β + 53326.4)]2

[l(β + 132.1) − 1959.5(β + 150.3)]−3 [l2 (β + 132.1) − 2293.4l(β + 12.2) + 487247(β + 366.4)]−3
(2.22c)

¨lml + lc
˙ l + lkl = Fl (2.22d)

d2 d
2
(578.5 sin β)mβ + (578.5 sin β)cβ + 578.5 sin βkβ = Fβ (2.22e)
dt dt

2.9 Double swingarm model

An interesting case of motorbike geometry is the double swingarm system. It has seldom been used
in production bikes, the Yamaha GTS 1000 and Bimota Tesi, and even racing applications, which in a
sense tend to use more extreme designs, has a single known case, the Elf 500cc, raced during the 80’s
decade (see Figure 2.17).
The benefits and disadvantages of such geometry have been an interesting topic of discussion only
in motorbike forums, where maneuverability is the main evaluation criterion.
The reason for developing such an innovative geometry is argued to be an attempt to separate the
steering dynamics and the damping ones. In fact, neither the steering angle nor the suspensions do not

19
(a) Yamaha GTS 1000 (b) Bimota Tesi

(c) ELf 500

Figure 2.17: Double swingarm motorbikes

need any forward velocity fork anymore, and as a consequence there will not be any γ rake angle or
trail.
The particular steering system, shown in picture 2.17(b), where the forward velocity wheel is con-
trolled indirectly by the driver through two pairs of rods connected to the handlebar. This steering method
is indeed completely independent from damping system.
A great advantage of the double swingarm innovation is the fact that driving control is not disturbed by
the forces that are transmitted from the floor to the driver’s hands through the forward velocity fork. In the
conventional motorbikes, the torque applied by the driver has to face constantly the effects of the forward
velocity fork tensions produced by road imperfections and by the variations of load on the forward velocity
wheel. These tensions are continuously modified and often poorly absorbed by the damping system.
By separating the steering control and the forward velocity suspensions, human driving might become
more comfortable in terms of controlling the handlebar.
On the drawbacks, the fact that forward velocity wheel position is not directly handled by the driver
implies that driving automatisms have to be modified. The system then loses maneuverability in terms
of how controlling the motorbike and following a required trajectory. The hand movements by the driver,
and the applied torques, are not directly related with the forward velocity wheel anymore, so that driving
may yield a different feeling when compared to conventional motorbikes.
Because of the different steering and damping systems, kinematic equations for a double swingarm
motorbike need to be changed to,

ẋ = θ̇grRr cos ψ (2.23a)

ẏ = θ̇grRr sin ψ (2.23b)

π θ̇grRr δ
ψ̇ = (2.23c)
w(β, β2 )180 cos ϕ

20
where w depends on both forward velocity and rear swingarms’ position. The application to the kinematic
model requires that some constant parameters have to be determined, namely w in rest conditions (1369
mm as in previous model), forward velocity swingarm length (578.5 mm, as the Honda CBR600RR’s rear
one) and its angle in rest conditions (β2 = 10.5◦ , as the rear one). The new model geometry is shown
in diagram 2.18.

Figure 2.18: Double swingarm geometry

As a result, w and final kinematic equations can now be calculated,

w = s cos β + s2 cos β2 + 231 (2.24a)

ẋ = θ̇grRr cos ψ (2.24b)

ẏ = θ̇grRr sin ψ (2.24c)

π θ̇grRr δ
ψ̇ = (2.24d)
(578.5(cos β + cos β2 ) + 231)180 cos ϕ

where β and β2 values will be between 0.4◦ and 13.4◦ .

2.10 Model assessment

The kinematics was tested in a Matlab environment. The analysis was carried out considering ϕ, the
leaning angle, and ψ, the rotation angle. In both cases variables θ̇, the engine speed, Fl , the forward
velocity fork force, Fβ , the swing arm force, and δ, the steering angle, are preset. In the leaning angle
analysis, the ψ angle is fixed and ϕ is calculated at each time step. In the rotation angle analysis the ϕ
is fixed and ψ is constantly recalculated.
The evolution of the leaning angle ϕ is obtained from a function that calculates which inclination the
motorbike would have as the equilibrium configuration when running through a specified path, with a
determined speed, steering angle and suspensions forces. After having introduced the preset variables
values, the simulation starts with a loop, where for each time step ϕ is calculated along with l and β
values. ψ is obtained from ψ̇ using the Euler integration, which comes from a simple calculation of v, the
forward speed, and r1, the curve radius,

21
v θ̇Rr gr
ψ̇ = = (2.25)
r1 r1

The results are shown in section 2.11.


The rotation angle analysis is obtained using a function that calculates which path the motorbike
would follow when running with determined v forward speed, ϕ inclination, δ steering angle, and Fl
forward velocity and Fβ rear suspensions forces.
The suspension model is introduced by discretizing equations 2.26a to obtain β and l values for each
time step,

2ml cl ml
Fl + ( 2
+ )l1 − l0
l2 = ∆t ∆t ∆t2 (2.26a)
ml cl
+ + kl
∆t2 ∆t
Fβ 2mβ cβ mβ
+( 2 + ) sin β1 − 2
sin β0
β2 = arcsin( 578.5 ∆t ∆t ∆t ) (2.26b)
mβ cβ
+ + k β
∆t2 ∆t
Including these equations in the main function loop yields l and β for each time step as a function of
preset Fl and Fβ external forces.
In the case of the double swingarm model, functions are equal to the conventional motorbike model
ones apart from the fact that the kinematic relations between the preset variables and ϕ or ψ will now
follow equations 2.24b, 2.24c and 2.24d.

2.11 Model validation

The model validation consists in simulating the motorbike passing through a two-curves trajectory ob-
tained by adjusting the driving variables. Apart from the curve path, the analysis includes a loads change
similar to the one that appears in hard braking condition in order to test dumping model (shorter forward
velocity fork suspension length and bigger rear suspension beta angle). Finally the double swingarm
model is simulated and the results are compared to those obtained for the standard motorbike.
Introducing as input variables two 25m r1 radius curves, one to the right, after 2s, and one to the left,
after 8s, 100km/h of v forward speed, δ = 3◦ for the steering angle, Fl = 30Kg for the forward velocity
fork load variation, and Fβ = 80kg for the swingarm load variation, we obtain the results of Figure 2.19.
The motorbike is stable with a ϕ = 31.4◦ leaning angle in normal conditions and a ϕ = 28.5◦ angle
when the loads change as during hard braking after 5s from the beginning of the simulation. This means
that if driver brakes while turning, the equilibrium leaning angle will become smaller.
Introducing now as input variables two 30◦ ϕ inclination curves, one to the right, after 2s, and one
to the left, after 8s, v = 100km/h, δ = 3◦ , Fl = 30Kg, and Fβ = 80kg, the results of Figure 2.20 are
obtained.
The motorbike rotates with ψ̇ = 1.098◦ /s in normal conditions and ψ̇ = 1.124◦ /s when loads changes
after 5s of simulation. In other words, when braking during a curve, rotating speed increases, which

22
means that the driver can turn with a smaller radius curve.
Due to the geometry of the tires, the pneumatic effect, v is not constant but it depends on ϕ. In fact,
although θ̇ engine speed is constant, v modifies its values as the motorbike leans. As shown in pictures
2.19(b) and 2.20(b) forward velocity changes from 100km/h to 96km/h, because Rr rear wheel radius
is also decreasing due to the increasing ϕ leaning angle.
The graphs show how the motorbike behaves when forces generated by the suspension are applied
in the kinematics model, as the rider accelerates or brakes. l and β depend on Fl and Fβ , which are
the loads on the wheels that modify the system equilibrium. The Matlab functions include the differential
equations specified in section 2.7. The results are shown in Figure 2.23 after 5s of simulation. At that
instant l decreases 60mm approximately, whereas β augments 1.6◦ after some little oscillations until
second 7 of simulation, when their values eventually stabilize.
In the double swingarm kinematics, the effects on ϕ and ψ angles are strongly related with the w
wheelbase value. In fact, a shorter w will increase ψ̇ on one hand and decrease ϕ on the other hand (if
all other variables have constant values), as shown in figures 2.21 and 2.22.
The model has been tested under the same conditions of the previous simulations: two 25m radius
curves, one to the right after 2s and one to the left after 8s, v = 100km/h, δ = 3◦ , Fβ2 = 30Kg, and
Fβ2 = 80kg, and two 25◦ ϕ inclination curves, v = 100km/h, δ = 3◦ , Fβ2 = 30Kg, and Fβ2 = 80kg for ψ
analysis.
The results are shown in Figures 2.24(a) and 2.24(b) and show the variations of ϕ and ψ̇ (19.70 −
19.37 = 0.33◦ and 1.1845 − 1.1825 = 0.002◦ /s respectively).
This perturbations are smaller than the consequences of the same loads application in the conven-
tional motorbike model (see Figure 2.23) where variations are 2.9◦ and 0.026◦ . In fact, apart from the
benefits and the disadvantages explained in section 2.9, another interesting aspect of this kind of for-
ward velocity kinematics is the fact that when the load on the forward velocity wheel increases, w also
increases, whereas in a conventional motorbike w would decrease as seen in section 2.3. This effect
leads to a smaller perturbation of ϕ or ψ variables when braking or accelerating, which could mean a
more comfortable driving (thus in accordance to the subjective opinion of some riders). When braking,
the forward velocity swingarm load increases decreasing β2 value, whereas β augment as in a con-
ventional motorbike. These two variations can compensate each other with some well-chosen damping
constants and swingarm lengths.
In terms of motorbike driving easiness overall advantages and disadvantages appear in the alternative
system of connection between the frame and the forward velocity wheel: on the one hand additional
comfort is gained firstly by avoiding transmitting the damping forces to the hands of the driver and
secondly thanks to the fact that w wheelbase is less variable when the loads on the wheels change
during a strong acceleration or a hard braking; on the other hand, the indirect control of forward velocity
wheel position leads to a worse maneuverability.

23
(a) x − y path (b) v forward speed

(c) δ steering angle (d) ϕ leaning angle

(e) ψ̇ rotating speed (f) ψ rotating angle

(g) l forward velocity fork length (h) β swingarm angle

Figure 2.19: ϕ function


24
(a) x − y path (b) v forward speed

(c) δ steering angle (d) ϕ leaning angle

(e) ψ̇ rotating speed (f) ψ rotating angle

(g) l forward velocity fork length (h) β swingarm angle

Figure 2.20: ψ function


25
(a) x − y path (b) v forward speed

(c) δ steering angle (d) ϕ leaning angle

(e) ψ̇ rotating speed (f) ψ rotating angle

(g) β2 swingarm angle (h) β swingarm angle

Figure 2.21: ϕ function for double swingarm model


26
(a) x − y path (b) v forward speed

(c) δ steering angle (d) ϕ leaning angle

(e) ψ̇ rotating speed (f) ψ rotating angle

(g) β2 swingarm angle (h) β swingarm angle

Figure 2.22: ψ function for double swingarm model


27
Figure 2.23: Braking effect on ϕ and ψ

Figure 2.24: Braking effect on ϕ and ψ for double swingarm model

28
Chapter 3

Dynamics

A kinematics model describes how the geometry of the bike affects its performance without account-
ing for physical parameters such as masses, inertias, and constraints. These are accounted for in a
dynamics model.
In this section, a dynamic model is built including effects such as the rotation of the engine and the
variable loads on the wheels that appear thanks to the suspensions effects included in the kinematics
model. The motorbike dynamics in this chapter is developed using the Lagrangian approach. However,
opposed to most of the models developed in the literature, a symbolic model obtained using symbolic
computational resources (namely those provided by Matlab) is developed. The computational complexity
of such model rends a formal analysis difficult. However, it provides a valuable tool to simulation studies.
The Lagrangian technique follows the usual workflow. In the first step the free Lagrangian is com-
puted from the total kinetic energy. The contact forces that constrain the motion of the motorbike are not
included. In the second step these constraints are included in the model.

3.1 Kinetic energy

The kinetic energy in the motorbike is due to any moving component, namely frame, engine and wheels.
Let ϕ be the angle determining the rotational energy with respect to the x axis, ψ be the angle determin-
ing the rotational energy with respect to the z axis, and the total kinetic energy due to these rotations
be denoted by Kϕ̇ψ̇ . It is assumed that energy associated to elevation changes can be neglected. In a
sense this corresponds to say that the motorbike moves over an horizontal surface as it is most often
the case and hence there is no lack of generality.
The kinetic energy due to the linear motion, again assuming the motion in the horizontal plane, is
due to the motion along the x and y directions and it is denoted by Kẋẏ .
The rotational energy of the front and rear wheels and engine is denoted Kθ̇f , Kθ̇r and Kθ̇ , respec-
tively.
Some kinetic energies will not be included in the system because their values can be assumed
to be irrelevant in comparison with the included ones. This fact is due to the small masses or small

29
velocities of the different bodies. For example, the rotational energy of the swingarm, for the front fork,
the translation energy of the front and rear suspensions and the translation energy of the engine pieces
are not included.
The kinetic energy is thus calculated through following expressions,

  
1 T 1  h2 hb ϕ̇
Kϕ̇ψ̇ = Ω AΩ= ϕ̇ ψ̇ m    (3.1a)
2 2 hb b2 ψ̇
 
1  ẋ
Kẋẏ = ẋ ẏ m   (3.1b)
2 ẏ

1
Kθ̇f = mf rf2 θ̇f2 (3.1c)
4
1
Kθ̇r = mr rr2 θ̇r2 (3.1d)
4
1 2 2
Kθ̇ = mm rm θ̇ (3.1e)
4

where Ω is the angular speed vector and A is the inertia tensor, which depends on the distances between
mass center and the rotating axis (h = h(l, β) and b = b(l, β) are coordinates of total mass center,
illustrated in Figure 3.1). This inertia is calculated for the total mass as it were concentrated in a single
point and, therefore, the diagonal elements of the matrix correspond to the square of the distances
whereas the non-diagonal elements are a multiplication between them.
m is the total mass, mf is the front wheel mass, mr is the rear wheel mass, mm is the engine’s
rotating part mass, rf is the front wheel radius, rr is the rear wheel radius, rm is the engine’s rotating
part radius, θ̇f = θ̇f (θ̇, ϕ) is the front wheel angular speed, θ̇r = θ̇r (θ̇) is the rear wheel angular speed,
θ̇ is the engine angular speed. All masses and inertias are simplified as point-masses (m) or thin disks
(mf , mr and mm ).
The goal of the Lagrangian based dynamics is to obtain an energy balance as a function of the
variables ϕ, θ̇, δ, l and β. Therefore, the kinematics model of chapter 2 will be used, in addition to some
other kinematic equations to complete the equations system that represents the dynamic model:

ẋ = θ̇grRr cos ψ (3.2a)

ẏ = θ̇grRr sin ψ (3.2b)

θ̇grRr
θ̇f = (3.2c)
cos δrf

θ̇r = grθ̇ (3.2d)

Rr = Rr (ϕ) is the rear wheel radius including the pneumatic shape effect as explained in section 2.8,
gr is the rear wheel-engine angular speeds ratio, ψ = ψ(ϕ, θ̇, δ, l, β) is calculated with the kinematic
equation 2.16 of section 2.6 and h = h(l, β) and b = b(l, β) are the coordinates of total mass center
which depend on the length of suspensions.

30
Figure 3.1: Mass center position

h = d cos(α) (3.3a)

b = d sin(α) (3.3b)

where d and α are polar coordinates of the mass center in rest condition. Adding the front and rear
suspensions movements, h and b, the equations are,

h = d cos(γ − (γ0 − α)) (3.4a)

b = d sin(γ − (γ0 − α)) (3.4b)

γ0 is the front fork angle in rest conditions (23.5 ◦ ) and γ is generalized as an l and β-dependent function
(equation 2.15a),

h = d cos(0.464(l(β + 2.3054) − 1.95953(β − 0.11018)) − (γ0 − α)) (3.5a)

b = d sin(0.464(l(β + 2.3054) − 1.95953(β − 0.11018)) − (γ0 − α)) (3.5b)

all values are in meters and radians.


The kinetic energy is therefore calculated with following equations,

1
Kϕ̇ψ̇ = m(ϕ̇h + ψ̇b)2 (3.6a)
2
1
Kẋẏ = mRr2 θ̇2 gr2 (3.6b)
2
mf Rr2 θ̇2 gr2
Kθ̇f = (3.6c)
4 cos2 (δ cos γ)
1
Kθ̇r = mr rr2 θ̇2 gr2 (3.6d)
4

31
1 2 2
Kθ̇ = mm rm θ̇ (3.6e)
4

and the total kinetic energy of the system is,

K = Kϕ̇ψ̇ + Kẋẏ + Kθ̇f + Kθ̇r + Kθ̇ (3.7a)

m(ϕ̇h + ψ̇b)2 mRr2 θ̇m


2
gr2 mf Rr2 θ̇m
2
gr2 mr rr2 θ̇m
2
gr2 2 2
mm rm θ̇m
K= + + 2
+ + (3.7b)
2 2 4 cos (δ cos γ) 4 4

3.2 Parameters

To complete the dynamics model all parameters must be defined as constants and let the model be a
function that depends only on required variables (ϕ, θ, δ, l and β). In the kinetic energy expression
(equation 3.7b) the following parameters are present: the coordinates of total mass center in rest con-
ditions α and d, masses m, mf , mr , mm , radius rr , rm , and the speed ratio gr, which are assumed to
have the following values (from a Honda CBR600RR motorbike (pictures 3.2 and 2.1).

α = 0.6981317rad
d = 0.85m

m = 180Kg
mf = 3.3Kg
mr = 4.5Kg
mm = 2.5Kg

rr = 0.3149m
rm = 0.1m

gr = 0.192054

3.3 Lagrangian

Following the Lagrange method procedure, the dynamics equations are obtained by calculating the
kinetic energy derivatives. In this case, the model does not include the potential energy variation due
to vertical position changes, as the motorbike is assumed to be moving in a horizontal plane and hence
potential energy is constant.

d ∂K ∂K
− = Γϕ (3.8a)
dt ∂ ϕ̇ ∂ϕ
d ∂K ∂K
− = Γθ (3.8b)
dt ∂ θ̇ ∂θ

32
Figure 3.2: Real mass center position

This system of equations can also be written in matrix form,

 
ϕ̈
M +V =τ (3.9)
θ̈

where,

 
Γϕ
τ =  (3.10a)
Γθ

∂2K ∂2K
 
 ∂ ϕ̇2 ∂ ϕ̇∂ θ̇ 
 
M = (3.10b)
 

 2 2

 ∂ K ∂ K 
∂ θ̇∂ ϕ̇ ∂ θ̇2

∂2K ∂2K
 
∂K
 − ∂ϕ + ∂ ϕ̇∂ϕ ϕ̇ + ∂ ϕ̇∂θ θ̇ 
 
V = (3.10c)
 

∂2K ∂2K
 
 ∂K 
− + ϕ̇ + θ̇
∂θ ∂ θ̇∂ϕ ∂ θ̇∂θ

The whole system will be therefore summarized in following equation,

∂2K ∂2K ∂2K ∂2K


   
∂K
 ∂ ϕ̇2 − + ϕ̇ + θ̇  
∂ϕ ∂ ϕ̇∂ϕ ∂ ϕ̇∂θ 
  
∂ ϕ̇∂ θ̇ 
 
 ϕ̈  Γ
  ϕ 
+ = (3.11)
 
  
∂ 2 K  θ̈ ∂2K 2 Γθ
 2   
 ∂ K  ∂K ∂ K 
− + ϕ̇ + θ̇
∂ θ̇∂ ϕ̇ ∂ θ̇2 ∂θ ∂ θ̇∂ϕ ∂ θ̇∂θ

33
3.4 Constrained Lagrangian

3.4.1 Differential-algebraic solution for constrained dynamics

The dynamics model obtained from the free, that is unconstrained, Lagrangian it is required to account
for the constraints imposed by the kinematics of the bike and the contact with the ground. Motorbike
dynamics falls under the general class of constrained multibody mechanical systems. The theory for
this type of systems is well understood when the constraints are holonomic or semi-holonomic, Flannery
[2005]. The identification of such dynamics models follows well established principles. See for instance
Chessé and Bessonet [2001], Campion et al. [1996], McClamroch and Wang [1988] for structures with
holonomic constraints and Flannery [2005] for nonholonomic ones (classical texts on dynamics include
Arnold [1978], Bloch [2003]).
The generic dynamics model for a robot composed of multiple rigid bodies with n degrees of freedom
and m control inputs is of the form

M (q)q̈ + V (q, q̇) = Q + B + C (3.12)

where q ∈ Rn is the joint configuration vector, M (q) is the mass and inertia matrix and V (q, q̇) is a vector
containing the velocity terms imposed by the kinematic structure. Q, B, C are vectors of, respectively,
actuation variable (i.e., forces and torques generated by the actuators), dissipative forces acting on the
robot, and forces due to the contact between the robot and the environment, respectively. The dissipative
forces B are usually chosen as a combination of typical ad-hoc models (see for instance Olsson [1996]).
The geometric constraints that correspond to the contact between the wheels and ground induce the
contact forces C depend on the locomotion gait (classical examples are described in Bloch [2003]).
The common 2-wheel motorbike has two classes of constraints, (i) the contact with the ground which
constrains the vertical movement, and (ii) the car-like non-holonomic constraint, v = ψ̇r1 (which is also
commonly written as ẋ sin ψ − ẏ cos ψ = 0), that inhibits side-slipping (r1 represents the curve radius).
For the simplified kinematics model in section 2.2,

v = ψ̇r1 (3.13a)

v tan(δ cos γ)
ψ̇ = (3.13b)
w cos ϕ

In order to be included in the general model (3.12) the above constraints have to be expressed with the
coordinates used in the Lagrangian system (equation 3.9), that is ϕ and θ, and be of the form

 
ϕ̇
J =b (3.14)
θ̇

The C term in (3.12) can be shown to be of the form (see for instance Flannery [2005]),

C = JT λ

34
where the λ is a vector of Lagrange multipliers that represents the external forces (τext ) in the contact
points.

   
ϕ̈ Γϕ
M  + V = τ + τext =   + JT λ (3.15)
θ̈ Γθ

The solution of this differential-algebraic system can be obtained from,

  
ϕ ϕ̇
J   = withoutexceedingtheirvalue0 (3.16a)
θ θ̇
     
ϕ ϕ̈ ϕ ϕ̇
J   + J˙   =0 (3.16b)
θ θ̈ θ θ̇
 
ϕ̈
  = M −1 (−V + τ + τext ) = M −1 (−V + τ + J T λ) (3.16c)
θ̈
 
ϕ̇
λ = −J −T M J −1 J˙   − J −T (τ − V ) (3.16d)
θ̇

The model is now complete and it is held by two-equations system 3.15 added to equation 3.16d.
In a real motorbike the variety of situations rends difficult the identification of these constraints. For
example, the geometry of the contact between the tires and the ground imposes that there is indeed a
small amount of slide-slippage and hence the non side-slipping constraint is only valid under unrealistic
conditions.
In a sense, the above model is useful only when the constraints are well identified. The models in
chapter 2 have been shown to depend on multiple parameters and have a complex algebraic structure
and hence hardly fall in this class.

3.4.2 Simplified constrained dynamics

A common strategy when identifying constrained dynamics is to rely on a priori knowledge on the phys-
ical effects affecting the robots. This is by far the preferred strategy in the literature related to bike and
motorbike dynamics.
The τext is thus simplified and it will be considered the direct consequence of the reaction forces
applied on the center of the total mass. As it has already been used in Limbeer and Sharp [2006],
contact forces, which are calculated at each instant and depends on the global system state, are now
reduced to the sum of gravitational and centrifugal forces that the motorbike receives in the contact
points between the wheels and the floor referred to the center of mass.
These forces are calculated as,

N = mg (3.17)

35
T = mRr grθ̇ψ̇ (3.18)

where N is the vertical force due to gravity and depends on the total mass, m, and the gravity accelera-
tion, g. T is the centrifugal force due to ψ̇ rotating speed; in other words, the circular motion induces an
acceleration due to the variation of the forward velocity v,

d d
v = ψ̇r1 = ψ̈r1 + ψ̇ 2 r1 (3.19)
dt dt

The perpendicular component of this acceleration produces the centrifugal effect,

v
ψ̇ 2 r1 = ψ̇ 2 = v ψ̇ = Rr grθ̇ψ̇ (3.20)
ψ̇
These forces are applied ideally on the motorbike total mass center, and have to be related to ϕ as
torques,

Γϕext = N h sin ϕ − T h cos ϕ (3.21)

     
ϕ̈ Γϕ N h sin ϕ − T h cos ϕ
M  + V = τ + τext =  +  (3.22)
θ̈ Γθ 0

3.5 Implementation

The terms in (3.11) are in general highly complex. In fact, M and V are formed by derivatives of the
kinetic energy expression (3.7b) which depend on the complex kinematics equations, i.e. ψ̇ rotation
speed (equation 2.22), Rr radius (equation 2.20), h and b distances in equations (3.5a) and (3.5b) and
so on.
The implementation, in Matlab environment, has been thus divided into two different steps, (i) the
build of a Lagrangian based model, and (ii) the resolution of model’s equations after having specified the
simulation parameters. Diagram 3.3 explains graphically this procedure.

Figure 3.3: Blocs diagram of the Matlab implementation

36
3.5.1 Model building function

The first part will be accomplished with an m-file that calculates all terms in (3.10b) and (3.10c) starting
from the expressions of kinetic energy and kinematic. The structure of the m-file is very simple and
consists in five parts. Firstly all symbolic variables 1 and parameters are created. Secondly, the kinematic
expressions for h, b, ψ, Rr are formulated. Afterwards, kinetic energy K is entered following (3.7b).
Matlab calculates all kinetic energy derivatives that are used in M matrix and V vector, in (3.10b) and
(3.10c), using the symbolic language. Finally M and V are constructed and written in a text file that will
afterwards be used by the second m-file to simulate the model.

3.5.2 Simulation function

The second step is realized with the help of another m-file that simulates the behavior of the system mod-
eled by the matrix and vectors calculated in the first step. This function integrates the model equations
along consecutive time steps.

First of all the parameters values are introduced, such as the simulation time-step, the final time and
all masses, angles and lengths. The variables values for initial time (ϕ0 , θ0 , ϕ˙0 and θ̇0 ) are then entered
along with the torques (Γϕ Γθ ). The simulation can now start and ϕ¨0 and θ̈0 values are calculated for
each time step with following equation,

 
ϕ̈
  = M −1 (τ + τext − V ) (3.23)
θ̈

where M and V proceed from the model building function (Matlab language: M = model1; V = model2; ),
whereas τ and τext are formulated as external or contact forces functions. Afterwards, ϕ¨0 and θ̈0 values
are integrated wit the Euler numerical integration in order to obtain ϕ0 , θ0 , ϕ˙0 and θ̇0 , that will be used in
next time step calculations. After ending the simulation loop, the results are plotted in a graph.

3.6 Qualitative model validation

The dynamics model obtained from the equations previously developed and the Matlab functions of
3.5 must be validated before being used for control purposes. Several tests have thus been made,
encompassing different situations. The performance of the model will be then compared to that of a
real motorbike. In addition, the results will be commented against to what the intuition of a typical
rider suggests. This procedure permits to improve constantly the model and the Matlab functions, to
correct mistakes and to debug them. In this section the results are showed for diverse real and non-real
conditions of the motorbike riding.

1 Matlab language: variablename = sym(’variablename’);

37
3.6.1 No forward velocity

The first tests assess the behavior of the model in rest condition, that is without forward velocity. In
this case the motorbike behaves as a pendulum, that is a suspended mass who can rotate around
an axis. The pendulum rotates with ϕ leaning angle around the x axis, which is aligned with the line
defined by the contact points between the wheels and the floor. When the initial leaning angle is zero an
unstable equilibrium situation can be observed, which means that the system can maintain its position
until a perturbation appears (Figure 3.4(a)). When a torque is applied in ϕ direction, the motorbike falls
according to the sign of the torque (Figures 3.4(b) and 3.4(c)).

(a) Unstable equilibrium (b) Positive torque (10.74 Nm) (c) Negative torque (-10.74 Nm)

Figure 3.4: No initial leaning angle and no forward velocity. Leaning angle [◦ ] (red)

A small lean angle can act as a small perturbation in the unstable equilibrium system. As expected, the
motorbike falls while increasing initial leaning angle (Figures 3.5(a) and 3.5(b)). If the simulation time
is long enough, the oscillatory movement can be observed, as a common pendulum oscillating without
any energy dissipation. In fact, the initial leaning angle in Figure 3.5(c) is 30◦ degrees and the motorbike
oscillates between that position and 360 − 30 = 330◦ .

If the torque applied and the initial lean angle have opposite signs, the fall direction can change depend-
ing on the values of these variables as in Figure 3.6(a), where a negative initial inclination is compen-
sated by a positive torque. Figure 3.6(b) represents the simulation of a motorbike with a lower mass
center position. The oscillations show a smaller period than those of the real motorbike of Figure 3.5(c)
but preserve the amplitude, in accordance to the physics laws of the pendulum dynamics. Finally, the
system has been simulated without gravity, where the initial leaning angle of −30◦ does not change
during simulation time (Figure 3.6(c)).

38
(a) Positive initial leaning angle (1 de- (b) Negative initial leaning angle (-1 (c) Positive angle oscillations (30 de-
gree) degree) grees of initial leaning angle)

Figure 3.5: Initial lean angle without forward velocity. Leaning angle [◦ ] (red)

(a) Positive torque (21.48 Nm, -1 de- (b) Lower mass center (0.2907 m, -30 (c) No gravity conditions (-30 degrees
gree of initial leaning angle) degrees of initial leaning angle) of initial leaning angle)

Figure 3.6: Additional analysis with initial lean angle and without forward velocity. Leaning angle [◦ ]
(red); rotation speed [◦ /s] (blue)

3.6.2 Straight line driving

in this section the forward velocity is introduced, whereas the steering angle is maintained as zero. In
other words, the movement of the motorbike is simulated while following a straight line trajectory. The
simulation results show that the motorbike behaves still as a pendulum like in section 3.6.1. In fact,
Figures 3.7(a), 3.7(b) and Figure 3.7(c) are identical to Figures 3.4(a), 3.5(a), Figure 3.5(c) respectively,

39
apart from the forward velocity values.

(a) Unstable equilibrium with forward (b) Positive initial leaning angle (1 de- (c) Positive angle oscillations (30 de-
velocity (40 km/h) gree, 40 km/h of forward velocity) grees of initial leaning angle, 40 km/h
of forward velocity)

Figure 3.7: Straight line driving with and without initial lean angle. Leaning angle [◦ ] (red); forward
velocity [Km/h] (green)

The test in the absence of gravity shows that the initial leaning angle is maintained all simulation long
(Figure 3.8(a)). On the other side, a positive torque on the engine, produces a constant increase of the
forward velocity without any change on the other variables (Figure 3.8(b)).

3.6.3 Steering angle

δ steering angle is now added to the simulated system, so that the motorbike can execute curves. In
Figures 3.9(a) and (b) can be observed that a +/ − 1◦ steering angle makes the motorbike fall while
rotating around the vertical axis. For instance, a positive steering angle (to the right) produces a right
curve (positive ψ̇ rotation) while motorbike is falling due to the gravity torque, which is bigger than the
centrifugal one (equation 3.21). A negative steering angle has opposite effects. In Figure 3.9(c) the
equilibrium position is showed for 22m radius curve, 40km/h forward velocity, 3.785◦ and 30◦ . In that
situation, the sum of the torques is equal to zero, and that is way the motorbike can maintain its position.
However, this equilibrium position is unstable. Indeed, the motorbike tends to fall when any of the
other variables is disturbed. If the leaning angle increases, the gravity torque becomes bigger than
the centrifugal one and the motorbike will fall to the left, whereas when leaning angle decreases, it
will fall to the right because the centrifugal torque becomes bigger than the gravity one (Figure3.10(a)
and Figure3.10(b)). A new equilibrium for a bigger leaning angle is reached if the forward velocity is
increased, so that the centrifugal force can equalize the gravity one (Figure3.10(c)).
The unstable equilibrium is also broken when the steering angle is modified. An excessively big steering

40
(a) No gravity conditions (30 degrees (b) Positive engine torque (40 Nm, 1
of initial leaning angle, 40 km/h of for- degree of initial leaning angle, 40 km/h
ward velocity) of initial forward velocity)

Figure 3.8: Additional analysis of straight line driving. Leaning angle [◦ ] (red); forward velocity [Km/h]
(green)

(a) Positive steering angle (1 degree, (b) Negative steering angle (-1 degree, (c) Equilibrium angle (3.785 degrees of
40 km/h of forward velocity) 40 km/h of forward velocity) steering angle, 40 km/h of forward ve-
locity, 30 degrees of leaning angle)

Figure 3.9: Steering angle effects. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue)

angle will increase the rotation speed and thereafter the centrifugal forces, that’s why the motorbike will
fall to the left, whereas a smaller steering angle produces the opposite effect making motorbike fall to
the right (Figure 3.11(a) and Figure3.11(b)). A new equilibrium can be reached, in which the increase

41
(a) Positive leaning angle variation (31 (b) Negative leaning angle variation (c) New equilibrium with higher forward
degrees of leaning angle, 40 khm/h (29 degrees of leaning angle, 40 khm/h velocity (31 degrees of leaning angle,
of forward velocity, 3.785 degrees of of forward velocity, 3.785 degrees of 40.77 khm/h of forward velocity, 3.785
steering angle) steering angle) degrees of steering angle)

Figure 3.10: Leaning angle variation. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue)

of the centrifugal torque due to the bigger steering angle is compensated by a bigger leaning angle that
increase the gravity torque (Figure 3.11(c)).

(a) Positive steering angle variation (b) Negative steering angle variation (c) New equilibrium with bigger leaning
(30 degrees of leaning angle, 40 khm/h (30 degrees of leaning angle, 40 khm/h angle (31.44 degrees of leaning angle,
of forward velocity, 4 degrees of steer- of forward velocity, 3.5 degrees of 40 khm/h of forward velocity, 4 degrees
ing angle) steering angle) of steering angle)

Figure 3.11: Steering angle variation. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue)

42
The last variation considered that leads to an equilibrium loss is an increase or decrease of the forward
velocity, which is directly related with the centrifugal force and this is why a high velocity makes the
motorbike fall to the exterior of the curve (Figure 3.12(a)), whereas a lower velocity has the opposite
consequence (Figure 3.12(b)). When the velocity is higher, the centrifugal force can be reduced with a
smaller steering angle, which decrease ψ̇ rotation speed (Figure 3.12(c)).

(a) Positive forward velocity variation (b) Negative forward velocity variation (c) New equilibrium with bigger steer-
(30 degrees of leaning angle, 41 khm/h (30 degrees of leaning angle, 39 khm/h ing angle (30 degrees of leaning an-
of forward velocity, 3.785 degrees of of forward velocity, 3.785 degrees of gle, 41 khm/h of forward velocity, 3.6
steering angle) steering angle) degrees of steering angle)

Figure 3.12: Forward velocity variation. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation
speed [◦ /s] (blue)

3.6.4 Driver action

The driver can change the system behavior by applying external torques. For instance, the mass center
position can be modified by moving the body horizontally in y direction. When the mass center moves to
the interior of the curve the gravity torque increases so that the motorbike tends to fall in that direction
(Figure 3.13(a)). This effect can be corrected by an increase in the forward velocity (Figure 3.13(b)) or
the steering angle (Figure 3.13(c)), so that in both cases the centrifugal force increases.

Another important external torque that results from the action of a driver is applied with driver’s feet as
a force on the footpegs. When this torque is on the internal foot with respect to the curve, the motorbike
tends to fall in that direction (Figure 3.14(a)). As in previous case, this tendency can be compensated
with an increase of the centrifugal force by augmenting the forward velocity or the steering angle (Figure
3.14(b) and Figure 3.14(c)).

43
(a) Variation of driver’s mass position (b) Correction with higher forward ve- (c) Correction with bigger steering an-
(0.1 m, 30 degrees of leaning angle, locity (30 degrees of leaning angle, gle (30 degrees of leaning angle, 40
40 khm/h of forward velocity, 3.785 de- 41.9 khm/h of forward velocity, 3.785 khm/h of forward velocity, 4.16 degrees
grees of steering angle) degrees of steering angle) of steering angle)

Figure 3.13: Torque applied with variation of driver’s mass center position. Leaning angle [◦ ] (red);
forward velocity [Km/h] (green); rotation speed [◦ /s] (blue)

(a) Torque applied (10.74 Nm, 30 de- (b) Correction with higher forward ve- (c) Correction with bigger steering an-
grees of leaning angle, 40 khm/h of for- locity (30 degrees of leaning angle, gle (30 degrees of leaning angle, 40
ward velocity, 3.785 degrees of steer- 40.45 khm/h of forward velocity, 3.785 khm/h of forward velocity, 3.87 degrees
ing angle) degrees of steering angle) of steering angle)

Figure 3.14: Torque applied through the action of the driver’s feet. Leaning angle [◦ ] (red); forward
velocity [Km/h] (green); rotation speed [◦ /s] (blue)

44
3.6.5 Further tests

Some other tests have been made to validate the model. For example, an increase of engine speed due
to a constant engine torque, which makes the motorbike fall to the exterior of the curve because of the
bigger centrifugal force (Figure 3.15(a)). The simulation without gravity force and centrifugal force shows
the torque related to the gyroscopic effect (Figure 3.15(b)). This negative torque is due to the system’s
conservation of the angular moment. In fact, when the wheels and the engine are rotating in y direction
while motorbike is turning around z axis, a torque in x appears as results of the following equation:

Γgyroscopic = Ω × A Ω (3.24)

where Ω is the vector of the angular speeds and A is the inertia matrix.

(a) Positive engine torque (4 Nm, 30 (b) No gravity nor centrifugal forces (30
degrees of leaning angle, 40 khm/h of degrees of leaning angle, 40 khm/h
initial forward velocity, 3.785 degrees of of forward velocity, 3.785 degrees of
steering angle) steering angle)

Figure 3.15: Further analysis. Leaning angle [◦ ] (red); forward velocity [Km/h] (green); rotation speed
[◦ /s] (blue)

3.6.6 Simulink model

The Simulink model (Figure 3.16) has been built at this stage just to verify the m-file model.
The results have not been included in this chapter, because all of them were exactly identical to the
ones obtained from the m-file simulation. This fact completes the model validation, makes the dynamic
model even more reliable and gives more confidence to build a control system on this basis.

45
Figure 3.16: Simulink model

46
Chapter 4

Control

The objective of this chapter is to demonstrate the performance of the dynamics model developed in the
previous chapters under autonomous control that can mimic the behavior of a human rider. Humans ride
bikes, motorbikes, drive cars using a relatively simple set of principles namely, estimating information
on (i) the current state of the vehicle, (ii) the expected variation in the state, and (iii) the riding/driving
history. These have close relations with the well known PID structure.
The actuation points can be identified with the actions taken by a human rider. These are essentially
the steering angle (or torque applied to the handlebars), the torque applied through the force applied by
the feet on the footpegs, the mass center position, which can be influenced by the motion of the rider’s
body, and the engine torque. These are indeed the variables that any rider controls when riding and
hence the aforementioned argument suggests that PID based control is worth trying specially if it can
be compared with an actual ride directly controlled by a human.
Moreover, a simple proportional control of the forward acceleration is used based on modifying the
engine torque.

4.1 Theoretical basis

As aforementioned, the method chosen to govern the system is a PID control as it can emulate human
actions. In addition it has a relatively simple implementation.
Basically, there are two values that have to be controlled: ψ rotating angle and ϕ leaning inclination.
The first depends primarily on θ̇ engine speed and δ steering angle, whereas the second depends on
Γϕ feet torque, the mass center position, and ψ̇ rotation speed.
A human rider has available four actuation points. The first is the engine torque which, without
loosing generality can be assumed to be directly available to the driver to control engine speed, which
is strongly connected to the forward velocity. Braking actions can be also represented with the engine
torque variable1 . The second actuation point is the steering angle in the front fork, imposed by applying
a torque through the handlebar. The third point is the torque applied through the footpegs, used by riders
1 In fact, the current motorbike engines have excellent motor-braking properties.

47
to control the leaning angle. The fourth point is the mass center position variation on y axis, obtained by
changing the position of the body, namely the contact of the upper torso with the seat and the position
of the legs.
The Γϕ torque results from the sum of the torques induced by (i) the feet at the footpegs and (ii) the
change of mass center position. Though generated by two different driver movements, these reach the
same objective, namely, a torque on x axis direction.
The steering angle δ will therefore control the ψ rotating angle, meaning that the system will try to
follow a trajectory by knowing in which angle the motorbike should be in any time instant and modifying
the front fork angle in order to reach that position. The PID method is applied to minimize the error (ψ )
between the real ψ angle and the one that corresponds to the trajectory to be followed by the motorbike.

ψ = ψgoal − ψ (4.1a)
Z
v
ψgoal = dt (4.1b)
r1
Z
ψ = ψ̇dt (4.1c)
Z
d
δ = Pψ ψ + Dψ ψ + Iψ ψ dt (4.1d)
dt

On the other hand, the ϕ leaning angle is reached by adjusting Γϕ following the same scheme of δ
control. The error to be corrected is now the sum of the torque due to the centrifugal force, the one due
to the gravity force, and the gyroscopic ones, which are included in V (1) (the first element of the vector
V ).

ϕ = V (1) − Γϕext (4.2a)

∂K ∂2K ∂2K
V (1) = − + ϕ̇ + θ̇ (4.2b)
∂ϕ ∂ ϕ̇∂ϕ ∂ ϕ̇∂θ

Γϕext = N h sin ϕ − T h cos ϕ (4.2c)


Z
d
Γϕ = Pϕ ϕ + Dϕ ϕ + Iϕ ϕ dt (4.2d)
dt

4.2 PID implementation

To implement this PID control additional Matlab code was implemented. The δ and Γϕ are calculated at
the end of each time step, so that they can be used in next loop. Values for Pψ , Dψ , Iψ , Pϕ , Dϕ and Iϕ
are introduced at the beginning of the function and are constant during all simulation.
The results for the empirical control (Figure 4.1) show that with well-chosen P , D and I values, the
motorbike follows the reference trajectory. The criteria followed during the search of the adequate P ,
D and I are, on the one hand, a quantitative analysis of the actuation variables that must be able to
control the motorbike keeping the values within realistic bounds and, on the other hand, the quality of
the system response, namely by approaching the critically damped system. For example, when starting

48
with ϕ = 3◦ , δ = 0◦ , v = 50km/h, Pψ = 0.3, Dψ = 0.0001, Iψ = 0.05, Pϕ = 1, Dϕ = 0.1, and Iϕ = 0, the
motorbike can perform a curve of 100 meters radius.

(a) Controlled model (b) Controlled model with torque limitation

Figure 4.1: Autonomous driving during a 100m-radius curve. Leaning angle [◦ ] (red); rotation speed [◦ /s]
(blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg] (black)

In both cases, the transient for the kinematics variables ends approximately after 1.5 seconds, whereas
the driver stops applying the Γϕ torque after 3 seconds. Figure 4.1(b) shows that a limit of 20 · 9.81N
force has been imposed, because that is the approximate maximum that a human driver can actually
produce with a feet force and without a mass center position change. As expected, in this case the
system response is slower respect to the non-limited one but it eventually manages to reach equilibrium
position
More extreme driving conditions have been simulated and the results are plotted in Figure 4.2(a).
In this simulation, the forward velocity is 100km/h and the curve radius is 50m, whereas the control
constants are Pψ = 0.2, Dψ = 0.0005, Iψ = 0.08, Pϕ = 0.95, Dϕ = 0.1 and Iϕ = 80. Under these con-
ditions, the system manages to control the motorbike and approximately after 2 seconds the equilibrium
is reached with δ = 1.218◦ and ϕ = 53.37◦ .
Figure 4.2(b) shows the effects of a simulation under the same conditions but with the feet force
limited to 280 · 9.81N . This constraint corresponds indeed to the equivalent of the approximate maximum
effect that a typical rider can apply to the motorbike by moving the mass center position. As in the pre-
vious simulation (Figure 4.1), the system responds similarly to the non-limited apart from being slightly
slower.

49
(a) Controlled model (b) Controlled model with torque limitation

Figure 4.2: Autonomous driving during a 50m-radius curve. Leaning angle [◦ ] (red); forward velocity
[Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque
[Kg] (black)

Further analysis can be done in an even more realistic situation: the motorbike moves in a 55m-radius
curve at 70km/h while applying an engine torque Γθ of 20N m (Figure 4.3). This torque induces a forward
acceleration, and this is what actually happens when riding a motorbike in a curve. The forward velocity
v increases along with the leaning angle ϕ and rotating speed ψ̇, whereas a constant positive torque Γϕ
is applied by maintaining mass center position in the interior of the curve, exactly as it happens in real
human driving. The control constants are in this simulation Pψ = 0.2, Dψ = 0.0005, Iψ = 0.08, Pϕ = 0.9,
Dϕ = 0.15 and Iϕ = 65.

The system response can be modified by changing the values of the PID control constants. For instance,
the results of changes on ψ-δ control constants are showed in Figures 4.4, 4.5 and 4.6 in comparison to
the reference simulation of Figure 4.3.

Figure 4.4 shows how rapidly the system reaches the stationary state and its dependence on the propor-
tional constant of the steering angle control. In fact, when Pψ increases, the system reaches the required
δ angle faster and with less oscillations. Still, this rapid response has some negative consequences on
the leaning angle control, since the fast δ response causes a high initial peak in the Γϕ torque, due to
the sudden variation of the geometry of the system.

In Figure 4.5 the effects of modifying integral control constant are represented. By increasing Iψ , the
overshot of δ response also augment, having the same consequence on the Γϕ evolution.

50
Figure 4.3: Controlled model with acceleration while turning. Leaning angle [◦ ] (red); forward velocity
[Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque
[Kg] (black)

(a) Response with smaller Pψ (0.05) (b) Reference response (Pψ = 0.2) (c) Response with bigger Pψ (0.5)

Figure 4.4: System behavior when modifying Pψ . Leaning angle [◦ ] (red); forward velocity [Km/h] (green);
rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg] (black)

Finally, the derivative constant have been analyzed and the results show that variations of Iψ in the
range between 0 and the one which makes the system unstable do not modify substantially the response

51
(a) Response with smaller Iψ (0.01) (b) Reference response (Iψ = 0.08) (c) Response with bigger Iψ (0.25)

Figure 4.5: System behavior when modifying Iψ . Leaning angle [◦ ] (red); forward velocity [Km/h] (green);
rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg] (black)

(a) Response with smaller Dψ (0) (b) Reference response (c) Response with bigger Dψ
(Dψ = 0.0005) (0.00577)

Figure 4.6: System behavior when modifying Dψ . Leaning angle [◦ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg]
(black)

(Figure 4.6).

Figures 4.7, 4.8 and 4.9 shows the consequences of the changes on ϕ-Γϕ control, always compared
with the reference simulation in Figure 4.3.

A similar analysis can be made to understand the importance of the constants for ϕ leaning angle control.

52
(a) Response with smaller Pϕ (0.85) (b) Reference response (Pϕ = 0.9) (c) Response with bigger Pϕ (1.1)

Figure 4.7: System behavior when modifying Pϕ . Leaning angle [◦ ] (red); forward velocity [Km/h] (green);
rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg] (black)

(a) Response with smaller Iϕ (40) (b) Reference response (Iϕ = 65) (c) Response with bigger Iϕ (100)

Figure 4.8: System behavior when modifying Iϕ . Leaning angle [◦ ] (red); forward velocity [Km/h] (green);
rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg] (black)

Figure 4.7 explains that an excessively small Pϕ value leads to a constant increase of the torque that can
not manage to converge to an acceptable value, whereas, when Pϕ value is too big some oscillations
appear and, as a consequence, other oscillations are originated in δ response too.
The integrator constant must be big enough to avoid the rise of a fast increase of the torque (Figure
4.8(a)), which resembles the same problem that appears when using a small Pϕ (4.7(a)). Figure 4.8(c)
shows that when Iϕ increases, the negative peak reaches values that do not respect the normal riding

53
(a) Response with smaller Dϕ (0.08) (b) Reference response (Dϕ = 0.14) (c) Response with bigger Dϕ (0.3)

Figure 4.9: System behavior when modifying Dϕ . Leaning angle [◦ ] (red); forward velocity [Km/h]
(green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque [Kg]
(black)

patterns. For instance, changing the direction of Γϕ torque from positive to negative at the beginning of
the curve would be a waste of effort by the rider. Normally, a great positive torque is firstly applied and
then reduced, always in the positive direction, which corresponds to the interior of the curve.
Figure 4.9 demonstrates that the value of the derivative constant of ϕ control is strongly connected
to the response rapidity, which augments when Dϕ increases. A side effect of this improvement is the
growth of the initial peak of the applied torque.
Because of all these reasons, the six control constants have to be carefully chosen in order to avoid
the diverse negative effects that appears after having assigned both excessively small or excessively big
values.

54
Chapter 5

Results

5.1 Simulation background

The first three curves of the Montmeló circuit, located in the outskirts of Barcelona, Spain, have been
chosen to simulate the model in realistic conditions. The particular path to be followed by the motorbike,
shown in Figure 5.1, was obtained from telemetry data of a real racing motorbike. The control model can
thus be tested in fast and slow curves, left and right rotation, under braking and accelerating conditions.

Figure 5.1: Montmeló circuit sections

The simulation must analyze if the control strategies are able to make the motorbike follow the circuit
path by changing the values of control variables δ and Γϕ . The engine torque will be previously set
independently from the other variables in order to obtain fixed forward velocities that are normally used
in each section of the circuit (Figure 5.2 shows the theoretical forward velocity velocity for each circuit
section, whereas 5.3 shows the results for v according to the Matlab simulation function).
Some driving parameters have been simplified for this simulation, so that the analysis can focus on
the control when all boundary values are completely known. As shown in table 5.1, the path has been
divided in nine different sections, where the driving conditions are similar. This process permits to set en-

55
Figure 5.2: Speed evolution during first three curves of Montmeló circuit (telemetry data)

Figure 5.3: Speed evolution (km/h) through time (s) in Matlab simulation

gine torque, curve radius and time steps number for each circuit section during simulation. For instance,
in sections A, C and E the motorbike should try to follow a straight line trajectory while accelerating (A1,
C1 and E1) and successively braking (A2, C2, E2), whereas sections B, D and F represent the curves
where the motorbike is constantly accelerating.

56
Section r1 [m] v [km/h] time [s] ψ [◦ ]
A1 ∞ 250-265 1.185 0
A2 ∞ 265-73 0.295 0
B 45 73-94.3 2.700 0-90
C1 ∞ 94.3-98 0.466 90
C2 ∞ 98-76 0.237 90
D 45 76-93.4 2.250 90-22.78
E1 ∞ 93.4-106 1.603 22.78
E2 ∞ 106-95 0.122 22.78
F 121 95-201 3.605 22.78-111.45

Table 5.1: Proprieties of path’s sections

5.2 Empirical control

The circuit has been simulated with control constants that have been empirically found using a trial and
error strategy until acceptable results have been obtained. This strategy reflects in some sense the
driving learning by a human rider. In fact, by changing the control parameters for each circuit section,
the trajectory is gradually improved in order to make coincide the motorbike trajectory with the curves of
real path (Figure 5.4).

Figure 5.4: Montmeló curves and trajectory followed by the simulated model [m] (blue) and real motor-
bike [m] (red)

The evolution of the variables during the simulation is shown in Figure 5.5. One can see that Γϕ torque
has been limited to a correspondent feet force of 280 · 9.81N for the reasons explained in section 4.2.
The results of Figures 5.5 and 5.4 are reasonable and seem close to the real human riding. As

57
a matter of fact, the forward velocity behaves as expected as a consequence of the engine torque
externally imposed. The green line in graph 5.5 shows that v follows the data from telemetry (Figure
5.2), which are approximately real values.
The steering angle δ and rotation speed ψ̇, that are strongly related to each other, behave as a
second degree system reaching the final value after a delay time of approximately 1s. The inclination
angle ϕ has a smooth evolution when its value is modified, especially at the beginning of the curves
when the transition from the vertical position to a leaning angle is not as sudden as the behavior of δ and
ψ. The values for the permanent δ variable respect the kinematic equations, since the system search
the equilibrium position for a determined curve at a concrete speed. These values are approximately
1.6◦ degrees in the first two curves and 0.8◦ in the third faster curve.
The torque Γϕ appears when entering in the first curve and decreases gradually after having man-
aged to incline the motorbike. The torque does not disappear after having arrived to the required leaning
because the forward velocity is constantly increasing, as explained in section 4.1. A greater torque is
also needed after the curve in order to recover the vertical position, for instance after approximately 4.2s
or 7.4s. The ranges for Γϕ values during the permanent state are around −/ + 90N m. They reach the
280 · 9.81N force limit when changing ϕ leaning angle rapidly whereas they arrive to 200 · 9.81N when
leaning variation is slower.

Figure 5.5: Evolution of the main variables through time [s]. Leaning angle [◦ ] (red); forward velocity
[Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent mass for ϕ-torque
[Kg] (black)

When Γϕ is not limited new control constants must be found and the evolution of the relevant variables
is different from the one of the limited simulation (Figure 5.6). In this case, the torque is higher than
the correspondent force limit of 280 · 9.81N when the motorbike is leaning to the left in order to enter in

58
the ”D” curve. The fact that the torque can exceed the previous limitation modifies the whole dynamic
equilibrium, since the system can adapt faster to the required leaning angle.

Figure 5.6: Variables evolution through time [s] without Γϕ torque limitation. Leaning angle [◦ ] (red);
forward velocity [Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equivalent
mass for ϕ-torque [Kg] (black)

Another change that can be included in the system is the limiting ϕ leaning angle in order to avoid
the pneumatics sliding that appears when the real motorbikes lean excessively. In fact, the tires can
guarantee no sliding just until a certain motorbike inclination, which depends on the specific design of
the pneumatics. Due to the variability of this leaning range, which is not an object of this study, the model
has been limited to 60◦ , since this is normally a standard boundary value for the ϕ angle. The results
of Figure 5.7 show how after approximately 11s a torque variation appears and manages to find a new
equilibrium position when leaning angle reaches 60◦ and continues being restricted to this value.

59
Figure 5.7: Variables evolution through time [s] with 60◦ ϕ leaning angle limitation. Leaning angle [◦ ]
(red); forward velocity [Km/h] (green); rotation speed [◦ /s] (blue); steering angle [10x◦ ] (magenta); equiv-
alent mass for ϕ-torque [Kg] (black)

60
Chapter 6

Conclusions

During this study a great improvement in the motorcycle modeling has been made. The additional
features that have been incorporated in the kinematic model made possible to simulate it at a more
realistic level. The dynamic model served as a basis for the control analysis. The motorbike system has
been controlled with a simple strategy that nonetheless resembles that used by a typical human rider.
The model control has been finally tested in a real simulation and the results obtained suggest that the
model is valid.

6.1 Achievements

The major achievements of the thesis can be assessed on the one side to the kinematics and dynamic
equations that model the motorbike system, and on the other side to the results of the controlled model
simulations.
As a matter of fact, the kinematic equations (2.22a, 2.22b, 2.22, 2.22d and 2.22e) at he end of
chapter 2 summarize the work done in order to complete the existing models. For instance, the gradu-
ally incorporated improvements are the front fork inclination, the damping system with the consequent
variable front fork length and swingarm angle and the pneumatic shape effect on the real back wheel
radius. The implementation has been realized through Matlab, which allows to simulate the evolution
of the model in different conditions. The tests results reveal that, after having defined the front speed
and the suspensions loads, equilibrium position depends on the steering angle δ and leaning angle ϕ .
Two class of simulations have been analyzed: the search of the leaning angle once the steering one is
fixed, and vice versa. In both cases the results are realistic and reveal some interesting aspects. One
of this is the fact that the pneumatic shape causes a variation in the forward speed when the motorcycle
inclines. Moreover, changes in suspensions loads affect the equilibrium kinematic position. During hard
braking, the load on the front wheel increases whereas the one on the back wheel decreases and, as
a consequence, the geometry variation produces a change on the relation between leaning angle and
steering angle. The simulations show indeed that the leaning angle is smaller when braking, so that, in
that case, maintaining a curved trajectory would require a minor torque on the inclination direction as

61
dynamic analysis will explain.

Besides, a kinematic model for the motorbikes based on a double swingarm damping system has
been developed. The results of the different simulations, along with some empirical and intuitive con-
siderations, expose advantages and disadvantages of this particular kind of motorcycles. A benefit is
the fact that when the loads on the wheels change (i.e. when braking or accelerating) some variations
appear in the swingarms angles, which could compensate each other in terms of maintaining the wheel-
base length constant. In addition, tensions received by the system from the floor are not transmitted
directly to the driver’s hand, since there is no front fork and steering angle is independent form the
damping system. Contrary to these features, some maneuverability is lost due to the fact that the angle
of the front wheel is not directly controlled anymore, but a complex systems of special bars transmits the
movement of the handlebars to the wheel.

The dynamics model was obtained through the Lagrangian method based on the kinetic energy of
the system. After having calculated the free Lagrangian equation system, the model has been com-
pleted with the incorporation of the external forces due to the contact between the motorbike and the
floor. These forces have been firstly formulated following the exact algebraic strategy of constrained La-
grangian and then simplified in order to make them useful for a standard implementation and simulation
(the complete formulation is indeed complex). The model is implemented through two Matlab functions,
using to symbolic calculation, is able to operate with the complex elements of M and V that are different
derivatives of the total kinetic energy expression. The model is finally validated with several simulations
that tested diverse driving situations. These tests simulate the system with and without forward velocity,
straight and curve trajectories, turning right and left, applying and not applying torques, with and with-
out gravity forces and with different initial conditions. The results of the simulations show the validity of
the model and demonstrate how the equilibrium position is reached or lost by the system by modifying
the variables. Moreover, they explain the importance of each one of the dynamic and kinematic vari-
ables for the system and how they affect the behavior of the motorbike. The dynamics model has been
implemented with Simulink obtaining results identical to those obtained with the m.file functions.

The above model is unstable if there are no controls. The system by itself is indeed not capable of
maintain the equilibrium position for the driving situation. Therefore a PID control has been designed
by manipulating the input variable, i.e. the torque applied by in the direction of the leaning movement
and the steering angle of the front fork. The objective of the feedback control is twofold. On one side
to eliminate the difference between the torque due to the centrifugal effect, the gravitational force and
the gyroscopic one by applying a torque that corresponds to the change of the mass center position
produced by the movement of the driver’s body. On the other side the steering angle attempts to remove
the error that appears between the rotation of the bike end the trajectory to be followed. The main issue
of this strategy is the reach of the right values of the P,I and D control constants. The stability, the velocity
of the response, the oscillations and other effects depend indeed on the values of these constants. After
several trial and error tests, the adequate control has been found and acceptable results have been
obtained. This procedure, that reflects in some way the learning stage of a typical human rider and
provides a good indication on alternative strategies for control of a fully autonomous motorbike.

62
The final model, controlled with both strategies, has been finally tested on a real driving situation.
The behavior of the motorbike model has been simulated in first part of the circuit of Montmeló, where a
rich variety of driving conditions can be found.

6.2 Future Work

The arrival point of this thesis can be considered the starting one for several further developments.
The result of this study consists in the achievement of a realistic motorcycle model, its simulation and
assessment. Future work should extend the control implementation to all feasible situations encountered
in racetrack driving ensuring stability and performance requirements, namely in the fully autonomous
motorbike scenario.

63
64
Bibliography

V.I Arnold. Mathematical Methods of Classical Mechanics, volume 60 of Graduate Texts in Mathematics.
Springer, 1978.

Karl J. Aström, Richard E. Klein, and Anders Lennartsson. Bicycle dynamics and control. IEEE Control
System Magazine, August 2005.

Bikervoodoo. Honda unit pro-link. http://www.bikervoodoo.com/2010/08/06/hondas-unit-pro-link/, 2010.


[Online].

A.M. Bloch. Nonholonomic Mechanics and Control. Interdisciplinary Applied Mathematics. Springer,
2003.

G. Campion, G. Bastin, and B. D’Andrea-Novel. Structural Properties and Classification of Kinematic


and Dynamic Models of Wheeled Mobile Robots. IEEE Trans. on Robotics and Automation, 12(1),
February 1996.

S. Chessé and G. Bessonet. Optimal dynamics of constrained multibody systems. Application to bipedal
walking synthesis. In Procs of the IEEE Int. Conf. on Robotics and Automation, 2001. Seoul, Korea,
May 21-26.

M. R. Flannery. The enigma of nonholonomic constraints. American Journal of Physics, 73(3):265–272,


2005.

David J.N. Limbeer and Robin S. Sharp. Bicycles, motorcycles, and models. IEEE Control System
Magazine, October 2006.

N. H. McClamroch and D. Wang. Feedback Stabilization and Tracking of Constrained Robots. IEEE
Transactions on Automatic Control, 33(5):419–426, May 1988.

H. Olsson. Control Systems with Friction. PhD thesis, Dept. Automatic Control., Lund Institute of Tech-
nology, Lund, Sweden, 1996.

A.A. Popov, S. Rowell, and J.P. Meijaard. A review on motorcycle and rider modelling for steering control.
Vehicle System Dynamics, 48(6):775–792, 2010.

A.L. Schwab, J.P. Meijard, and J.M. Papadopoulus. Benchmark results on the linearized equations of
motion of an uncontrolled bicycle. in Proc. 2nd Asian Conf. Multibody Dynamics, August 2004.

65
R.S. Sharp, S. Evangelou, and D.J.N. Limebeer. Advances in the Modelling of Motorcycle Dynamics.
Multibody System Dynamics, (12):251–283, 2004.

Sportrider. 2004 honda cbr1000rr chassis tech.


http://www.sportrider.com/bikes/2004/146 04 honda cbr1000rr chassis/index.html, 2005. [Online].

Francesco Vitale and Gabriele VirzA Mariotti. Innovative progressive suspensions.


http://www.atnet.it/lista/casuni.htm, 2007. [Online].

66

Você também pode gostar