Characterisation and Utilisation of Steering Feel in Heavy Trucks Malte Rothhämel

Characterisation and Utilisation of Steering Feel in Heavy Trucks Malte Rothhämel
Characterisation and Utilisation of
Steering Feel in Heavy Trucks
Malte Rothhämel
Doctoral Thesis in Vehicle Engineering
TRITA-AVE 2013:21
ISSN 1651-7660
Postal address
Visiting address
Royal Institute of Technology Teknikringen 8
KTH Vehicle Dynamics
Stockholm
SE-100 44 Stockholm
Sweden
Telephone
Internet
+46 8 790 6000 www.ave.kth.se
+46 8 790 9304
Telefax
ATEX
Typeset in L
c
2013, Malte Rothhämel
Preface
This doctoral thesis has been written during my employment as industrial Ph.D. student at Scania R & D, Vehicle Dynamics, in collaboration with KTH Vehicle Dynamics.
This thesis is an extension and continuation of the previous work presented in the licentiate thesis "Capturing Steering Feel - A step towards implementation of active
steering in heavy vehicles" [1] by the author.
Reuse of text of this licentiate thesis
occurs and will not be labelled as citation.
I would like to express my gratitude to my supervisor at Scania, Jolle IJkema, for the
support provided during the project, with his encouragement and patience. Thanks
are also extended to my academical supervisor team, Annika Stensson Trigell and Lars
Drugge, for feedback and discussion, especially during the critical phases of this work.
Especially at Scania R & D I got a lot of help during this project. The steering group
that consisted next to the before named members also of Jon Andersson, Peter Karlson
and Magnus Juhlin.
I appreciated to have a wide freedom in a project but being
supported immediately when necessary. Additionally colleagues I would like to name
are Robert Skaba, Klas Bogsjö, Markus Agebro and Johannes Slettengren, Matthias
Andersson (ÅF), Joseph Ah-King and especially Tom Nyström and posthumously
Rickard Lyberger for extra support and a calm environment during turbulences.
I
also wish to thank Håkan Sehammar at VTI and Andreas Erséus, Adam Rehnberg,
Fredrik Svahn, Johannes Edrén and Daniel Wanner at KTH Vehicle Dynamics and
Olaf Jeschke at Lufthansa for their support during the course of this work.
My parents Ilse and Wulf and my brother Jörg and his wife Sophie, thanks for all
support and encouragement before and during this work.
I would like to thank my wife Mirja for her support, her patience and her love even
during the turbulent times of my thesis work.
Acknowledgement
The nancial support by Scania, KTH Vehicle Dynamics, IVSS (Intelligent Vehicle
Safety Systems Programme) and FFI (Strategic Vehicle Research and Innovation Programme) is gratefully acknowledged. Moreover, I would like to thank ebmPapst and
AUDI for their support.
Stockholm 13
th
May 2013
I
The important thing in science is not so much to obtain new facts as to discover new
ways of thinking about them.
William Lawrence Bragg
Abstract
Steering is next to braking the most important control feature of a road vehicle. The
driver gives input through the steering wheel and the vehicle reacts in a certain way,
which results in a transfer function between input and output. The transfer function
can be modied by means of an actuator in the steering system to improve safety,
handling and steering feel. To be able to use active steering systems to improve the
steering feel, we need to understand how steering feel comes about.
There have been several investigations to nd out how drivers experience a change in vehicle steering and handling behaviour and how a change in vehicle handling behaviour
aects the driver. However, as yet, there is no standardised way to nd mutual corresponding measurements, assessments and ratings, nor is there a consistent denition
of steering feel.
An important part of investigating steering feel concerns how to measure what drivers
feel. One of the essential prerequisites in the present research work is that steering
feel, as perceived by human beings, can be allotted in dimensions. To dene this noninstrumental space, a method to nd the dimensions that people use to perceive and
describe steering feel has been developed. It is shown that it is possible to extract up to
nine dimensions describing the steering feel of road vehicles. This was experimentally
evaluated using a driving simulator. In the test, drivers assessed truck steering system
settings that diered in friction, damping, inertia and stiness, due to ve dimensions
of steering feel.
The same steering system settings were also tested in accordance
with ISO standards for vehicle handling to acquire characteristic instrumental quantities.
The instrumental measurements and the non-instrumental assessments were
then analysed with respect to their correlation with each other. The results show that
there are indeed correlations and also which of the handling quantities inuence which
dimension of steering feel.
One possible use of the increased knowledge of steering feel evaluation is to inuence
the driver's behaviour by a directed change of steering feel. In a track test the steering
wheel torque of a truck was modied depending on the lateral dynamic vehicle driving
state. During the experiment the cornering behaviour of truck drivers was evaluated
regarding lateral acceleration, which is related to rollover accidents. Statistical evaluation showed a decrease of maximum lateral acceleration values while cornering when
the steering wheel torque was decreased at high lateral acceleration.
There are also more possibilities to modify the transfer function between driver input
and vehicle response. Articial understeering and yaw rate gain acceleration are two
functionalities that were developed during this work and were evaluated by simulation.
V
Thus, based on the knowledge about steering feel, an application-oriented hypothesis
could be formulated and evaluated. The fundamental part of this thesis contributed
a puzzle piece to the mapping of steering feel while the advanced part established ties
to future applications with active and semi-active steering as well as driver assistance
systems.
Keywords:
Steering feel, instrumental, non-instrumental, subjective, objective, vehi-
cle handling, driver behaviour, active steering, heavy vehicles, neural network, regression analysis
VI
Sammanfattning
Styrning är tillsammans med bromsarna de viktigaste kontrollelementen i ett vägfordon.
Föraren ger input via ratten och fordonet reagerar på ett visst sätt.
Det kan
sammanfattas till en överföringsfunktion mellan input och fordonsrespons. Överföringsfunktionen kan modieras med en aktuator i styrsystemet för att förbättra säkerhet,
handlingegenskaper och styrkänsla. För att kunna använda aktiv styrning för att förbättra styrkänslan, behöver vi förstå hur styrkänsla uppstår.
Det nns många undersökningar om hur förare upplever en förändring i fordonets
styrsystem eller handlingegenskaper och hur de ändringarna påverkar föraren. Men än
så länge nns det inte en enhetlig denition av, eller något standardiserat sätt, för att
detektera, mäta, beskriva och bedöma styrkänsla.
En viktig del inom forskningen om styrkänsla handlar om hur man mäter vad föraren
upplever. Ett av de essentiella antagandena i det här arbetet är att man kan dela in
styrkänsla i era dimensioner. Därför har en metod utvecklats för att deniera detta
icke-instrumentella rum och detektera de dimensioner som förarna använder för att
uppleva och beskriva styrkänsla.
I avhandlingen visas att det är möjligt att extra-
hera upp till nio dimensioner som beskriver styrkänsla hos vägfordon. Dimensionerna
utvärderades också experimentellt i en körsimulator. I experimentet beskrev förarna
med hjälp av de denierade dimensionerna styrkänslan hos fordon som karakteriserades
av olika styrsysteminställningar.
Samma inställningar utvärderades enligt ISO stan-
darder för fordonshandling för att få fram objektiva instrumentella mätvärden.
De
instrumentella och icke-instrumentella beskrivningarna analyserades med avseende på
statistiska samband. Resultaten visar att det nns samband och vilka handlingvärden
som påverkar vilken dimension av styrkänsla.
En möjlig användning av kunskapen kring styrkänsla är att kunna påverka förarens
körbeteende genom att modiera styrkänslan. I ett experiment modierades rattmomentet i en tung lastbil som funktion av sidaccelerationen. Under experimentet uppmättes förarnas kurvtagningsbeteende för en riskbedömning angående vältning. Den
statistiska utvärderingen visade att topparna i sidacceleration kapades när rattmomentet minskades vid hög sidacceleration.
Det nns många möjligheter att modiera överföringsfunktionen mellan förarens input och fordonets respons.
Articiell understyrning och girvinkelacceleration är två
funktioner som utvecklades och utvärderades inom detta arbete.
Baserad på den utökade kunskapen om styrkänsla kunde en användningsorienterad
hypotes formuleras och utvärderas. Den grundläggande delen av avhandlingen ger ett
VII
bidrag till kartläggningen av hur styrkänsla kan beskrivas och den tillämpade delen
etablerar anknytningar till framtida fordonsapplikationer med aktiva och semi-aktiva
styrsystem och förarassistanssystem.
VIII
List of publications
Publications included in this thesis
Paper A
Rothhämel, M., IJkema, J. & Drugge, L.: On correlation between steering feel and
handling in heavy trucks, F2008-02-047, FISITA World Automotive Congress, Munich,
Germany, 2008.
Contributions of authors:
Rothhämel designed the scenarios, recruited the drivers
and performed and supervised the test, did the analysis and wrote the paper. IJkema
provided useful ideas and supported the procedure. IJkema and Drugge supervised the
work, provided useful comments and proofread the paper. Rothhämel also presented
the paper at FISITA 2008.
Paper B
Rothhämel, M., IJkema, J. & Drugge, L.: On a method to generate a word pool for
the description of steering feel, Proceedings of the 10
th
International Symposium on
Advanced Vehicle Control (AVEC 10), Loughborough, UK, 2010.
Contributions of authors: Rothhämel designed the interrogation, recruited the interviewees and performed the interviews, did the analysis and wrote the paper. IJkema
provided useful ideas and supported the procedure. IJkema and Drugge supervised the
work, provided useful comments and proofread the paper. Rothhämel also presented
the paper at AVEC 2010.
Paper C
Rothhämel, M., IJkema, J. & Drugge, L.: A method to nd correlations between steer-
ing feel and vehicle handling data using a moving base driving simulator, Vehicle System Dynamics, 2011, Vol. 49(12), pp. 1837-1854.
Contributions of authors: Rothhämel and IJkema designed the scenarios, recruited the
drivers and performed and supervised the test, Rothhämel developed and performed
the analysis and wrote the paper. IJkema and Drugge supervised the work, provided
useful ideas, valuable comments and proofread the paper.
IX
Paper D
Rothhämel, M., Drugge, L. & IJkema, J.: Finding correlations between handling values
and the driver's performance using a moving base driving simulator, Proceedings of
nd
the 22
International Symposium on Dynamics of Vehicles on Roads and Tracks,
Manchester, UK, August 14 - 19, 2011.
Contributions of authors: Rothhämel and IJkema designed the scenarios, recruited the
drivers and performed and supervised the test, Rothhämel developed and performed
the analysis and wrote the paper. IJkema and Drugge supervised the work, provided
useful ideas, valuable comments and proofread the paper. Rothhämel also presented
the paper at IAVSD'11.
Paper E
Rothhämel, M., IJkema, J. & Drugge, L.: Finding correlation between steering feel
assessments and the drivers' performance using a moving base driving simulator, Proceedings of the FAST-zero'11 1
st
International Symposium on Future Active Safety
Technology towards zero trac accidents, Tokyo, Japan, September 5 - 9, 2011.
Contributions of authors: Rothhämel and IJkema designed the scenarios, recruited the
drivers and performed and supervised the test, Rothhämel developed and performed
the analysis and wrote the paper. IJkema and Drugge supervised the work, provided
useful ideas, valuable comments and proofread the paper. Rothhämel also presented
the paper at FAST-zero'11.
Paper F
Rothhämel, M., IJkema, J. & Drugge, L.: Inuencing driver chosen cornering speed
by means of modied steering feel, submitted for publication.
Contributions of authors: Rothhämel designed the scenarios, recruited the drivers and
performed and supervised the test, developed and performed the analysis and wrote
the paper. IJkema and Drugge supervised the work, provided useful ideas, valuable
comments and proofread the paper.
Paper G
Rothhämel, M., IJkema, J. & Drugge, L.:
Articial understeer by means of active
rd
steering an investigation of proper handling test methods, accepted for the 23
Inter-
national Symposium on Dynamics of Vehicles on Roads and Tracks, Qingdao, China,
August 19 - 23, 2013.
X
Contributions of authors:
Rothhämel and IJkema designed the model, Rothhämel
implemented the model and performed the simulation, developed and performed the
analysis and wrote the paper. IJkema and Drugge supervised the work, provided useful
ideas, valuable comments and proofread the paper. Rothhämel will present the paper
at IAVSD'13.
Publications referred to in this thesis but not appended
Patent
Rothhämel, M.: Method for replacement of two-circuit steering systems for heavy ve-
hicles by means of active steering with angle overlay (SE 534 469 C2).
Patent
Rothhämel, M.: Method for active steering with torque overlay for HPS systems with-
out an additional electric motor (SE 534 331 C2).
Pending patent
Rothhämel, M.: Method for articial understeering and acceleration of vehicle reaction
(1250406-4).
XI
XII
Contents
1 Introduction
1
1.1
Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.2
Research scope
4
1.3
Research approach
1.4
Reader's guide
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
2 Steering systems and steering parameters
11
2.1
Power steering system gain
. . . . . . . . . . . . . . . . . . . . . . . . .
11
2.2
Contemporary truck steering systems . . . . . . . . . . . . . . . . . . . .
11
2.3
Requirements for a truck steering system
12
2.4
Parameters describing the truck steering system . . . . . . . . . . . . . .
13
2.4.1
Steering ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
2.4.2
Friction
2.4.3
Damping
2.4.4
Inertia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
2.4.5
Servo characteristic . . . . . . . . . . . . . . . . . . . . . . . . . .
15
2.4.6
Stiness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
. . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
3 Steering feel and vehicle handling
3.1
3.2
3.3
3.4
19
Steering feel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.1.1
What is steering feel?
3.1.2
Human steering control
19
. . . . . . . . . . . . . . . . . . . . . . . .
20
. . . . . . . . . . . . . . . . . . . . . . .
21
Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
22
3.2.1
Steering torque feedback . . . . . . . . . . . . . . . . . . . . . . .
22
3.2.2
Correlation subjective/objective
. . . . . . . . . . . . . . . . . .
24
Driver command interpretation . . . . . . . . . . . . . . . . . . . . . . .
30
Feeling the steering and the vehicle . . . . . . . . . . . . . . . . . . . . .
32
3.4.1
32
Psychophysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.5
Non-instrumental quantities - Evaluation by the semantic dierential . .
34
3.6
Instrumental quantities
. . . . . . . . . . . . . . . . . . . . . . . . . . .
35
3.6.1
Quantities at the steering wheel . . . . . . . . . . . . . . . . . . .
36
3.6.2
Quantities at the vehicle . . . . . . . . . . . . . . . . . . . . . . .
36
3.6.3
Driver performance quantities . . . . . . . . . . . . . . . . . . . .
41
XIII
Contents
4 Asking for steering feel development of test methods
4.1
4.2
4.3
4.4
43
Preparation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Driver level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
4.1.2
Common language
. . . . . . . . . . . . . . . . . . . . . . . . . .
44
4.1.3
Questionnaire . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
4.1.4
Evaluation
45
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Multivariate Data Analysis
. . . . . . . . . . . . . . . . . . . . . . . . .
46
4.2.1
Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
46
4.2.2
Choice of method . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
4.2.3
Neural networks
. . . . . . . . . . . . . . . . . . . . . . . . . . .
Track test with three vehicles on low friction
50
4.3.1
Vehicles and test track . . . . . . . . . . . . . . . . . . . . . . . .
50
4.3.2
Test drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
50
4.3.3
Results
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulator experiment
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulator experiment planning
. . . . . . . . . . . . . . . . . . .
54
4.4.2
Test manoeuvre . . . . . . . . . . . . . . . . . . . . . . . . . . . .
54
4.4.3
Pre-test and simulator experiment
56
Denition
5.2
Superposition of steering angle
5.5
XIV
50
52
4.4.1
5.1
5.4
48
. . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . .
5 Active steering systems
5.3
43
4.1.1
57
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . .
57
57
5.2.1
Angle overlay by means of a planetary gearbox . . . . . . . . . .
58
5.2.2
Angle overlay by means of a harmonic drive . . . . . . . . . . . .
58
5.2.3
Angle overlay and steering ratio
59
. . . . . . . . . . . . . . . . . .
Superposition of steering torque . . . . . . . . . . . . . . . . . . . . . . .
60
5.3.1
Torque overlay by means of electric motor . . . . . . . . . . . . .
60
5.3.2
Torque overlay by means of electronically controlled hydraulics
Safety aspects for active steering
.
60
. . . . . . . . . . . . . . . . . . . . . .
61
5.4.1
Superposition of steering torque
5.4.2
Superposition of steering angle
. . . . . . . . . . . . . . . . . .
61
. . . . . . . . . . . . . . . . . . .
62
Active steering and driver assistance systems
. . . . . . . . . . . . . . .
62
5.5.1
Assisted and automatic obstacle avoidance
. . . . . . . . . . . .
62
5.5.2
Fail-safe-properties of active front steering systems . . . . . . . .
64
5.5.3
Inuencing driver behaviour by modied steering feel in buses . .
65
Contents
6 Prototypes and experiments for active steering evaluation
6.1
6.2
6.3
6.4
67
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
6.1.1
Superposition of steering angle (AFS)
. . . . . . . . . . . . . . .
67
6.1.2
Superposition of steering wheel torque (EPS/MDPS) . . . . . . .
68
Implementation and environment . . . . . . . . . . . . . . . . . . . . . .
68
6.2.1
ECU . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
68
6.2.2
AFS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
6.2.3
EPS
69
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Functionalities
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
69
6.3.1
Modied steering feel Rollover indication
. . . . . . . . . . . .
69
6.3.2
Articial understeering . . . . . . . . . . . . . . . . . . . . . . . .
71
6.3.3
Yaw rate gain acceleration - reducing the vehicle reaction time lag 71
6.3.4
Yaw rate gain acceleration with variable steering ratio . . . . . .
Safety concept
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
72
73
6.4.1
Superposition of torque
. . . . . . . . . . . . . . . . . . . . . . .
73
6.4.2
Superposition of angle . . . . . . . . . . . . . . . . . . . . . . . .
73
6.5
Virtual test environment . . . . . . . . . . . . . . . . . . . . . . . . . . .
74
6.6
Modied steering feel Track test
75
6.7
. . . . . . . . . . . . . . . . . . . . .
6.6.1
Test vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
6.6.2
Test drivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
6.6.3
Test track . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
76
6.6.4
Test procedure
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
77
6.6.5
Measurements
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
77
6.6.6
Data analysis methods Tests on dierent populations
. . . . .
77
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
7 Results papers in summary
83
7.1
Word Pool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
7.2
Correlation between handling and steering feel
7.3
Correlation between handling and driver performance
7.4
Correlation between steering feel assessment and driver performance
. .
87
7.5
Inuencing driver behaviour by means of steering feel
. . . . . . . . . .
88
7.6
Articial understeer handling test investigation . . . . . . . . . . . . . .
88
8 Scientic contribution
. . . . . . . . . . . . . .
. . . . . . . . . .
83
84
86
91
XV
Contents
9 Conclusions and recommendations to future work
93
9.1
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
93
9.2
Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
94
9.2.1
Word pool and extending the mapping of steering feel
. . . . . .
94
9.2.2
Modied steering feel
. . . . . . . . . . . . . . . . . . . . . . . .
94
9.2.3
Driver behaviour manipulation by means of superposition of
steering angle . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
95
9.2.4
Triggering the driver's reaction . . . . . . . . . . . . . . . . . . .
95
Bibliography
Appendix
97
105
Notation
105
List of abbreviations
106
Schematic diagram of hardware wiring for angle overlay
107
Appended Papers
111
A
On correlation between steering feel and handling in heavy trucks
111
B
On a method to generate a word pool for the description of
steering feel
121
C
A method to nd correlations between steering feel and vehicle
handling data using a moving base driving simulator
129
D
Finding correlation between handling values and the drivers'
performance using a moving base driving simulator
149
E
Correlation between steering feel assessments and the drivers'
performance using a moving base driving simulator
157
F
Inuencing driver chosen cornering speed by means of modied steering
feel
165
G
Finding proper methods for the investigation of articial understeer
by means of active steering
XVI
183
1 Introduction
This chapter gives an introduction to steering feel, human perception and the
correlation to instrumentally measured vehicle dynamics. The hypotheses will
be presented as well as the resulting research questions. The approach will also
lead to the methods that were used to answer the research questions guiding the
reader through the thesis and the appended papers.
1.1 Background
If one wants to investigate how a surface feels like, one would use the nger to sense
the structure. If one does not want to use the nger, one could use some object (e. g.
a pencil) to extend the own body and the own sense organ as well. Instead of feeling
where the sensor cells in the hand are situated, the person moves the situation of feeling
to the tip of the pencil. Of course, it depends on the characteristic of the pencil how
good in particular he can feel. A similar phenomenon do car drivers use. They replace
the pencil with the car's steering system to feel the tyre-road contact. How well the
driver can feel it, is on the one hand a matter of the driver's skills and experience and
on the other hand a matter of the properties of the steering system.
Steering is next to braking the most important control feature of a road vehicle. Right
from the beginning of automotive development the control of lateral position has been
a major focus of engineering (e. g. steering trapezium, tyre development, shimmy avoidance, ESC). Contemporary driver assistance systems exist for all areas of the drivers'
vehicle control - including steering.
Steering can be described as consisting of three levels: navigation, track following
2
stabilisation
1
and
and there are driver assistance systems for all of these levels: Navigation
systems, Lane Departure Warning or Lane Assist and ABS, ESC. Lane Assist and
ESC could be realised or supported by active steering. Navigation could be executed
by active steering as well which would be called autonomous driving. This is not the
topic of this work. However, if there is an actuator that makes the steering system an
active one, it will inuence the steering feel of the vehicle as soon as it is active and
probably also when it is inactive as well. To be able to use active steering systems to
improve the steering feel, there is a need to understand how steering feel comes into
existence. However, a cross-company standardised way of measuring steering feel has
1 also
2 also
called: guidance
called: control phases
1
1 Introduction
yet to be fully developed. Despite this, methods for measuring handling with regard
to road vehicles are well dened. Thus, the rst hypothesis of the present work is:
There exists a correlation between instrumentally measured handling values and
non-instrumentally measured human perceived values regarding steering feel.
Fig. 1.1 illustrates the concept of the hypothesis.
Assume two equal vehicles that
dier in only one or a few parameters, e. g. friction and inertia in the steering system.
These dierences will result in dierent handling values (see section 3.6) that
describe the lateral dynamic vehicle behaviour in an instrumental and objective way.
However, the behaviour of the vehicle enables the single driver to perform in a certain
way. So the changes in vehicle parameters propagate to subjective (meaning driverindividual) but instrumentally measurable driver performance values. In the same way
the changes in vehicle parameters will inuence the assessment of steering feel. These
non-instrumental quantities that are measured by test drivers will give an objective
description of steering feel. However, every driver has personal preferences. While the
assessment is assumed to be independent of the drivers, the ratings can dier quite
considerably. The most gurative example is a heavy and an easy steering feel. Both
drivers will agree which one is easier and which one is heavier - however some drivers
prefer the heavy steering while others prefer the easy one.
Objective
Subjective
Instrumental measurements
Char
Driver
performance




x,P = 

...






Char

Understeering gradient
Torque gradient
Response deadband
Bandwidth





...
Char
2



x,R = 



x = 






x,A = 

Char
Parameters

Figure 1.1:

Ratings
good - bad
good - bad
good - bad
good - bad
good - bad
Resistance :
Response1 :
Stability :
P lay :
Response2 :






Assessment
Handling
values




x,H = 

Char
SW angle changes
SW angle maximum
Time-to-collision
Unsymmetry
Non-instrumental measurements
F riction
Damping
Stif f ness
Inertia
Servo
heavy - easy
direct - indirect
stable - unstable
play - no play
slippery - not slippery
Resistance :
Response1 :
Stability :
P lay :
Response2 :












Illustration of interaction. Correlations are supposed between all blocks.
This model was developed during this work. In Paper C there is a simplied
version of this gure where there is no dierence between assessment and
rating.
1.1 Background
An analogous process is the description and gauging of an arbitrary object: the noninstrumental dimensions weight and size are obviously orthogonal.
Moreover, they
would correlate to the instrumental values mass and, for example, volume (see Fig. 1.2).
There are also more factors than only mass which will inuence the human perception
of weight, e. g. the handlebar design or ergonomic aspects. There may also be some
kind of saturation where a human being asserts that the object concerned is too heavy,
independent of the (excessively large) mass, which means a non-linearity in a linearly
experienced quantity.
heavy
Volume [m3 ]
Rating
Measurement
Correlation
light
Transformation
small
Figure 1.2:
Mass [kg]
large
Analogy for the correlation between non-instrumental ratings and instrumental measurements. The conscious interchange of the obvious correlating measurements and ratings on ordinate and abscissa demonstrates the
necessity for transformation.
In addition, the description of the object could be insucient and could then be
completed with, for example, the dimension softness (see Fig. 1.3), which must be
ft
so
light
h
a
rd
heavy
correlated to another instrumental characteristical quantity such as shore hardness.
small
Figure 1.3:
large
Analogy for non-instrumental ratings in three dimensions.
When transferring this complex of problems to instrumental and non-instrumental
measurements of the steering feel of road vehicles, this may obviously be even more
complex.
3
1 Introduction
Since human beings usually use words to describe the dimensions they experience (see
the words in Fig. 1.3), it might be possible to nd the dimensions that the words
describe, and their correlations to the instrumentally measured handling properties of
road vehicles. Thus, the second hypothesis for the present work is:
Steering feel can be described by a well dened word pool non-instrumentally
but objectively.
Finding and dening a word pool that distinguishes the dimensions and contents the
words that describe the dimensions, would result in a useful toolbox for several future
tests. Here, it should in the rst place help to map steering feel.
The logical next step is then to use this knowledge about steering feel for the investigation of active steering functionalities i. e. advanced driver assistance systems (ADAS)
that use active steering.
As early as 1991, Bösch stated that if the presentation of
information for the driver is manipulated or if manipulated information is oered articially, the manipulation will only be rated as progress if the ordinary driver is led
3
subconsciously to correct acts
[2].
This means that if advanced driver assistance
systems are to support the driver, they must give a good steering feel, and thereby
improve the driver's performance. Thus, the third hypothesis for the present work is:
Systematically modied steering feel can be utilised to induce the driver to
change his driving behaviour.
1.2 Research scope
The goal of this thesis is to nd a method to capture and map steering feel as well as
to use this information for the design of steering feel that results in a desired vehicle
experience.
This work has focussed on heavy semi-trailer-tractor combinations and
this results in the following aspects:
•
•
Mapping of steering feel of two-axled heavy tractors
How do drivers describe steering feel?
Into which dimensions can steering feel be divided?
Mapping of the correlations between instrumental and non-instrumental steering
feel values of two-axled heavy tractors
How can the correlations between instrumental and non-instrumental dimensions of steering feel be evaluated?
3 Original
citation: Wird an der Informationsdarbietung für den Fahrer etwas geändert bzw. werden
manipulierte Informationen künstlich angeboten, ist eine solche Informationsverzerrung nur dann
als Fortschritt zu betrachten, wenn der normale Fahrer ohne Training subbewusst zu richtigem
Handeln verführt wird.
4
1.3 Research approach
Which experimental methods are usable for nding correlations?
Which statistical analysis methods are usable to detect correlations?
This includes a strong focus on how people experience and especially describe and
express steering feel.
It also includes a study to evaluate the correlation between
handling quantities and measured steering feel.
Further aspects are:
•
Utilising articial steering feel by means of superposition of steering wheel torque
How can the driver be inuenced intuitively by active modied steering feel
for better (meaning safer) performance of the driver-vehicle combination?
•
Utilising articial steering feel by means of superposition of steering angle
How can the vehicle behaviour be designed to be more linear, meaning more
predictable?
Which handling test methods are usable?
The investigation for active manipulation of steering feel was performed for heavy vehicle applications dealing with active steering interventions for rollover prevention. Moreover, an initial study shows an articial linearisation of vehicle behaviour focussing on
understeering and appropriate measuring methods.
1.3 Research approach
The hypotheses and general research questions from the previous section lead to more
specic questions, to answers to which describe the methodology of this work. In the
following, the questions and the respective methods are presented, linked to the thesis
chapters and appended papers and in addition visualised in Fig. 1.4.
The rst hypothesis was There exists a correlation between instrumentally measured
handling values and non-instrumentally measured human perceived values regarding
steering feel.
Working with this hypothesis led to the following research questions (RQ) and methods
(MT) to approach the hypothesis:
RQ How to nd correlations?
MT Measure both, instrumental and non-instrumental, using data mining and
correlation methods.
If the interaction model illustrated in Fig. 1.1 is valid, it can be assumed that
correlations between the dierent blocks will be detectable (see Papers C,
D and E).
5
1 Introduction
Figure 1.4:
Process map with experiments, simulations and the resulting papers. LL
represents the lessons learned from each experiment which were input for
the subsequent investigations.
RQ How to measure instrumentally?
MT Using well established methods (ISO).
There are several standardised tests to identify vehicle dynamic properties.
The tests are described in ISO standards [36] and the quantities in section 3.6.
RQ How to measure non-instrumentally?
MT There will be need for a new method.
The methods found did not cover the present ideas of steering feel mapping;
therefore, a separate method was developed, which led to the following
questions:
RQ How can non-instrumental quantities be measured?
6
1.3 Research approach
MT Descriptive and Evaluative (see Fig. 1.1).
RQ Which drivers can do this?
MT Drivers who are used to assessing and describing (see subsection 4.1.1 and
Paper C).
RQ Which manoeuvres are appropriate?
MT Manoeuvres where the drivers have to change the steering input to feel
dierent states (see subsection 4.4.2 and Paper C).
RQ Which environment?
MT Track test and moving base driving simulator (see section 4.4 and Paper A
and C).
Track tests represent realistic conditions but oer less variability and hardly
inuenceable test conditions.
In contrast, the driving simulator oers a
constant environment, high variability as well as time saving because it
is independent of real test tracks, but it shows a lack of of reality.
So a
combination of both test facilities using the respective vigorousness seems
preferable.
RQ Which vehicle?
MT Tractor semi-trailer combination.
RQ How to correlate?
MT Multivariate data analysis methods (see section 4.2).
There are several statistical methods to detect correlations of data. In section 4.2 multivariate data analysis methods are presented and distinguished
where the choice of method depends on the data and the expected correlations.
The second hypothesis was Steering feel can non-instrumentally but objectively be de-
scribed by a well dened word pool.
Working with this hypothesis led to the following research questions and methods to
approach the hypothesis that are described in Paper B:
RQ How to nd appropriate words for the pool?
MT Word pool mining.
RQ How to nd out which words are part of the pool?
MT Word pool ltering I (Election process).
7
1 Introduction
RQ How to classify the words found?
MT Word pool ltering II and Statistics.
RQ How to make sure that people agree and use the words?
MT Pre-test and Simulator-experiment.
The third hypothesis was Systematically modied steering feel can be utilised to induce
the driver to change his driving behaviour.
Working with this hypothesis led to the following research questions and methods to
approach the hypothesis that are described in Paper F and G:
RQ How to change the steering feel to modify the driver's behaviour?
MT Change the usual transfer function between steering wheel input (angle
and/or torque) and vehicle response output.
RQ How to inuence the driver by means of active steering to decrease vehicle speed
during cornering?
MT Increase additionally or decrease the steering wheel torque or angle at high
lateral acceleration.
RQ How to test such a system?
MT Software in the loop and in a prototype vehicle on an isolated test track.
The software in the loop can be used early in the development process and
for the adjustment of parameters without any danger. Tests in a prototype
vehicle are the next step and should take place rst on an isolated test track.
Depending on the system and the degree of development, a hardware in the
loop test of the programmed controller can be recommended.
RQ How to measure driving behaviour?
MT By means of subjective instrumental measurements meaning driver performance quantities.
The choice of quantity depends on the experiment.
RQ How to evaluate a change in driving behaviour?
MT Evaluate measurements of systems and references statistically on dierences.
MT The evaluation may be driver-specic or track-specic (e. g. for each bend)
or for the whole population. The experiment design results in an appropriate statistical method e. g. the comparison of populations.
8
1.4 Reader's guide
A topic that is not described in detail in this thesis is the application work of providing CAN bus simulation for the active steering actuators.
Actuators from serial
production cars normally require information for safety reasons (e. g. ignition on or
vehicle rolling forward ). When transferring the actuator in another environment, the
necessary information must be simulated.
1.4 Reader's guide
Chapter 2 provides basic knowledge about truck steering systems. More information
in detail is given in chapter 3 including a literature review about steering feel, instrumental and non-instrumental quantities and their correlation and known challenges
with the interpretation of the driver's intention. Chapter 4 describes the experiments
on the track and in the simulator to correlate instrumental and non-instrumental
quantities of steering feel including the evaluation methods. Information about active
steering systems and their advantages and applications as well as previous research
results is provided in chapter 5.
In chapter 6 the developed functionalities and the
prototype are described as well as the driving experiment on a test track. This chapter
also includes the statistical analysis methods. Chapter 7 summarises the results and
the appended papers while chapter 8 summarises the scientic contributions of this
work. Finally, chapter 9 completes the thesis with conclusions and recommendations
for future research work.
9
1 Introduction
10
2 Steering systems and steering parameters
This chapter introduces contemporary truck steering systems with their requirements and the main parameters that are used and modied later on.
2.1 Power steering system gain
Steering of a road vehicle that is equipped with a steering wheel consists for the driver
mainly of the parameters steering wheel angle
δSW
and steering wheel torque
MSW .
Both of them are characterised by a gain factor. In a vehicle without any assisting
system (even without any power steering) the gain is only characterised by the steering
ratio which increases the torque and decreases the angle through the steering system
from the steering wheel to the steered wheels.
When adding power steering, the
torque gain will be dominated by the power steering's characteristic. Depending on
the principle of the power steering a more free choice of characteristics will be possible
than only a more or less linear gain. An example is an electrical power steering (EPS)
which can be programmed with a dependency of several input parameters.
Such a
system can be called active if it is able to apply a torque independent of the drivers
input e. g. to help the driver driving into a car park (without the driver's hands at
the steering wheel). This enables the system also to be more than an amplier and
add e. g. a negative torque to help the driver staying in the lane (Lane Keeping Assist
System).
In this case the gain would be negative.
Active steering systems will be
described more in detail in chapter 5.
2.2 Contemporary truck steering systems
The recirculating ball steering (see Fig. 2.1) is the state of the art steering system for
heavy trucks.
It is a variation on the older worm and sector design.
The steering
column turns a large screw (the worm gear) which meshes with a sector of a gear,
causing it to rotate about its axis as the worm gear is turned. An arm attached to the
axis of the sector moves the Pitman arm, which is connected to the steering linkage
and thus steers the wheels. The recirculating balls reduce the considerable friction by
placing large ball bearings between the teeth of the worm and those of the screw.
The servo assistance is realised by hydraulic pressure that is adapted to the driver's
input torque. This pressure results in an auxiliary force operating in the worm gear.
11
2 Steering systems and steering parameters
Figure 2.1:
Example of a contemporary hydraulic power steering system in a heavy
vehicle [7].
A torsion bar in the steering column provides the opening and closing of valves that
regulate the hydraulic ow. Torsion bar, valve geometry as well as hydraulic ow and
pressure create the characteristics of the power steering.
2.3 Requirements for a truck steering system
There are several requirements for the steering system of a vehicle and especially for
heavy vehicles. Generally, the steering system's task is to guarantee the lateral control
over the vehicle by the driver. This must be enabled independently of the vehicle load
and of the vehicle speed in spite of the fact that these parameters naturally inuence
the steering system's characteristics and behaviour. The control itself, however, is not
enough. It must be a safe control over time. Especially in a commercial vehicle, where
12
2.4 Parameters describing the truck steering system
drivers operate several hours every day, the ergonomics is an important part.
If a
vehicle for example is exhausting for the driver, (s)he will not be able to control it
safely at the end of the day. So safe control comprises a certain level of comfort. This
comfort means a distinct correlation between steering input and vehicle movement but
also a controlled, distinct (not wobbly) behaviour. It means an easy steering which,
with respect to necessary steering wheel angle and torque, does not demand too much
workload. Moreover, it means a sucient steering wheel return as well as a certain
feedback from tyre-road-contact.
Not driver related properties are that the steering system should ensure as little energy demand as possible. Further on, it shall be lightweight (with respect to energy
consumption of the vehicle) and should contribute to as little tyre wear as possible.
A lot of these properties are dicult to measure, therefore, it is dicult to set requirements and limits.
2.4 Parameters describing the truck steering system
This section describes the the steering system parameters of contemporary trucks that
are important for steering feel. The parameters are steering ratio, friction, damping,
inertia, servo characteristics and stiness.
2.4.1 Steering ratio
The steering ratio is the kinematic ratio between steering wheel angle and the mean
value of the front wheels' steering angles (see Equation 2.1). It is generally a mechanically xed ratio that can be non-linear (e. g. by a change of module in the steering
gear). When measuring the steering ratio the static or kinematic ratio must be differenced from the dynamic ratio. The static steering ratio describes the mechanical
correlation, the kinematics. This is only valid if the measured angles are large enough
(e. g.
δ = 20◦ ) and if the kinematics is linear.
A mechanical non-linearity, like a change
of module, must be taken into consideration.
i=
∆δSW
l
∆ δr +δ
2
(2.1)
The dynamic steering ratio takes also mechanical play and especially elasticity in the
steering system into consideration. This leads to an innite steering ratio in extremely
small intervals.
A plain case is a small steering wheel angle with a torque that is
smaller than the friction in the steering system.
The front wheels will not deect
while the steering wheel is deected. In Equation 2.1 the denominator will be equal
to zero while the numerator will be non-equal to zero.
13
2 Steering systems and steering parameters
The steering ratio inuences the steering feel by means of several reasons: It enforces
the driver's steering torque input and attenuates the experienced aligning torque of
the tyres.
It causes how friction in the steering system is perceived at the steering
wheel. Moreover, a too high steering ratio worsens the driving quality because of too
high required steering wheel angles that the driver cannot apply as fast as necessary.
By contrast, a too low steering ratio worsens the driving quality as well because the
vehicle becomes too sensitive at high speeds [8]. Further on, low steering ratio can lead
to a bad directional stability caused by the driver's inability to apply the, in this case
necessary, extremely small steering wheel angles.
Finally, especially heavy vehicles
need to prove that the vehicle is steerable even without power steering due to fail-safe
reasons, which demands a quite high ratio [9]. Thus, the choice of the steering ratio
is a compromise, which depends on the vehicle and its operation purpose.
Consequently, a non-linear steering ratio is common in a lot of production vehicles.
Generally the non-linearity is realized by a change of module over the rack in the
rack-and-pinion steering in the on-centre area which results in a change of the eective
lever arm. Another kind of non-linearity comes to existence by the linkage consisting
of Pitman-arm, steering drag and steering knuckle arm. However, this non-linearity
is at minimum around the center of the steering and becomes larger with increasing
steering wheel angle.
Sweatman and Joubert [10] found that drivers perceive a change in steering ratio in
the same way as they perceive a change in speed. This would result in a vehicle speed
depending steering ratio if the steering sensitivity shall be felt equal in all situations.
However, non-linearities are in ergonomics discussed controversially. The question in
this context to be asked is: Which system is (non-)linear?
A vehicle with a linear
static steering ratio will have a yaw rate gain that is not linear with respect to vehicle
speed.
Hence, it will be perceived as non-linear regarding vehicle speed.
However,
even a vehicle speed dependent steering ratio will have some kind of non-linearity (e. g.
the dynamic steering ratio on-centre). Moreover, the steering linkage between steering
servo and steering knuckle arm has a non-linearity resulting from the crank mechanism
Pitman-arm Steering drag Steering knuckle arm since Pitman-Arm and Steering
knuckle arm do not operate in the same plane. The author of this thesis assumes that
familiarisation and learning have a quite big inuence on what the driver likes and
which system that leads to higher performance and/or higher ratings.
The steering ratio is, when ignoring the servo assistance, also the ratio of driver input
torque i. e. steering wheel torque and the steering torque at the front wheels. However,
since the driver does not apply a torque but a force at the steering wheel, the steering
wheel radius is an important parameter, especially when transferring results from one
vehicle to another or when comparing results of dierent vehicles.
14
2.4 Parameters describing the truck steering system
2.4.2 Friction
The steering system of a road vehicle is built of several mechanical elements that are
joined to each other in a way to transform the rotation of the steering wheel to a
steering angle at the front wheels.
Between all steering system parts that perform
relative movement there will be friction.
By help of bearings and lubrication the
amount of friction is reduced to a low level.
The eect of friction in the steering system is bipartide. On the one hand does friction
lter small movements in the steering system and reduce vibrations that originate from
the road or from the engine. On the other hand does friction block small amplitude
signals from the front axle that could be used as information by the driver. It can be
concluded that the amount of friction is a compromise that has to be found for each
vehicle depending on target user group and other parameters like e. g. damping and
inertia.
2.4.3 Damping
Damping comes up in production steering systems by means of the damping properties of tyres and is therefore speed-dependent. In some vehicles there is need for an
extra damper in the steering system. In simulation or by means of force- or torquegenerating actuators damping can be programmed after measuring the steering angle
velocity. When operating a machine, damping is positively experienced when adjusting operating levers since small movements used to be done slowly and need therefore
less force while bigger movements used to be done faster and need therefore higher
forces. Whether these preferences are valid for steering systems of road vehicles is not
yet known.
2.4.4 Inertia
Inertia in the steering system is dominated by the steering wheel's inertia around the
vertical axis because of its dimension and weight distribution and the high steering
ratio (see Equation 2.2). Its eect is similar to friction (see subsection 2.4.2) but with
a dierent phase.
Itot =
Iwheel
+ ISW
i2
(2.2)
2.4.5 Servo characteristic
The servo characteristics inuences the steering feel, too. The servo amplies at rst
the driver's input torque (steering wheel torque). Depending on the characteristic this
15
2 Steering systems and steering parameters
amplication can be more or less linear. The servo characteristic is dependent on the
torsion bar's stiness and the valve geometry.
The torsion bar itself represents the
part with the lowest stiness in the whole steering system (see subsection 2.4.6) and
holds up the drivers steering wheel torque. The torsion bar's stiness together with
the valve geometry causes the opening and closing of the valves and consequently the
supporting pressure in the steering servo. The boost characteristic of the steering servo
is usually situated near zero with low steering wheel torque, increases rst slowly, then
faster and saturates at its possible maximum (see Fig. 2.2 red curve in graph).
In near future electromechanic actuators will oer a more free choice of the servo characteristic which enables a dependency of dierent parameters, e. g. vehicle speed (see
Fig. 2.2). Electromechanic actuators that perform the assisting steering torque without any hydraulic help have nowadays lack of power density for use in heavy vehicles.
However, in extreme driving situations such as e. g. dry-park (changing the steering
angle of the wheels while the vehicle is standing still) there is need for high power density. For heavy vehicles, with typical axle loads of eight tonnes, this is nowadays only
possible with hydraulics. Admittedly, the development of electromechanic actuators
for power steering continues for increasing axle load of the steered axle. However, a
combination of hydraulics and electromechanics (sometimes also called hybrid steering
- see section 5.3) oers wider possibilities.
Figure 2.2:
Example of a power steering characteristics (red curve in graph). The blue
and yellow curves show possible vehicle speed dependent variations [11].
2.4.6 Stiness
The stiness of the steering system is mainly dependent on the torsion bar in the
steering servo.
The torsion bar contributes to around 50% of the stiness in the
system. The stiness of the steering system inuences the dynamic steering ratio (see
16
2.4 Parameters describing the truck steering system
subsection 2.4.1) since, depending on the upholding torque, the driver's input torque
rst has to saturate the stiness before the front wheels begin to follow the steering
input.
However, the body stop around the torsion bar that is available for safety
reasons is not reached while driving normally. The stiness' inuence on the dynamic
steering ratio propagates to the understeer gradient (see section 3.6.2).
17
2 Steering systems and steering parameters
18
3 Steering feel and vehicle handling
One of the main hypotheses in this work states that there is a correlation between how instruments measure steering characteristics and how drivers assess
and rate the resulting steering feel. This chapter describes the denitions and
the background on related work and presents important results from earlier research work regarding both steering feel and human steering control.
3.1 Steering feel
Steering feel comes to existence during the interaction between driver and vehicle
which especially happens via the steering wheel, see Fig. 3.1. From the driver's point
of view the driver gives an output to the vehicle's steering wheel by means of his arms.
This output is realised by force on the steering wheel rim (which becomes a torque
around the steering column) and by distance of the steering wheel rim (which becomes
an angle around the steering column). The same values are the driver's input from the
vehicle's steering system: i. e. the steering wheel torque and the steering wheel angle
that the driver recognises.
Output
−
−−
−−
−−
→
←−
Input
Figure 3.1:
Interaction between driver and steering wheel.
Beyond that the driver perceives vehicle reactions that are expected to coincide with
the steering wheel quantities. These vehicle reactions are lateral acceleration, roll, yaw
rate and with certain limitations body slip angle.
19
3 Steering feel and vehicle handling
3.1.1 What is steering feel?
Steering feel is a myth wrote Setright in 1999 [12,13]. In contrary Aristoteles said (not
directly about steering feel but more generally): If touching is not a single perception
but a plural, then its objects are a plurality, too. Following Aristoteles and assuming
that touching (steering feel) is a plural perception, then it must be based on several
causes.
These causes might be the parameters mentioned in the previous chapter.
Obviously, Setright assumes steering feel to be a single perception and concentrates
on the feeling of the self-aligning torque. He concludes to plead the joystick without
any force-feedback to manipulate vehicle steering.
Dorsch [14] wrote that feeling cannot be dened, it can only be described since it is
not traceable back to anything.
Personal comments become manifested in feelings
1
depending on personal experience . This means that the description of steering feel is
dependent on each person since it is dependent on the personal experience. Further on
this may explain why car drivers that are not used to rate vehicle dynamic properties,
have diculties with repeatable ratings.
Following Dorsch' statement Paper B shows a method on how to describe steering
feel by arranging descriptive words in nine dimensions. These dimensions can be used
to measure steering feel, both subjective i. e. for each single driver, and objective i. e.
valid for all drivers. Objective steering feel is descriptive where more or less all drivers
are expected to agree to each other.
Subjective steering feel is evaluative including
own bias.
Every dimension of steering feel is like a colour: one can describe it and - independent
of the description - one can like it (or not). An example can illustrate the denition,
see Table 3.1.
Table 3.1:
Example for subjective and objective steering feel described in one
dimension.
Vehicle A
Driver 1
Driver 2
Vehicle B
heavy
easy
like it
do not like it
heavy
easy
do not like it
like it
objective
subjective
objective
subjective
In order to make the dierence clear the content of discussion is on steering feel, not
steering wheel feel. The latter could be dened as steering wheel haptics.
1 Translation
from the original citation: Gefühl lässt sich nicht denieren, sondern nur umschreiben,
da es sich auf nichts zurückführen lässt. In Gefühlen manifestieren sich persönliche Stellungnahmen
des einzelnen zu den Inhalten seines Erlebten. 20
3.1 Steering feel
For steering feel there is no standardised denition but below some possibilities on
how to describe it follows:
1. Steering feel can be dened as the combination of steering wheel angle and steering wheel torque, including their derivatives over time.
Steering wheel torque
will be perceived as a force depending on where the driver holds the steering
wheel.
2. Steering feel equates felt road contact.
3. Steering feel can also be dened according to Fig. 1.1. There the right column
labelled non-instrumental can stand for steering feel including both subjective
and objective quantities.
4. Steering feel is the perception of a complex sensation while steering a vehicle. Different sensing organs provide information beginning, of course, with the sensors
in the ngers and arms that return information about forces and positions. This
is dened by Wolf [15] as the common sense. But the cerebric compares this with
the information from the vestibular system and the visual information. Steering
feel is in this denition the relation between the perception of the vehicle's real
response and the driver's steering input with an expectation by the driver what
the vehicle is going to do.
This is dened by Wolf [15] as the extended sense.
The perception is assumed to correlate to the vehicle's reaction in consideration
of every driver feeling in a dierent way.
The latter denition (No. 4) is used in this work since drivers experience steering feel
not only as steering wheel angle and torque but also as the vehicle's response to their
steering input as to be seen in Paper B. The last but one denition is expected to
cover Wolf 's denition. However, while Wolf 's denition is from the theoretical point
of view, the visualised denition in Fig. 1.1 is made from the experimental point of
view.
3.1.2 Human steering control
Vehicle drivers perform the steering task in two dierent ways regarding control technique: closed-loop (with usage of feedback) and open-loop (without usage of feedback).
Wolf [15] concludes that the driver will steer predominantly open loop if he is highskilled and/or if he can look ahead over a long distance which means good weather
conditions and driving straight on, or in long drawn-out curves. By contrast the driver
will steer predominantly closed loop if he is unskilled and/or if he can look ahead only
over a short distance (like fog).
Open-loop means action while closed-loop control means re-action including a certain
time of waiting for the response and a certain human based reaction-time. This leads
to the conclusion that a good steering feel is related to short response time and a high
21
3 Steering feel and vehicle handling
percentage of open loop steering. In this case the driver feels safe in his driving situation. Open-loop control means that the driver knows what to do, knowing previously
how the result will be. Actually driving open-loop in this case means, of course, some
kind of feed-back-control, too, but only a verifying feed-back not a deviating feed-back.
This seems to contradict to the results of Neukum & Krüger [16].
They tested the
eect of theoretical malfunctions of an active front steering (AFS) system that superposed steering angle, and found that car drivers would not realise if an AFS-system
superposed a steering angle by mistake while cornering. Normally car drivers do not
realise this until leaving the curve when the steering wheel is still deected. However,
one reason for these contradictive theses could be that the steering angle dierences
analysed by Neukum & Krüger were not big enough to inuence the human steering
control. Another reason could be that car drivers are able to detect and quickly adapt
to the steering system's properties and can extrapolate for their open-loop control.
3.2 Previous work
There are several public presented research works about steering feel and vehicle dynamics.
Very comprehensive research works about steering feel were presented by
Buschardt [17] and by Barthenheier [18, 19].
Beyond that the following important
researchers who performed investigations with passenger cars, are summarised.
3.2.1 Steering torque feedback
Buschardt [17] mapped from 1998 to 2002 consequently the human perception of
changes in the steering wheel torque by means of psychophysical methods (see subsection 3.4.1). Based on these ndings he developed steering wheel torque feedback
models for use in steer-by-wire vehicles. The models were tested in slalom and double
lane change against conventional steering systems where Buschardt could show that
steering torque feedback based on yaw rate or lateral acceleration were rated as good
as contemporary conventional steering systems.
Barthenheier [19] investigated in his dissertation in 2004 the design of steering wheel
torque as a function of driver and driving situation. By measuring more than 20 passenger cars that were on the market at that time, he created a model to characterise
the torque of steering systems.
The model contained four parameters that were re-
turning torque, saturation of returning torque, damping and friction and two input
variables that were steering wheel angle and vehicle speed. The model is only valid for
the area of normal driving i. e. up to a limit of lateral acceleration
22
ay = 5 m/s2 .
Tests
3.2 Previous work
showed that even experienced drivers could not dier between real and synthetic steering wheel torque. In the validation experiment for the model subjects drove dierent
settings of the synthetic steering feel and evaluated the settings against each other.
The criteria were
•
•
•
experienced steering comfort
fun of driving
safety.
The parameter saturation did not inuence the evaluation. This was expected since
the manoeuvre in the experiment resulted in lateral acceleration up to
ay = 4 m/s2 ,
however, the saturation took place at higher lateral acceleration.
In the main experiment nine dierent settings were evaluated. The experiment was
carried out in three dierent manoeuvres that represented the three driving situations
of motorway, country road and urban trac. The criteria were
•
•
•
•
experienced steering comfort (empfundener Komfort)
experienced fun of driving (empfundene Sportlichkeit/Fahrspaÿ)
experienced safety (empfundene Sicherheit)
overall priority (allgemeine Bevorzugung).
The experiment was performed with more than 60 drivers. Anyhow, the evaluation of
the results was a challenge. First when taking age and gender of the drivers into consideration, the results became evaluable since some groups had opposite preferences.
A high steering damping decreases the experienced steering comfort but results in an
increased experienced safety.
The self-aligning torque has only a little inuence on
steering comfort (less than on safety).
On motorway and country road the friction
stands for the higher dissipative proportion while in urban trac the damping's proportion is higher. Younger drivers like less dissipative proportion regarding comfort
while older drivers prefer more dissipative proportion.
The experienced safety depends on the composition of the dierent torque proportions
friction, damping and self aligning torque.
However, the ratio of the proportions
depends of driving situation (motorway, country road, urban trac). The self-aligning
torque seems to be important for urban trac as well as motorway driving. For the
driving situation country road a medium high self-aligning torque gain was rated best.
The absolute torque was in the same range like in urban trac and motorway driving.
On the motorway self-aligning torque and friction lead to more experienced safety,
on country road driving the friction becomes even more important. Regarding safety
older drivers like higher damping while younger drivers prefer less damping.
Summarised, Barthenheier rejected the hypothesis that the steering wheel torque
should be a simple function of vehicle speed. Overall priority correlates with experienced safety and steering comfort while driving fun could not be correlated to another
23
3 Steering feel and vehicle handling
criteria.
However, regarding driving fun women associate low level steering wheel
torque with much fun of driving while men associate high level steering wheel torque
with fun of driving.
Schmidt [20] investigated directed steering wheel torque as haptic information. The
reason for using haptic signals is that the haptic perception channel is the fastest of
the human perception channels.
One of his research questions was When does the
driver feel anything at all? Schmidt compared steering wheel torque disturbances in
dierent situations:
•
Cornering
•
Narrow straight road
•
Lane change
In his work Schmidt found that a steering wheel torque signal is easier to detect if the
so called basic steering wheel activity is very low. While cornering (smooth curves) and
while driving on straight but narrow roads (e. g. road works area) the basic steering
wheel activity is quite low so changes in steering wheel torque can be felt very easy.
During a lane change the basic steering activity is quite high, and in addition the
steering wheel torque is non-uniform.
This results in that changes in the steering
wheel torque are harder to detect during a lane change for human beings. Moreover,
Schmidt found that a change of steering wheel torque is easier to detect for the driver
◦
if the steering wheel angle changes with at least 3 .
Authors comment: in a truck the basic steering wheel activity is assumed to be even
higher because of larger experienced road-induced disturbances which is caused by
the hard chassis suspension and the superposed cab suspension. This is expected to
impede the detection of information by means of steering wheel torque by the driver
compared to the situation in passenger cars. In summary the driver can detect steering
wheel torque signals easier when the vehicle is in steady-state on even road surface
since the low basic random noise facilitates the driver to detect. In addition, the driver
has better conditions to detect steering wheel torque signals if the driving task has a
certain diculty and is dened quite narrow.
3.2.2 Correlation subjective/objective
Deppermann [21] investigated in 1989 the subjective directional stability of passenger
cars. In the experimental part of his work he performed subjective and metrological
assessment of 9 test vehicles. The subjective assessment resulted in that the drivers
use at least two dierent quantities for the assessment of directional stability. The rst
24
3.2 Previous work
is the steering eort, the second is the steering feel at small steering movements. Moreover, the results from the open-loop-stability tests could be related to the subjective
ratings.
In the questionnaire he asked for both assessment and rating.
The answers were
to be given on a seven-step scale where four represented equality to the reference.
For the metrological assessment small steering wheel sine waves were the input to
δ = [−3, 3]◦ ,
frequency f = 0.15 Hz vehicle
vx = [0, 100] km/h). The hysteresis loops in the plots of steering angle against
steering torque at vx = 100 km/h were evaluated and showed friction in the range of
[0.41, 0.78] Nm.
the test vehicles (steering angle range
speed
In the evaluation Deppermann realized that when driving, the breakaway torque is
lower and the torque increase is higher. At vehicle speed
vx = 0
it is exactly the op-
posite, the breakaway torque is higher and the torque increase is lower. Moreover, the
torque consists of three parts: breakaway torque, static restoring torque and dynamic
self-aligning torque.
•
The breakaway torque is equal to the friction in the steering system
•
The static self-aligning torque is equal to the elasticities of the steering system
•
The dynamic self-aligning torque is equal to lateral forces at front wheels
Example: Vehicle F in Deppermanns investigation and the corresponding ranges over
all tests:
•
0.78 Nm
Breakaway torque:
•
Gradient:
•
Static self-aligning torque:
•
0.33 Nm/
◦
(Test range
(Test range
[0.41, 0.78] Nm)
[0.15, 0.44] Nm/◦ )
The gradient is equal to the
sum of static and dynamic self-aligning torque.
0.143 Nm/◦
Dynamic self-aligning torque:
0.187 Nm/◦
(calculated from gradient minus static
self-aligning torque)
Corresponding to Deppermann it was shown that a change of steering wheel angle is
easy to feel for the driver if the breakaway torque is low and the gradient is high which
corresponds to Weber's law [22] (see subsection 3.4.1).
The relative self-aligning torque is a value that makes dierent vehicles easier to compare. It is dened as the gradient of the (sum of the) self-aligning torque(s) and the
breakaway torque.
It shows the proportion of the self-aligning torque to the whole
necessary steering wheel torque, see Equation 3.1.
It is a quantity for the driver's
ability to perceive steering wheel angle.
In the same way a relative self-aligning torque can be calculated to nd the dynamic
self-aligning torque proportion of the whole necessary steering wheel torque.
This
25
3 Steering feel and vehicle handling
quantity is even more important to be high since it gives the actual information that
is related to the vehicle's lateral acceleration and yaw rate, see Equation 3.2.
ML,rel =
ML,stat + ML,dyn
ML,stat + ML,dyn
=
ML,0 + ML,stat + ML,dyn
ML
ML,rel,dyn =
ML,dyn
ML,dyn
=
ML,0 + ML,stat + ML,dyn
ML
(3.1)
(3.2)
For parameters, see Nomenclature
The characteristic of the single proportions of the steering wheel torque over steering
wheel angle visualised Deppermann as in Fig. 3.2.
MSW
ML,dyn
ML,stat
ML,0
δSW
Figure 3.2:
Components of the steering wheel torque, consisting of friction and static as
well as dynamic self-aligning torque, plotted over steering wheel angle [21].
Steering wheel angle and steering wheel torque were measured and the distribution was
analysed. The Kurtosis describes the shape of the distribution. The excess describes
the shape deviation from the Gauss distribution:
•
Negative excess in the steering wheel torque distribution is correlated to friction
•
Negative excess in the steering wheel angle distribution is correlated to play in
in the steering system.
the steering system.
From the evaluation of the measurements by means of power spectral density Deppermann concluded:
26
•
Steering frequencies over
•
Slow changes in steering wheel torque and angle have high amplitudes, while fast
1.5 Hz
do eectively not appear.
changes in steering wheel torque and angle have low amplitudes.
3.2 Previous work
Masaru Koide [23] evaluated steering feel by correlating describing words and handling
measurements. A glossary of 60 expressions was extracted from papers, questionnaires
and magazines and compressed to 10 evaluative expressions. These expressions were
used by evaluating 28 vehicles (passenger cars). These vehicles were measured regarding 47 combinations of steering eort, steering angle, lateral acceleration and yaw
velocity.
Koide found correlation between steering eort, steering hysteresis and the feeling of
solidity. Furthermore he found correlation between yaw rate gain, steering eort and
the feeling of smoothness.
Randwijk et al. [24] tested ve medium heavy trucks in loaded and unloaded condition.
On the one hand, they performed an assessment of steering diculty. 12 test drivers
drove an ISO double lane change and rated the dierent trucks according to the twolevel rating scale. On the other hand, they performed handling measurements in openand closed-loop manoeuvres.
Finally, a principal component analysis with the pa-
rameters of the open-loop manoeuvres indicated a correlation between measurements
and ratings.
High ratings, meaning easy experienced double lane change, correlate
according to Randwijk to modest steady-state yaw rate gain and lateral acceleration
gain, low equivalent yaw rate time constant, high bandwidths of yaw rate and lateral
acceleration transfer functions and low roll gradients.
Riedel et al. [25] measured nine vehicle settings with 15 non-professional test drivers
in a double lane change instrumentally and non-instrumentally. On the instrumental
side, they evaluated 33 characteristical quantities based on steering wheel angle, lateral
acceleration, yaw rate and body slip angle over time. For the ratings they used the
questionnaire that was developed for the double lane change manoeuvre, mentioned
by Zomotor [26]. In this questionnaire the driver rates eight questions in seven steps
which can be subdivided again and results in a number of points between 0 und 60.
To avoid the driver individual use of the rating scale they introduced a correcting
factor for each driver.
This factor was calculated as the ratio between the driver's
averaged ratings and the averaged ratings of all drivers.
In their tests they found
correlations between averaged ratings and a new elaborated quantity which is called
2
KD-value . This value consists of the sum of a dierence of two peaks of the steering
wheel angle and the time delay between steering wheel angle and lateral acceleration
(see Equation 3.3). The values are weighted with a ratio that was extracted from a
regression analysis.
2 The
KD = ∆δmax,2 + 2.5 · t0(δ,ay )
original German word is KD-Wert as abbrevation for
means Characteristic metric for double lane change
Kennwert Doppelter Spurwechsel
(3.3)
which
27
3 Steering feel and vehicle handling
The parameter variation for the nine vehicle settings realised Riedel by means of added
weight in the front, the rear, on the roof and by means of change of tyres.
manoeuvre was meant to be evaluated at two dierent speed levels: at
80 km/h
at the individual maximum speed, however, the latter was not evaluable.
The
and
A rst
evaluation of results showed that the drivers could not dierentiate suciently between
the eight dierent questions. There is the assumption that these non-professionals do
not have enough experience to rate the vehicle that dierentiated. Therefore, Riedel
evaluated all questions as one combined rating value. So one subjective rating judges
all the vehicle's subjective properties.
Lincke et al. [27] compared subjective judgements of non professional drivers with
typical vehicle characteristics and driving performance values in a driving simulator.
Subjective judgements were realised by a question and a given answer by two descriptions on the extremes of a ve-step-scale.
The questions (examples are here given
interpreted) were partially quite detailed:
Q: What do you think about the necessary angle of steering wheel?
A: Extremely large vs. Extremely small
Contrariwise there were quite comprehensive and abstract questions:
Q: What is the summary of your subjective judgement of the vehicle?
A: Car handling is extraordinary easy vs. Car handling is very dicult
Also metaphorical expressions were used:
Q: In which manner does the vehicle react to a sudden steering input?
A: Without overshooting vs. Overshoot several times
(Car goes like on rails vs. Car is skidding )
The questionnaire was only presented in extracts. The driving performance was characterised by time driven (measured in seconds) and penalty points. The latter described
how often the driver crossed the lane markings. There was no information about any
more characteristic values specifying driving performance in the work of Lincke.
Typical car characteristics were described by metrics of the vehicles response.
The
overshoot behaviour regarding yaw rate and body slip angle, the delay times and the
maximum and static value of the body slip angle were considered.
Lincke found correlation between the drivers summarising judgement of the vehicle
and the vehicle speed, the driving time, the distance from the centre line and the
number of penalty points that were given for crossing the lane markings.
Another
correlation seems to be between subjective judgement and the vehicle's natural yaw
frequency. In the investigated range of
0.4 Hz
up to
1.0 Hz
the higher frequencies led
to a better rating. Vehicles with the same natural yaw frequency were better rated if
the yaw damping was lower.
28
3.2 Previous work
Jaksch [28] investigated in 1979 the driver
vehicle interaction with focus on controllability. He used passenger cars and diversied the vehicles by means of extra load
and dierent tyre pressure.
Jaksch per-
formed several tests and simulations and
analysed the yaw rate response time as
the most important vehicle handling parameter for controllability. One of the test
results regarding response time was that
there is an optimum which is visible in
Fig. 3.3. The yaw rate response time covered an interval from
70 ms
up to
220 ms.
The best rating gave a response time of
about
100 ms.
A shorter response time
was connected to extreme understeering
which made the vehicle dicult to control
again. In addition, the steering gradients
increased in this case. Steering angle gra-
dδSW
day
dient
and steering torque gradient
dMSW
were important factors, too, after
day
the yaw rate response time. These two
gradients can be summarised to their product which is equal to steering eort. However, Jaksch expects an optimum vehicle
response in a steering angle gradient range
◦
◦
90 /g to 125 /g and a yaw rate response
time below 100 ms at a steering torque gradient of 30 Nm/g.
of
Figure 3.3:
Figure from Jaksch [28]
showing steering eort and
subjective rating versus
yaw rate response time t .
The optimum rating can be
seen around t = 100ms.
ψ̇
ψ̇
Crolla and Chen [29] presented in 1998 their investigation about comparisons between
rated and measured vehicle handling behaviour. After a pilot-study they performed a
track test with two vehicles where one was the reference vehicle while the other was
modied by means of eight parameters that were combined in a fractional factorial
experiment to 16 set-ups. Eight experienced test drivers rated the dierent settings in
relation to the reference. They were used to the language of the questionnaire about
vehicle handling that consisted of 49 structured and tied up questions. A 7-step ratingscale was used with the anchors better, same and worse as well as a don't know
option. For the experiment the manoeuvres steady-state circle, J-turn, and impulse
steer test were performed and 46 metrics measured. The evaluation was done by means
of multiple regression which resulted in a matrix of 27 questions correlating to certain
29
3 Steering feel and vehicle handling
of 28 metrics. However, not all of the mathematical correlations could be explained
physically. Moreover, in spite of the fact that the test drivers were professionals there
was a poor consistency of the drivers' ratings. Anyhow, Crolla and Chen could point
out that the frequency response metrics are very important for driver experienced
vehicle behaviour.
Agebro [30] performed a simulator experiment with 30 subjects. They rated 16 settings
consisting of four levels of steering ratio as well as four levels of steering eort in a 300m
course with multiple bends. The result was a complex of single results, therefore, it
was hard to draw general conclusions. However, Agebro showed a correlation between
the rated manoeuvrability and the performed o-tracking. Moreover, the evaluation
indicated that drivers were more sensitive to changes in the steering ratio than eort
when investigating manoeuvrability and lane keeping.
Another interesting nding
especially for further experiments is that high skilled drivers (according to Agebro's
denition) have less scatter in their ratings.
In another experiment with 18 subjects Agebro showed the inuence of the steering
properties kingpin torque, inertia, friction and damping on drivers' performance and
preference with respect to the drivers' skills and mapped the correlations. An increasing friction in the steering system was preferred with increasing vehicle speed.
In
the same manner drivers performed better regarding o-tracking so rating and driver
performance coincide with each other.
3.3 Driver command interpretation
There are some contents where it is hard to nd literature references but that are
topics of discussion anyway.
In this section some of these topics around the driver
command are shown to discuss their matter.
Driver command
The driver's commands are usually interpreted by the activation of steering wheel,
accelerator and brake pedal. This is important when driver assistant systems act on
their own. A quite old example is cruise control. The system works by itself as long
as there is no other driver command like switching o, switching o by the direct
driver command braking or overriding by means of accelerating. In the same manner
production solutions for lane keeping assisting systems do work. The system adds a
steering input (torque or angle controlled) that can be accepted or overridden by the
driver. The overriding by the driver is interpreted as driver wish or driver command.
However, there might be situations where the driver acts in a certain way which cannot
30
3.3 Driver command interpretation
be interpreted as a driver command. Though, this input might be unconscious or at
least not wanted. See some examples below.
Steering wheel angle on bumpy road
A quite common situation in a heavy truck that usually has a spring mounted cabin,
is steering wheel angle disturbances on bumpy roads. If the truck chassis (and with
it the truck cabin) gets excited single-edge by a road unevenness, the driver will be
shaken in lateral direction. The intuitive part to hold on is the steering wheel. Though,
the driver's wish is probably stabilising the truck, and the appropriate steering input
would be a straight on orientated steering wheel or counter steering. So the input at
the steering wheel in this moment is not the driver's request!
Accelerator pedal during autonomous braking
A similar phenomenon is the driver's action during autonomous braking. If a driver assistance system acts with an autonomous braking, the driver's foot will press down the
accelerator pedal because of the foot's inertia [31]. Anyway, when realising the trac
situation the driver might be convinced that braking was the best idea. Nevertheless,
the foot presses down the accelerator by means of inertia which could spuriously be
understood as an overriding of the autonomous braking process.
Steering wheel angle during ESC intervention
The steering wheel angle is the driver's input and is therefore often used as the drivers
request. There is no doubt that the driver will wish to turn right if he turns the steering
wheel to the right. So the common programming of ESC-systems makes absolute sense
to use the steering wheel input as an indirect quantity for the road curvature, since
we can assume that the driver tries to follow the road. However, there may come up
extreme situations like driving on snow where the driver gives the steering wheel angle
as input and the tyres saturate. The ESC system will operate at its limit and perform
the yaw rate that will be possible at the given speed and friction coecient. If the
driver requires even higher yaw rate by turning the steering wheel even more, it will
not be possible to full this requirement. The requirement could appear unconscious
since there is no feedback due to the input. The driver does not realise that he requires
more yaw rate since the yaw rate does not increase. When the friction coecient in
this situation changes rapidly to a higher value, the requirement is suddenly possible
to full. The front tyres will be more turned in than the cars real corresponding yaw
rate. This will boost the yaw rate suddenly. In addition, the ESC system will help
to increase the yaw rate even more if the tyres saturate on this higher friction surface,
31
3 Steering feel and vehicle handling
too, and the vehicle will shoot into the curve. This will cause a very dangerous case if
it is a left hand curve since the vehicle can suddenly cross the oncoming trac lane.
The described example is certainly a very seldom situation but it shows that the
steering wheel angle not always can be a quantity for the driver's wish of yaw rate
or direction, especially when driver assisting systems already intervene.
The longer
the intervention lasts, the more uncertain the steering wheel angle will represent the
driver's wish.
3.4 Feeling the steering and the vehicle
The steering system parameters mentioned in section 2.4 can be measured with instruments, partially directly, partially indirectly. The instrumental part of measurements
is well established and dened e. g. in ISO-standards and will be described in section 3.6. Nevertheless the non-instrumental part is not that well known. Therefore
it needs more focus in research. The aim is here to measure what people feel. Since
these measurements are to correlate with instrumental measurements later on, they
have to be standardised to get comparable results.
3.4.1 Psychophysics
The words proprioceptive, haptic tactile, vestibular and kinaesthetic are in dierent
literature used in dierent ways. Therefore it is important to explain the way these
words are interpreted here:
Haptic tactile
is the component of haptic perception to feel especially with the
ngertips the character of surfaces. Sensorcells in the skin detect mechanical eects
and deliver information about pressure and local velocity in the skin. In a car haptic
tactile is important for the steering wheel's haptics (surface) and in some way force
detection.
Kinaesthetic perception
(Greek kinesis, movement, and aesthesie, perception)
describes the system of some of these mechano-sensitive cells that are also placed in
joints and bres and deliver information about the position and forces of parts of the
body to each other.
The kinaesthetic perception let a human being know that e. g.
his arm is lifted independent of whether he has lifted it himself or the arm was lifted
by someone or something else. Regarding steering a vehicle kinaesthetic perception is
important for the driver's arm's position while steering and the detection of forces.
The
vestibular system is the sensory system that provides the dominant input about
3
movement and equilibrioception .
3 Sense
32
of balance
It is located in the ear and returns information
3.4 Feeling the steering and the vehicle
about longitudinal and rotational movements which are synchronised with the visual
feedback.
Regarding driving a car this is one way for driver and passengers to feel
lateral acceleration and it is the fastest way to feel yaw rate.
Proprioceptive perception
(lat. proprius, own, and recipere, reception) is the
combination of kinaesthetic, the vestibular system's and other senses' perception. In
the proprioceptive way e. g. lateral acceleration would be detected by the vestibular
system, by the haptic tactile of the chest feeling the seat, by the kinaesthetic perception
of the arms' joints and muscles feeling the inertia of the arms and by the optical
sensation of the environment moving relative to the own body.
Steering feel in terms of the extended sense seems to be that comprehensive that the
proprioceptive perception is assumed to be the relevant way of perception.
The
Just Noticeable Dierence (JND) is the smallest dierence between two stimuli
which a person can feel. Of course the JND can vary between human beings. However,
in psychophysics a lot of standardised situations were tested with numerous people
dening average values. The absolute threshold is a special case of the JND where the
comparative stimulus is equal to zero.
According to Weber's law
4
the smallest noticeable dierence in weight (that is equal to
force) is proportional to the starting value of the weight. That is to say, if the weight is
1 kg, an increase of 3 g
will not be noticed since the JND is an increase of
when the mass is decreased to
100 g,
an increase of
3g
30 g.
Rather,
in weight is perceived suddenly.
If the mass is doubled, the threshold (JND) will also double. However, Weber's law is
only valid in the medium range of perception. It is not valid at the limits of perception.
Buschardt [17] extracted from several publications that the JND for steering wheel
1.0 Nm to 1.2 Nm at a steering wheel diameter of
2.6 N to 3.2 N at the driver's hand). However, people
torque was located in the range of
38 cm
(which means a force of
dier about each other and the JND for a certain stimulus may dier from person to
person. Therefore there are dierent thresholds that are valid for 50% of the subjects
and 75% of the subjects. According to Buschardt [17] the JND50% for steering wheel
torque is located around
0.5 Nm
or
1.3 N,
the JND75% is located around
0.7 Nm
or
1.8 N.
These JNDs represent conscious recognition. However, there is unconscious recognition,
too. Buld et al. [32] showed that car drivers use haptic information that is lower than
the JND. Depending on the intention of an active steering system, a that low steering
wheel torque can be usefully utilised. In combination with a camera (like what is used
in a Lane Departure Warning system) a feeling similar to driving in moderate lane
grooves could be emulated. This could help the driver to steer the vehicle more often
in the middle of the lane while LDW or LKA systems only prevent the driver from
leaving the lane which could lead to a zick-zack-course.
4 k = ∆S
S
with
S
equal to Stimulus,
∆S
equal to Dierence of stimulus and
k
equal to the ratio that
is constant
33
3 Steering feel and vehicle handling
Above the JND drivers will recognise consciously.
An evaluation of these conscious
perceptions needs an own method since it is hardly possible to measure inside the
driver. Instead the driver will be used as measuring gauge but this includes that the
drivers' output-signals must be interpreted.
3.5 Non-instrumental quantities - Evaluation by means of the
semantic dierential method
People express what they feel in words so there must be keywords describing the focus
of perception.
A method to nd the words is described in Paper B developing the
semantic dierential for steering feel. With the assumption that people have dierent
dimensions of perception and feeling, dierent keywords for the description of nine
dimensions are the result of Paper B. These dimensions consist of several words that
describe nearly the same perception or exactly the opposite of each other. An example
may be the three words heavy, inertia and easy. The latter is the opposite of the rst
while the second is in the context of steering feel very close to the rst.
Often the words in a dimension can be divided into opposite words like heavy and easy
and their synonyms. If there are no antonyms as e. g. for steering wheel return there
is need for supplementary adverbs like much or little. The semantic dierential joins
the feeling to steps in-between the extremes. By means of evaluating the steps, the
perceptions can be expressed as numbers, as is exemplied in Fig. 3.4.
more easy
-3
-2
-1
0
1
2
3
more heavy
3
more heavy
7
more heavy
↑
equal to reference
more easy
3
2
1
0
1
2
↑
equal to reference
more easy
1
2
3
4
5
6
↑
equal to reference
Figure 3.4:
Evolution of a 7-step ordinal scale for evaluation with the semantic dierential method. The rst owns risk for negative associations, the second for
misunderstandings, the third was used in Paper C.
The Semantic Dierential is a method developed by Charles E. Osgood [33] to measure
the connotative meaning of objects, events, and concepts. There is no use of direct
questions like What do you think of ...? but of indirect questions where people are
34
3.6 Instrumental quantities
asked for their rating how much they connote certain words with an object, event or
concept.
The semantic dierential for steering feel is on the one hand an important word pool enabling the subjective distinguished evaluation of the driver's perception (see Table 3.2).
On the other hand it is a tool that shows which part of the perception is important for
the driver. With known correlation to handling properties this will lead to parameters
that inuence steering feel which is important to know for the improvement of vehicles.
Table 3.2:
Identied non-instrumental dimensions of steering feel and related words
(Paper B). Since the interrogation was made in Swedish the words in the
table are only interpretations.
Dimensions
Related words
Stability
(un)stable
rate-stable
directionally stable
wobbly
sensitive to lane-grooves
Response
controlled
delayed
(in)direct
distinct
erratic
obedient
quick
reactive
precise
(in)exact
sensitive
slippery
Play
play
Resistance
heavy / easy
inertial
light as a feather
force-requiring
Ratio
large / small steering wheel angle required
Road feel
road feel
road contact
Jerk
jerky
pulsing
(Comfort)
shaky
stabbing
steering wheel jerk
vibrating
Steering wheel return
steering wheel return
3.6 Instrumental quantities
Measurements that are performed by measuring equipment have the advantage of high
repeatability, independency of the individual driver and less complexity and eort than
experiments with test drivers as sensors. In this work, standardised vehicle handling
tests according to ISO 13674-2 (Transition) [5], ISO 13674-1 (Weave) [6], ISO 14792
(Circle) [4] and ISO/TR 8726 (Random) [3] were performed.
35
3 Steering feel and vehicle handling
3.6.1 Quantities at the steering wheel
At the steering wheel there are two values measurable:
•
Steering wheel torque
•
Steering wheel angle
MSW
δSW
Based on this the derivatives with respect to time can be calculated, too. However,
the driver generates force with his arms - not torque. Therefore, the steering wheel
radius
rSW
is an important factor. Moreover, generated forces are only comparable if
the steering wheel angles are small. Otherwise the driver's arms' movements become
more complex using dierent muscles in dierent phases of the movement. This may
result in non-linear forces.
According to Wolf [15] steering wheel angle and torque
describe the central steering feel.
3.6.2 Quantities at the vehicle
By contrast the extended sense comes to existence by combining the steering wheel
torque and angle with the vehicle's response values that are:
•
lateral acceleration
•
yaw rate
•
body slip angle
•
roll angle
ay
ωz
β
α
Combining these leads to characteristical values, see Table 3.3. Vehicle speed is not
involved directly in any of the the quantities but is an important conditional value
that usually is constant but not equal to zero in these tests.
Torque gradient
(Transition test)
dMSW
describes the change of steering wheel torque as result of
day
a certain change in lateral acceleration in a ramp steer, which is called transition test.
The torque gradient
Response deadband
Response deadband
MSW
at
◦
ω = 0 /s
is a dierence of torque when the yaw rate is
equal to zero when rising the steering wheel angle from zero in a uniform slalom (see
Fig. 3.5(a)).
It describes the sum of all resistances that the steering system has to
overcome before the vehicle shows a response.
36
3.6 Instrumental quantities
Table 3.3:
Test type
Transition
test [5]
Weave
test [6]
Vehicle-dependent instrumental measurements: Handling values from
Paper C.
Instrumental measurement
Value
Unit
Torque gradient
dMSW
day
Nm/ m
@
ω = 0 /s
ay = 0 m/s2
ay = 0 m/s2
Nm
m
s2 /◦
ay
@
MSW = 0 Nm
m/s2
Torque gradient
∂MSW
∂ay
@
ay = 0 m/s2
Understeer gradient
Kus
Response deadband
Steering sensitivity
Steering angle deadband
Lateral acceleration
hysteresis
Circle
test [4]
Body slip angle gradient
Steady-state yaw rate gain
Torque magnitude
Random
test [3]
s2
MSW
@
δSW
◦
∂ay
∂δSW
@
s2
s2
1/s
@
dα
day
Bandwidth
f
Peak gain
∂ωz
∂δH
Peak gain relative to
∂ωz
∂ωz
∂δH / ∂δH
Time delay at
fδSW = 0.5 Hz
∆t
s2
◦/ m
Roll angle gradient
Steady-state gain
Nm/ m
◦/ m
dβR
day
dωz
dδSW
MSW
◦
ay = const.
Nm
◦/ m
s2
Hz
◦/ s
◦
1
ms
Steering sensitivity
∂ay
m 2
∂δSW at ay = 0 /s describes the change of lateral acceleration
as response to a certain change of steering wheel angle when rising the steering wheel
Steering sensitivity
angle from zero in a uniform slalom (see Fig. 3.5(b)).
It describes how strong the
vehicle reacts on a steering input.
Steering angle deadband
Steering angle deadband
δSW
at
ay = 0 m/s2
describes how much steering wheel angle
is necessary to get a vehicle response in lateral acceleration. Summing up the absolutes
of these necessary steering wheel angles is assumed to result in some feeling like play
in the steering system.
37
3 Steering feel and vehicle handling
6
Weave test measurement
2
ωz [◦/s]
Weave test measurement
ay [m/s2 ]
1.5
4
1
0.5
ay [m/s2]
wz [°/s]
2
0
0
−0.5
−2
−1
−4
−6
−5
−1.5
MSW [Nm]
−4
−3
−2
−1
0
M SW [Nm]
1
2
3
4
−2
−50
δSW [◦ ]
−40
−30
−20
−10
0
δ SW [grad]
10
20
30
40
(a) Yaw rate against steering wheel torque.
(b) Lateral acceleration against steering wheel an-
The arrow visualises the response deadband.
gle. The red lines visualise the steering sensitivity,
the green arrow visualises the steering angle deadband.
Figure 3.5:
Weave test measurement.
Lateral acceleration hysteresis
Lateral acceleration hysteresis
ay
at
MSW = 0 Nm
describes the remaining lateral
acceleration when the steering wheel torque crosses zero in a uniform slalom. Summing
up the absolutes of this remaining lateral acceleration is assumed to result in some
feeling like returnability of the steering system.
Torque gradient
(Weave test)
∂MSW
at ay = 0 m/s2 describes the change of steering wheel torque as
∂ay
a result of a change of lateral acceleration when the latter crosses zero in a uniform
Torque gradient
slalom which is called weave test.
Understeer gradient
Understeer gradient
Kus describes the dierence between the theoretic (static) steering
angle and the real (steady-state) steering wheel angle at a certain speed at constant
radius cornering (see Equation 3.4). The latter equals to a certain lateral acceleration.
The gradient is visualised in Fig. 3.6 with the red derivative triangle. The lines on the
right side of the diagram show the theoretical correlation between Ackermann steering
angle
δ0 ,
vehicle speed
vx
and lateral acceleration
ay
where the former is calculated
according to Equation 3.5. Driving the same circle with a higher vehicle speed results
38
3.6 Instrumental quantities
in a higher lateral acceleration. The real steering wheel angle
δSW
is normalised by
the static steering ratio (see subsection 2.4.1). Then the dierence between theoretic
steering wheel angle and real steering angle over lateral acceleration is equal to the
understeer gradient. Chassis elasticity is included in the real steering wheel angle and
is therefore also included in the understeer gradient. A vehicle is understeered if the
steering wheel gradient
∂δ
∂ay
> 0.
ay
vx,4
vx,3
Kus =
∂( δSW
i −
∂ay
L
R)
vx,2
L
ay
δ0 =
=L 2
R
vx
δSW0 =
(3.4)
δ0
i
δSW
(3.5)
vx,1
(3.6)
(δSW − δSW0 )
Figure 3.6:
δSW0
Handling diagram for a heavy
truck. Understeer gradient illustrated in red. Vehicle speed isolines on the right.
Figure 1: Active steering functionality in a simulation environme
Body slip angle gradient
dβR
day describes the change of body slip angle at the rear
axle caused by a certain change of lateral acceleration. This quantity gives information
The body slip angle gradient
about the absolute stiness in wheels and chassis.
Referring back, the understeer
gradient is a quantity describing the balance between body slip angles at the front and
at the rear axle. A low body slip angle usually gives a more safe feeling while driving.
However, an increase with lateral acceleration is expected and indicates the vehicle's
stability limit.
Yaw rate gain
dωz
dδSW describes the change of yaw rate as a result of a
change of steering wheel angle. It is a factor that describes the enhancement of the
Steady-state yaw rate gain
39
3 Steering feel and vehicle handling
vehicle's reaction. A very low yaw rate gain means a very weak vehicle reaction and
will be experienced as dead. A very high yaw rate gain means a very strong vehicle
reaction and will be experienced as nervous.
A good solution must be in-between
these two extremes.
Torque magnitude
Torque magnitude
MSW
at a constant lateral acceleration
ay
is a standardised torque
measure. This quantity is dependent on lateral force and therewith of vehicle speed
and radius, thus the vehicle speed is standardised at
70 km/h
for all of these ISO-tests.
Roll angle gradient
dα
day describes a change of chassis roll angle as result of a change
in lateral acceleration. It is measured as a dierence of the chassis suspension to the
The roll angle gradient
right and to the left. This dierence is calculated for the front axle and the rear axle
each. By means of a gyrometer it can be measured at any other location at the chassis
as well. Especially in a truck this quantity does not describe the roll angle the driver
experiences since the cabin has its own suspension which results in an additional roll
angle.
Bandwidth
Bandwidth is the frequency band
between
0 Hz
and the frequency
where yaw rate gain falls below
90% of its value at
0.2 Hz
(see
dω
dδSW
angle input, the vehicle's response
Ratio
Fig. 3.7). It is a quantity for that
when giving faster steering wheel
◦
[ s /◦ ]
Peak gain
100%
90%
decreases signicantly.
Peak gain
Bandwidth
0.2Hz
Peak gain is the maximum value
of yaw rate gain as a function of
steering wheel angle frequency (see
Fig. 3.7) measured in ISO-random
test.
40
Figure 3.7:
fδSW [Hz]
Yaw rate gain as function of steering wheel angle frequency.
3.6 Instrumental quantities
Ratio between peak gain and steady-state gain
Ratio between peak gain and steady-state gain describes the ratio between the maximum value of yaw rate gain as a function of steering wheel angle frequency and its
value at
0.2 Hz
(see Fig. 3.7).
Time delay at fδSW = 0.5 Hz
Time delay at
0.5 Hz
is the dierence in time between the peak of the steering wheel
angle amplitude and the yaw rate amplitude which is the vehicle's answer to the
steering wheel input. This quantity is standardised for a steering frequency of
0.5 Hz
since this is a typical steering frequency for e. g. single lane change manoeuvre.
3.6.3 Driver performance quantities
Instrumental quantities are characterised by being measured by measuring instruments.
However, not all of these instrumental measured quantities are only vehicle dependent,
they can also be dependent on the driver. Driver dependent instrumental quantities
must be dened as exactly as vehicle dependent quantities must be. A typical example
is the lap time of a certain vehicle on a certain track. It can be assumed that the vehicle enables a certain performance, however, dierent drivers are expected to perform
dierently even though they use the same vehicle.
Another example (that probably has a strong connection to the lap time) is the corridor
deviation.
Nilsson [34] and Neukum [35] showed independently of each other that
each driver had some kind of handwriting when driving a double lane change - some
kind of a personal corridor through the manoeuvre.
This is assumed to be valid
for all manoeuvres. A deviation from this personal corridor indicates a disturbance,
something that goes wrong for the driver. A single deviation might be reasoned because
of a moment of inattention. However, if the same deviation occurs several times or for
several drivers for a certain vehicle characteristic, it can be assumed that this vehicle
characteristic leads to worse driver performance.
Riedel et al. [25] dened several of this kind of driver performance quantities for the
double lane change. Several of them were adapted to the tests in this thesis and are
described more in detail in Paper E.
41
3 Steering feel and vehicle handling
42
4 Asking for steering feel development of test
methods
This chapter shows the methodology for the mapping of steering feel which was
investigated by two experiments. The rst experiment was a driving test with
three trucks on a closed test track (see Paper A). The second experiment built
on the experience from the rst one and was performed in a moving base driving
simulator (see Paper C). The general preparation is described followed by the
evaluation methods and the experiments for data collection.
4.1 Preparation
Both of the tests had in common that handling values were measured with measuring
equipment and steering feel was assessed by the test drivers. The instrumental measurements are well developed and standardised, delivering high repeatability. In contrary,
the non-instrumental part, the test drivers' ratings, are to be developed. Therefore the
non-instrumental measurements had to be well prepared taking the existing diversity
of the drivers into account.
4.1.1 Driver level
To make the driver level of experience in rating clear, all possible drivers were classied
according Table 4.1.
Table 4.1:
A
Driver level denition (cf. Paper C)
Professional test drivers in vehicle dynamics development
B
Test engineers in vehicles dynamics development
C
Test engineers in chassis development
D
Remaining people with driving licence
The demands for the drivers diered in the tests.
For the track test the aim was
to get professional truck drivers with as high experience in truck driving as possible.
However, experiences from this test lead to dierent demands for the following test
in the simulator.
A high familiarisation with rating and evaluation was required.
However, this requires also a certain experience since the evaluation of perception and
feelings is related to the personal experience [14].
43
4 Asking for steering feel development of test methods
4.1.2 Common language
An important aspect when considering ratings of human feeling, was a high standardisation since the ratings should serve as a base for direct comparison later on. Thus,
there was an intensive focus on human feeling and how people express or describe
feelings regarding steering feel (see Paper B).
In general, there are two dierent kinds of description, a direct and a metaphoric.
When describing how e. g. a human voice sounds there are typically two dierent
kinds of description, a metaphorical and a direct sense. An example: Everyone can
detect, only by listening to the voice, whether this human being has a cold. Describing
the sound of the voice in this way is metaphorical. It describes a situation, a picture
that we are familiar with. However, when a host in radio broadcasting has a cold, it
is not meant hearing this on the radio. In radio broadcasting the sound of a voice can
be modied. So the sound of the voice will be altered by means of ltering. When
describing the changes made by parameter variation of frequency lters the description
is direct, e. g. lighter. Here the description is done with direct connected words that
are normally used in this context.
For the general description of steering feel it was decided to use the direct description
since this seems to be more general.
For specic steering feel investigations e. g. in
combination with the development of certain driver assistant systems, a metaphorical
description could be preferred.
For the track test an interrogation was made with 15 people extracting a common
language. This seemed to be insucient since in the test the drivers had diculties to
assign the words to certain perceptions of steering feel. Therefore, for the simulator
test, a clear dened word pool was elaborated.
The methodology is described in
Paper B. The result was a compact word pool that is subdivided in eight dimensions,
of the description of steering feel.
4.1.3 Questionnaire
For the track tests, a questionnaire was developed and evaluated. It was designed in
two columns. The rst column contained the objective description, the second column
was the subjective view where the subjects were to describe whether they liked it
or not. This was done to emphasise the dierence between description and personal
evaluation. However, for the simulator test the subjective view was removed to shorten
the test procedure. Nevertheless, the test drivers were instructed to divide between
description and evaluation and were at the same time invited to give comments.
Human beings usually describe their feelings and perceptions in words. The questionnaire tries to cover the dimensions of people's perception and converts it into numbers.
44
4.1 Preparation
The conversion is made by help of the semantic dierential method. As seen in section 3.5 a dimension is always described with two representative words that describe
the opposite of each other.
In-between the two words there was a scale.
Whether
there are numbers on the scale or not depends on the test design. In this case seven
steps with numbers were chosen which were distributed over the range of
−3
to
+3
in
the track tests. For the subjective rating (what the driver likes or not) the idea of the
scale was the same. The rating was absolute.
For the simulator test the range was shifted to the interval
1
to
7
to avoid negative
numbers. On this scale the subjects rated the driven vehicle regarding the asked dimension relative to the reference. The scale was deliberately chosen with an odd number
of steps to obtain a dened middle of the scale, which was equal to no dierence.
In the pre-test to the simulator experiment, the subjective rating of what the driver
likes or not was tested, too. The scale of the track test was taken over but with only
three steps.
−1
represented
feels worse, while
+1
represented
feels better and 0 was
reserved for no dierence and/or the dierence does not matter. However, a pre-test
in the simulator showed that it took too much time to ask for both assessment and
rating. Therefore, from the simulator experiment only assessment data are available.
4.1.4 Evaluation
By help of the questionnaire with the semantic dierential method the drivers' assessments are converted into numbers. These values can be used to elaborate characteristical quantities for the non-instrumental part. If an assessment (Q) of a single driver
(D) for a certain vehicle (V) regarding one question (x) is called
over all drivers for one vehicle can be calculated
QxV .
QxDV
the average
With an increasing number
of drivers this value becomes more and more objective. Regarding this average, the
spread is interesting, too.
Focussing on the pattern of all ratings of one driver for one vehicle, this can be correlated to the similar pattern of other drivers. Finding correlations means that dierent
drivers experience a vehicle in a similar way. This is possible for several vehicles as well.
However, an even better method is a cluster analysis over the ratings (see Fig. 4.3)
since this method detects parts of similar patterns as well which cannot be found
manually or by correlations only.
The instrumental quantities are described in section 3.6. For the comparison of instrumental and non-instrumental quantities rst the correlation coecient was calculated.
When evaluating the simulator test regression analysis and neural networks were used
(for details see section 4.2).
45
4 Asking for steering feel development of test methods
4.2 Multivariate Data Analysis
Multivariate data analysis comprehends methods that allow an evaluation of data with
several input and/or output channels. When searching for correlation between what
engineers measure (M) and what drivers feel (Q), the instrumental values (M) can be
seen as input data (independent variable) and the non-instrumental values (Q) can be
seen as output data (dependent variable). This will enable engineers to conclude from
later measurements to the feeling the vehicle will give the driver.
4.2.1 Methods
Each method is characterised by input, output and the conclusion that can be made
over the relation. Most of the methods only detect linear relations. A short overview
follows:
•
Correlation analysis
A correlation indicates the strength and direction of a linear relationship
between two random variables.
The correlation analyses one input against one output variable.
The correlation analysis is a good pre-test for multivariate data analysis.
Q=k·M
•
(4.1)
Linear regression analysis
The linear regression analysis examines linear relations between metric data
consisting of values of one dependent variable (also called response variable
or measurement) and of one or more independent variables (also known as
explanatory variables or predictors).
The linear regression analysis covers only one output variable.
To cover
more than one dependent variable the regression analysis can be performed
several times.
Q=
n
X
j=1
•
k j · Mj
(4.2)
Analysis of variance (ANOVA)
The ANOVA analyses linear relations between statistical data consisting
of values of a non-metric dependent variable (also called response variable,
here the assessment) and of one or more metric independent variables (also
known as explanatory variables, here the instrumental measurements). Nonmetric dependent variable means that the scale is qualitative or a ranking.
46
4.2 Multivariate Data Analysis
The main equation calculates the empirical F-statistic with the quotient
of the variance between the groups (index b) and the variance within the
groups (index w) for each steering system parameter and each combination
of parameters.
Femp =
•
Varb
Varmodif ication + Varspread
=
Varw
Varspread
(4.3)
Principal component analysis (PCA)
The PCA reduces large amounts of statistical data by nding factors that
can nearly be linear combinations to the observed variables. Each combined
assessment
pkq
Akj
was a sum of a factor
over all assessments
ajq
Akj =
Q
X
q=1
•
and a corresponding loading factor
q.
ajq · pkq
(4.4)
Contingency analysis
The contingency analysis detects correlations between statistical data that
are provided in nominal or ordinal scale where
contingency,
χ
was the coecient of the
the setting.
C=
•
s
χ2 -test
C
and
χ2
χ2 + ns
was the coecient of
ns
was the number of
(4.5)
Cluster analysis
The cluster analysis groups variables (e. g. test persons or questions) by
means of the contained statistical data and can nd new data groups. It is
a data-mining method.
•
(Articial) Neural network
A NN analyses a sucient number of statistical data also in a non-linear
way.
A NN needs not to postulate non-linearities a priori.
A NN is dependent on the design made by the user.
~ →Q
~
M
(4.6)
47
4 Asking for steering feel development of test methods
4.2.2 Choice of method
In this work the correlation coecient was used to get a quick overview over data. To
nd correlations the linear regression analysis was used since input and output data
were available as metric data.
Neural networks were used parallel to the regression
analysis. This was done to get a higher reliability of the results, and this oers also
the possibility for non-linear correlations.
4.2.3 Neural networks
Articial neural networks, here just called neural network (NN), are mathematical or
computational models along the lines of biological neural networks. Articial neural
networks originate from computational neuroscience and neuroinformatics.
The for-
mer intends to understand how the (human) brain works; the latter studies how to
use neural networks in technical applications. Typical operation purposes are pattern
recognition (e. g. human faces) and non-linear statistical data modelling. A comprehensive introduction in neural networks is too complex in this context. Although, some
important information for the actual approach is presented here.
A simplied description of a neural network is that an input vector is transformed into
an output vector (see Fig. 4.1). Considering the fuzziness it is an associative memory
because even an input vector that diers in some way from the actual input vector
dδSW
day
∆t
fB
MSW
Output Layer
MSW
Hidden Layer
Input Layer
will result in the designated output vector.
indirect 1 ... 7 direct
stable 1 ... 7 unstable
necessary SW angle 1 ... 7
SW return 1 ... 7
easy 1 ... 7 heavy
Kus
Figure 4.1:
Example for a 3-Layer (6-5-5) neural network transforming a vector of six
handling values into a vector of ve non-instrumental values.
Neural networks consist of (articial) neurons (see Fig. 4.2) that perform data processing with input and output data and communicate with other neurons. They are
48
4.2 Multivariate Data Analysis
usually placed in layers. One input-layer and one output-layer are at least necessary.
One or more so called hidden layers in-between are possible. A neural network with
at least one hidden layer, a sucient number of neurons and the neurons' non-linear
activation function is able to approximate any non-linear function [36] [37].
Figure 4.2:
A neuron in a neural network. In the evaluation a logistics activation
function was used.
An NN consists of an interconnected group of neurons and processes information using
a connectionist approach to computation. In most cases an NN is an adaptive system
that changes its structure based on external or internal information that ows through
the network during the learning phase.
To make a NN performing think steps transforming input vectors into output vectors,
it has to be trained. A certain amount of data must be used to train the NN. The
input e. g. characteristical handling values is transformed to an output e. g. ratings.
The dierences between the NN's output and the known values are used in a special
learning process (here Backpropagation is used) to perform corrections inside the
network.
After several sessions of training the NN in the best case generalises the
transformation from input into output.
Investigating vehicle behaviour regarding handling and steering feel, handling values
can be transformed into steering feel ratings. After performing measurements the test
results can be used to train and to evaluate the network. The former will be done with
the larger part of the test results while the latter will be done with only a small part
of the test results.
49
4 Asking for steering feel development of test methods
4.3 Track test with three vehicles on low friction
4.3.1 Vehicles and test track
This study concentrates on two-axle heavy tractors with a dead weight of eight tonnes
and a maximum weight of 18 tonnes (see Paper A). Three tractors were used in the
experiment where one of them was made by another brand.
All three tractors had
dierent kinds of steering feel assessed by several drivers previously.
A semi-trailer
was not used due to safety concerns. To obtain a more realistic axle load distribution
the tractors were loaded on the rear axle with a ballast frame of three tonnes. The
sequence of vehicles for each subject was randomised as much as possible, within some
practical constraints.
One restriction of the standardised tests is that they are only valid under pre-dened
conditions (using ideal conditions like at, dry and clean surface (µ
= 1, no wind, etc.)
[38]. The transferability to other circumstances is limited [39]. For example, a certain
response from the tyre-road-contact is not as important for the driver under those
idealized circumstances as in extreme skidding situations. Therefore the track tests
was decided to be performed on a slippery surface. The manoeuvres were performed
on a closed test track. The track surface consisted of horizontal asphalt with dierent
friction coecients by means of dierent kinds of asphalt colour and water.
4.3.2 Test drivers
15 people were chosen as test drivers in the age from 19 to 62. Most of them were
professional truck drivers and had more than
50 000 km/year driving experience on heavy
trucks; two of them were female, six of the drivers were employed by Scania R & D. As a
preparation for the experiment, the driving began with a training phase to familiarise
the driver with the vehicle.
The experiment itself comprised seven dierent parts
belonging to dierent word pairs. With the word pairs a maximum range was taken
into account.
Since the detection of all driving feelings needs dierent manoeuvres,
a pilot test was done to develop the dierent manoeuvres. Depending on the driver
the test took about one hour per vehicle with the possibility for a break at any time.
Between the tests there was some resting time for the drivers.
4.3.3 Results
There are dierent interpretations of the results outcome. First of all, the correlation
coecient is a characteristical value for correlations between the instrumental and the
non-instrumental measurements. However, since the gathered data set contains only
three tractor samples, it is statistically uncertain to apply the correlation coecient
50
4.3 Track test with three vehicles on low friction
method. Therefore this method is abandoned here. But correlation coecients do not
show all relations as to be seen in the driver independency test. To analyse whether
there are characteristical groups of truck drivers, their answer series were correlated.
A high correlation coecient between two or more drivers would point to some kind
of similar sensation but no signs were found in this study.
In contrast to this, a
cluster analysis extracted two groups of drivers (see Fig. 4.3):
with one exception
each, the cluster analysis could distinguish the drivers that came from Scania from
the external drivers by their ratings.
any of the groups.
One external driver could not be attached to
The result is the same when analysing only a part of the data
which underlines the continuity of the eect.
This could indicate the dierences in
the drivers' evaluation experience level, where the Scania employees that came from
testing groups of the R & D department, are more used to assessing and rating than
professional truck drivers.
Figure 4.3:
Result plot of the cluster analysis over the drivers' answer series showing
two groups of drivers. The distance on the ordinate is a measure of the
similarity of the drivers' answer series (small distances correspond to high
similarities).
In contrast to the main hypothesis of this experiment, the described method and
analysis show no detectable correlation between the instrumental values (handling
properties) and the non-instrumental values (drivers' ratings). Furthermore, the condence intervals frequently showed overlapping.
This means that the ratings of the
51
4 Asking for steering feel development of test methods
three tractors with 15 drivers did not show sucient detailed dierences to distinguish
them from each other signicantly.
Since the test leader was sitting on the passenger seat during the tests assisting the
drivers, there was the possibility to get more information than the ratings on the questionnaire. Of course, the drivers told in addition what they liked and what they did
not. In this context the speculation came to existence that the drivers did not only rate
as objective as possible even though there was a separated part of the questionnaire
where to ll in the personal preference. In a discussion Gies [40] supported the assumption that brand loyalty inuences the drivers' ratings signicantly. This led to the idea
to go on with the investigation in a driving simulator where the surroundings could be
constant while changing only certain parts in the simulation. Table 4.2 visualises the
dierences of the track test and the simulator experiment that were performed with a
similar approach.
Table 4.2:
Test structure of steering feel tests.
Track Test
Simulator
Test manoeuvre
Several (matched to question)
Fixed corridor
Test objects
3 trucks
16 steering characteristics
Test procedure
One truck each
One to two questions each
Rating
Absolute
Relative
Questionnaire
First
Based on Paper B
Analysis
Correlation & Cluster
Neural Network & Regression
Conclusion
Rating dierences
Correlation
4.4 Simulator experiment
1
The simulator used in the experiment was the Driving Simulator II at VTI
in Linköping
[41], a lateral moving base simulator with roll and pitch motion (see Fig. 4.4). The
simulator is often used for investigations about driver assistance systems and driver
behaviour where a high grade of repeatability is required [42]. The control of roll angle
was used in combination with lateral movements for the simulation of lateral forces
due to cornering. The control of pitch angle was used for the simulation of braking and
acceleration. However, pitch angle control was not part of the experiment, only for the
start of each session, since the test manoeuvre was driven with constant speed. Yaw
rate was only simulated visually, i. e. there was no real yaw rate. The visual impression
to the driver was simulated by three video projectors that cover 120 degree forward
eld of view.
1 Swedish
52
National Road and Transport Research Institute
4.4 Simulator experiment
Figure 4.4:
Principle of the VTI moving base Driving Simulator II with a passenger
car cabin [41]. For the described experiment a truck cabin was installed.
The vehicle model was a validated model of a tractor and a semi-trailer that was
provided by the operator of the driving simulator.
Fig. 4.5 shows the metrics of
the modelled vehicle combination. Fig. 4.6 shows a principle view of the model that
consists of ve rigid bodies being connected with springs and the axles, which results
in ten masses. The tyre characteristics were modelled separately. The model of the
steering system was also separate with the possibility to change the parameters friction,
damping, inertia, stiness and boost curve separately.
Figure 4.5:
Metrics of the tractor and trailer model used in the simulator experiment.
53
4 Asking for steering feel development of test methods
Figure 4.6:
Principle view of the ve bodies the vehicle combination is modelled of.
The ve axles are modelled separately.
4.4.1 Simulator experiment planning
From other tests [18] it is known that non-professional test drivers cannot evaluate
strictly objective. They need a reference to compare with! Since there are only few
of these professionals around, rst the simulator test was performed with test drivers
of the categories B and C due to the denition made in Paper C (see Table 4.1). Second the simulator test was from the beginning planned as a comparative test.
The
disadvantage of this kind of test design is that the subjects become very familiar with
the reference characteristic. Together with the experience that any other characteristic in the rst moment feels strange and therefore worse, the ratings about personal
preference (like it vs. don't like it ) are hard to interpret.
Another problem in comparative tests is the quite short perceptive memory of human
beings. The information about the short time memory regarding impressions diers
in the literature from several seconds up to few minutes.
Especially regarding the
comparative impression of steering feel there is no reliable value known [43].
This
leads to a quite short test manoeuvre sequence which makes it even harder for the
driver to get familiar with the new characteristic. Therefore any test design will be a
compromise.
4.4.2 Test manoeuvre
There were several demands for the test manoeuvre:
•
First of all it should not be too dicult to manage since the drivers should have a
certain amount of their cognitive power left to feel and recognise the steering feel.
An extremely dicult manoeuvre would lead to the drivers perhaps managing
the task but without feeling (or remembering) anything.
•
Second the necessary steering wheel angle had to be non-constant since the task
was to evaluate non-steady-state properties. So the manoeuvre had to force the
driver to turn the steering wheel.
54
4.4 Simulator experiment
•
Third the available lane should be quite narrow like in a road construction works
area because this is a situation where drivers rate vehicles regarding the quality
of the steering feel. A good steering feel leads to managing the situation easily
while a bad steering feel leads to a lot of corrections and an unsafe feeling.
•
Further on the manoeuvre had to be short since the human recognition memory
accommodates only a few seconds.
Even the time between two dierent test
episodes should be as short as possible. The characteristic is to be changed from
reference to test only a short moment before the manoeuvre begins. Otherwise
drivers will get information about the certain properties by voluntary or unvoluntary motor activities
2
[44]. This means that the driver already while driving
straight realises dierences in the steering system. This covers the author's experience from another simulator experiment [7] that the change in characteristic
could be felt instantly.
Especially for tests in a driving simulator the attention has to be turned to a realistic
manoeuvre which includes the tracking and is limited by the lateral acceleration that
is experienced as realistic only in a certain bandwidth. And the manoeuvre should be
driven at constant speed on the one hand to avoid dierences between the drivers and
on the other hand to make an evaluation possible coupled to the standard handling
tests.
Dierent manoeuvres were tested and evaluated i. a. slalom with constant or decreasing amplitude, driving through a roundabout, double lane change etc. The resulting
manoeuvre (Table 4.3) is a combination of several of them, taking the ISO handling
tests into consideration and consisting of three parts framed by straight road driving:
1. A lateral deviation like a single lane change in the beginning of a road works
area
0.4 m/s3
0.2 Hz and
2. A closing curve with a jerk of
(approaching transition test)
3. A slalom with a frequency of
an amplitude of one lane width of
3.5 m
(approaching weave test).
The manoeuvre closes with a straight road in normal width (4.5 m). The vehicle speed
is at a constant speed of
70 km/h
realised by a cruise control system. As named above
the same speed for all test subjects is important to make the tests comparable, to
couple the results to the standard handling tests and to make sure that the driving
experience is the same regarding jerk in the closing curve part and the frequency
response in the slalom part.
2 Translation
from the original citation: Bestimmte motorische Aktionen werden willkürlich oder
auch unwillkürlich zunächst ausgeführt, um vermehrt Informationen über die Eigenschaften [...]
des Steuersystems zu gewinnen. 55
4 Asking for steering feel development of test methods
Table 4.3:
Manoeuvre
Single lane change
Closing curve
Opening curve
Slalom
Test manoeuvre in simulator.
Range
4m
< 0.4 m/s3
> −0.4 m/s3
0.2 Hz & 3.5 m
Vehicle speed
70 km/h
4.4.3 Pre-test and simulator experiment
The manoeuvre was tested in a pre-test on a closed test track. Three dierent trucks
were available - one as reference and two trucks that were to be rated.
In this pre-
test dierent test congurations were tested, e. g. how many questions can test drivers
process at the same time, how many repetitions were necessary etc. The result from
the pre-test was at rst that the manoeuvre matched the demands. The manoeuvre
could be driven only once, thereafter the driver had to full a lap on the test track
to reach the start again where he could change the vehicle or repeat the manoeuvre.
To drive the manoeuvre again directly but with a dierent characteristic could not
be tried out but was imagined favourably by the test drivers. According to the test
drivers it was long enough to feel the vehicle characteristics but at the same time it
was short enough to compare it in a directly following manoeuvre. Moreover, the test
drivers could feel dierent parts of steering feel and they could dier between the three
trucks.
An in-depth description of the experiment in the simulator and the results is done in
Paper C, a summary is given in section 7.2.
56
5 Active steering systems
This chapter provides the denition of and background information about active
steering systems showing the principles of torque and angle overlay. In a further
section safety aspects become an issue.
The chapter will end with previous
research about safety aspects and driver assistance with active steering.
5.1 Denition
2.3.
Active
steering
systems
Active steering systems are in this work dened as systems
that
can alter
the steering
torque and/or a steering angle additional to and independent of the driver's steering
wheel input (see Fig. 5.1). This excludes the servo characteristic if it is not electronisteering wheel input. This excludes the servo characteristic if it is not electronically
cally controlled enabling a torque that is more than only an amplication of the drivers
controlled enabling a torque that is more than only an amplifying of the drivers steering
steering wheel input. Steer-by-wire systems (SbW) full this denition since they oer
wheel input. By denition this excludes steer-by-wire systems that, of course, oer
the same possibilities, however, SbW is a further step in technology development and
the same possibilities but are a further step in technology development.
is excluded from this work.
- vx
ps
MSW
Extern
-
- m
-
- ...
Input
Conventional
Torque & Angle
Steering
Output
Torque & Angle
Input
Active
Torque & Angle
Steering
∆δSW
∆MSW
Output
Torque & Angle
Conventionalpower
powersteering
steeringvs.vs.active
active
steering:
output
a conConventional
steering:
TheThe
output
of aofconvenventional
steering
system
isdependent
only dependent
of theand
input
and the
power
tional
steering
system
is
only
of
the
input
the
power
steering
steering characteristic.
The
output
of steering
an active
steering
system is depencharacteristic.
The
output
of
an
active
system
is
dependent
of
the
dent of
andexternal
any chosen
extern parameters.
input
andtheanyinput
chosen
parameter.
Figure
Figure2.2.
5.1:
2.3.2. Superposition of steering angle
5.2 Superposition of steering angle
The used possibilities in hardware today for a superposition of steering angle are a
The usedgear
possibilities
hardware today
for adrive
superposition
angle
are a
planetary
box or ainsuperposing
harmonic
gear box of
in steering
the steering
column,
planetary
gear
box
or
a
superposing
harmonic
drive
gear
box
in
the
steering
column,
realised by e. g. BMW, Audi and Lexus. A planetary gearbox is characterised by
realised
by e.axles
g. BMW
[45],
Audia [46]
and Lexus
three
coaxial
which
allows
summation
of [47].
angle and torque following the same
principle as a dierential gear. Instead of dividing one input into two outputs there
are two inputs added to one output. The output is connected to the steering system.
The two inputs are the steering wheel and a joined electric motor. The electric motor
57
interferes a steering wheel angle with the driver's steering wheel angle input depending
on a controller. This system allows statically the superposition of an angle (e. g. for
side wind compensation) or dynamically a free design of steering ratio with possibility
for variation at any time. The harmonic drive gear box oers the same functionality
5 Active steering systems
5.2.1 Angle overlay by means of a planetary gearbox
A planetary gearbox is characterised by three coaxial axles which allow a summation
of angle and torque following the same principle as a dierential gear.
Instead of
dividing one input into two outputs there are two inputs added to one output. The
output is connected to the steering system. The two inputs are the steering wheel and
a joined electric motor (see Fig. 5.2). The electric motor superposes a steering wheel
angle with the driver's steering wheel angle input depending on a controller.
This
system allows statically the superposition of an additional angle (e. g. for side wind
compensation). Dynamically used, it can superpose an additional angle depending on
the actual steering wheel angle. This will be experienced as a change of the steering
ratio by the driver and enables therewith, regarding steering feel, a free design of
steering ratio with possibility for variation at any time. However, there are at least
two limitations: The actuator will have limited dynamic properties. This means that
the actuator cannot change the experienced steering ratio over the whole spectrum of
frequencies (which must not be a disadvantage!).
Figure 5.2:
Picture of the BMW angle overlay planetary gearbox [48].
5.2.2 Angle overlay by means of a harmonic drive
The harmonic drive gear box oers the same functionality as the planetary gear box
by using the principle of a strain wave gearing (see Fig. 5.3). However, a strain wave
gear has a very high stiness and because of its pretension is it free of play.
58
5.2 Superposition of steering angle
(a) Exploded view.
Figure 5.3:
(b) Gear tooth system.
Harmonic drive gear [49].
The gear box consists in general of three main parts, the wave generator, the exspline
and the circular spline (Fig. 5.3(a)).
The wave generator is an rigid oval which by
mean of rollers presses its exible outer shell against the exspline.
The exspline
is a exible shell as well which is pressed into the circular spline by means of the
wave generator. Since the wave generator is oval, the exspline is pressed against the
circular spline only along one radial axle where the splines engage (see Fig. 5.3(b) top
left). Along the other radial axle the splines do not engage (see Fig. 5.3(b) bottom
right). Because of the dierences of teeth between exspline and circular spline the
ratio comes to existence.
iHD =
nexspline − ncircular spline
nexspline
(5.1)
5.2.3 Angle overlay and steering ratio
Moreover, the steering ratio is not only an important parameter for steering feel but
also for steer-ability.
Especially heavy vehicles can hardly be steered without the
assistance by power steering. If the power steering fails, the vehicle has by law to be
steerable anyhow.
Changing the steering ratio by means of superposition of torque
may only be a limited help since the torque equilibrium in the gearbox must be equal
to zero, meaning the electric motor must be strong enough to counteract the driver's
steering wheel torque input and the vehicle's steering system response torque with
respect to the ratio in the gear box. However, the other way around this eect can be
utilised [50]. If the static ratio of the angle overlay gearbox is e. g.
i = 2, then the static
ratio of the steering system can be doubled which will result in a very easy steering
even without power steering. The normal steering ratio will then be installed by a
proper dynamic angle overlay for normal driving.
59
5 Active steering systems
5.3 Superposition of steering torque
Another principle of active steering is the superposition of a steering wheel torque. In
this case the steering ratio is not alterable but possibly non-linear. There are dierent
ways how to superpose the steering wheel torque, depending on the power steering
system.
5.3.1 Torque overlay by means of electric motor
In smaller passenger cars EPS (Electric Power Steering) is established and permits
a free composition of steering wheel torque in dierent situations independently of
the drivers steering wheel input.
An example of active steering functionality is the
parking assist realised with EPS: The car turns the steering wheel and directs the
driver whether to drive forward or rearward to move into a parking spot. EPS systems
must be distinguished from EHPS systems which are characterised by a electric driven
hydraulic pump. Even EHPS-Systems (Electro-Hydraulic Power Steering), enable a
variation of the assistant steering torque which manipulates the steering wheel torque
within certain limitations. However, EHPS does not allow an arbitrary torque that is
independent of the drivers steering input.
A combination of HPS (Hydraulic Power Steering) and EPS oers the freedom of the
electronically controlled electric motor and the power density of a hydraulic system.
Two types of realisation for heavy vehicles are the ServoTwin steering servo from ZFLenksysteme [51] and the ColumnDrive steering system from TRW [52]. These systems
dier in their implementation but both of them have the principle design in common:
At the steering column which is the input shaft to the hydraulic servo, an electric
motor is installed that can create a torque at the steering column. Since the driver
induces his input moment at the steering wheel that is coupled to the steering column,
both torques are superposed. This design can be used in both directions, of course:
assisting the driver to disburden him from heavy steering or resisting the driver e. g. to
warn him. The system needs another torque sensor next to the torsion bar to enable
a smooth cooperation between electric motor and driver.
5.3.2 Torque overlay by means of electronically controlled hydraulics
In an HPS system the driver's steering wheel torque input deects a torsion bar that
opens and closes valves to regulate the hydraulic pressure that assists the driver with
steering (see section 2.2). For a superposition of torque the hydraulic pressure in the
servo is used. Controlling the valve additionally to the drivers steering wheel torque
input enables to control the pressure in the servo. This can be realised by rotating the
inner or the outer parts of the valve. The superposition will already take place in the
60
5.4 Safety aspects for active steering
Figure 5.4:
Principle of two dierent forms of active steering with superposition of
torque by means of an electric motor (left) and superposition of angle by
means of a planetary gearbox (right).
torsion bar deection and will therewith propagate to the hydraulic pressure. There
are publications [53], patented designs [54] and market solutions [55] for this.
5.4 Safety aspects for active steering
5.4.1 Superposition of steering torque
For active systems that inuence the most important vehicle dynamic control areas like
braking and steering, the safety aspect is an important issue. The maximum torque of
the torque overlay system ServoTwin is equal to around
100 Nm steering wheel torque.
This torque amount is higher than the maximum tolerable steering wheel torque in
case of servo fail.
So it can be assumed that a driver cannot counter a that high
steering wheel torque. This means also that a that high torque set by mistake, can
lead to fatal accidents.
So in the development of such kind of systems sources for
mistakes must be eliminated systematically. One example is to limit maximum values
and gradients of the actuator.
Anyhow, normally actuators are chosen or specied
with a certain dynamic behaviour (power and bandwidth) since there are cases where
this is required. A guideline for these safety issues was developed over the last years
and is today available as ISO 26262 [56] that is valid for passenger cars, however, the
principles are employable for heavy vehicles as well.
One possibility is, of course, to limit the actuator. In the iHSA system from Tedrive [55]
for example the hardware is limited to a certain maximum torque. This limit can be
chosen by the OEM. A typical value could be in the range of
MSW = [4, 10] Nm.
This
61
5 Active steering systems
is an amount of steering wheel torque that the driver could counter easily. An advantage with this kind of solution is that (according to ISO 26262) from the beginning
a low automotive safety integrity level (ASIL) can be achieved which simplies the
development process.
5.4.2 Superposition of steering angle
Especially angle overlay actuators used to be very powerful because of the required
dynamics. This means that an actuator can steer faster than an average driver could
countersteer.
A particular problem for angle overlay is the lack of haptic feedback
which enlarges the reaction time. So a faulty steering intervention with high dynamics
will probably lead to a fatal accident since the driver will be overburden. Depending
on the functionalities that shall be implemented, a limitation of gradients is hardly
possible so mistakes must be eliminated by another solution like e. g. redundancy of
signals, sensors and ECUs.
5.5 Active steering and driver assistance systems
Active steering itself is not a driver assistance system (DAS) but it is a platform
for enabling DAS. It is more comparable to the brake system that is a tool that
can be used for e. g. braking, stabilisation (ESC) or traction improvement (electronic
dierential lock). In this way active steering can be used for e. g. lane keeping assist,
ESC-support, crosswind compensation, remote control, autonomous driving etc.
In
the present work assistance systems are not in focus. Instead focus is on requirements
for DAS especially with respect to active front steering systems (AFS). The point
of interest is AFS from the view of vehicle dynamics and especially vehicle handling
which leads to certain steering feel. In sections 6.3 and 6.6 there are functionalities for
active steering systems presented that inuence steering feel and driver behaviour.
5.5.1 Assisted and automatic obstacle avoidance
Eva Bender describes in her doctoral thesis how drivers judge automatic braking and
steering for collision avoidance [31]. In the rst PRORETA project where Eva Bender was involved, the lack of automatic driving assistance was compared to collision
avoidance by automatic braking and collision avoidance by automatic steering. The
evaluation was based on 450 driving laps with 102 subjects. The comprehensive investigation delivered the following results:
•
62
Press on the accelerator pedal while automatic braking cannot be interpreted as
the driver's request for accelerating. The accelerator pedal was primarily pressed
5.5 Active steering and driver assistance systems
when the driver's foot rested on the pedal. So the press on the pedal was a result
of inertia not of the driver's intention.
This coincides with other investigations. The inuence of the inertia of feet and calves
is e. g. known from Asp [57].
•
Automatic steering or braking initiates the drivers reacting by themselves.
Without any driver assistance, a third of the drivers did not react at all
trying to avoid the collision. With assistance nearly all drivers showed at
least some kind of reaction.
When the assistance system braked, the drivers predominantly braked by
themselves, too.
When the assistance system steered, the drivers predominantly steered by
themselves, too.
•
A soft automatic braking is NOT better than a hard.
There are no hints that a automatic soft braking is judged more safe.
Neither was a soft automatic braking judged more comfortable.
Neither was a soft automatic braking judged more acceptable.
However, a soft automatic braking was judged less incapacitating by the
drivers.
•
In one test variant the assisting system was steering the car on a trajectory
around the obstacle by means of superposition of steering angle without giving
the driver the possibility to inuence the vehicle's direction. Anyhow, 59% of the
drivers specied that the car would have followed completely or partially their
steering input.
•
In another testing variant the assisting system was only initiating an impulse
returning the control of the vehicle to the driver instantly. The very short impulse
makes the car avoiding the obstacle and arouses the driver who has to lead
the vehicle back on the track. This avoids the problem of when to return the
responsibility of the vehicle back to the driver, and gave the best results.
This agrees to the second aspect that the assisting system can initiate the drivers'
necessary reactions.
•
The assisting system can steer the car on a trajectory around the obstacle. Comparing decoupling the driver completely with enabling the driver to override, the
former was judged more safe.
These results are especially interesting for the future task of ADAS where the system
or the driver has to decide whether to turn left or right to avoid an obstacle. Bender's
results show that the system can trigger the driver. However, whether the trigger rules
the direction is not yet answered and maybe a question for future work.
63
5 Active steering systems
5.5.2 Fail-safe-properties of active front steering systems
At the University of Würzburg, the Department of Psychology the Chair of Psychology
III deals with methodology and trac science in the interdisciplinary center for trac
science. Several investigations were done, a lot of them in cooperation with BMW, to
prove the safety of active (front and rear) steering systems [16, 35, 58].
One important part was the drivers' tolerance for faults of the steering system [16]. In
track tests a steering angle fault was superposed while driving straight on with dierent
= 100 km/h; ay ≈ 2 m/s2 ). The
steering ratio was changed while cornering and U-turn driving (vx = 15...30 km/h), an
ISO double lane change (vx = 30...80 km/h) and a slalom (vx = 30...50 km/h). In both
speed (50...150 km/h) and while following a wide curve (vx
tests the 25 to 30 non-professional and the four to ve professional test-drivers were
asked to rate the fault on a two-level scale.
The rst level was the rating between
ve ratings beginning with nothing observed up to vehicle is uncontrollable. The
second level was a dierentiation more in detail. In the end the scale had overall 11
steps (see Fig. 5.5).
10
Vehicle uncontrollable
9
Disturbance dangerous
8
7
6
Disturbance annoying
5
4
3
2
Disturbance observable
1
Nothing observed
Figure 5.5:
0
Two-level scale used for fault-description of active front steering corresponding to Neukum [16].
Additionally, the drivers described their workload, their experienced driving safety and
their mastery of task on another two-level scale (see Fig. 5.6).
very low
1
Figure 5.6:
64
2
low
3
high
medium
4
5
6
7
very high
8
9
10
Two-level scale for the assessment of subjective driver workload corresponding to Neukum [16].
5.5 Active steering and driver assistance systems
The results showed that the range according to Fig. 5.5 was covered.
Changes in
steering ratio (switching o the variability which results in a jump back to the ordinary
steering ratio) do not impair driving, not either in dicult manoeuvres. Safety critical
situations would only come up if the manoeuvre was that dicult that it was hardly
manageable even without the help of a variable steering ratio.
In case of steering
angle faults there have to be well dened boundaries to assure that disturbances do
not lead to driving safety critical situations. This means that for the present vehicle
(5-series BMW (E39) with a wheelbase of
2.83 m)
an added steering angle
δ < 0.3◦
is
according to Fig. 5.5 observable as disturbance but not annoying. An added steering
◦
angle δ > 0.6 is at high speed assessed
δ > 0.9◦ , according to Neukum [16].
as dangerous, for low speed this is the case for
5.5.3 Inuencing driver behaviour by modied steering feel in buses
Juhlin [7] presented in 2009 a method to inuence the performance of bus drivers by
means of manipulated steering feel to improve crosswind performance. The crosswind
performance of a bus deteriorates when the center of gravity (CoG) moves rearwards.
This results also in a altered steering feel.
By means of an adaptation of steering
ratio and servo assistance Juhlin modied the steering feel. Buses with CoG situated
further ahead are less crosswind sensitive.
For these buses drivers prefer a low yaw
rate gain with respect to steering wheel angle and torque. In the same manner buses
with CoG situated further back are more crosswind sensitive and the drivers prefer
higher yaw rate gains. In a driving simulator study the drivers performed well in both
kind of buses with high yaw rate gain, however, this results in a special steering feel
which would not be accepted for normal driving. So Juhlin concludes that a variable
steering system would have potential to improve the crosswind performance of the
driver-bus-system.
65
5 Active steering systems
66
6 Prototypes and experiments for active steering
evaluation
This chapter gives an introduction to the prototypes used in this work and their
topology. Furthermore, the developed functionalities will be presented. Finally,
the track test experiment with one of the functionalities will be described including the preparation and evaluation.
6.1 Introduction
The tests performed during this thesis work were conducted partially in a driving
simulator and partially in prototype trucks equipped with active steering systems.
One truck was equipped with an active steering system with torque overlay and used
for tests. Another truck was prepared for an AFS system (angle overlay) and will be
used for future experiments. This chapter gives an overview over the implementation
and the developed functionalities. It will conclude with a summary of the steering feel
modication experiment and the statistics that is necessary for the evaluation of the
experiment.
6.1.1 Superposition of steering angle (AFS)
When building a prototype truck with AFS, existing hardware on the market for
passenger cars was used. AFS needs a dierential gearbox e. g. a planetary gearbox
or a strain wave gearing (Harmonic Drive), which enables the superposition of the
driver's input, the steering wheel angle, and the additional angle of an ECU controlled
electric motor. In this project a Harmonic Drive gearbox actuator was used that is
available in actual production passenger cars. Gearbox and electric motor as well as
sensors are integrated in a complex actuator which is controlled by a special controller
(SCU) that also takes care of the power electronics and safety features.
Since the actuator and the SCU came from a production car, the system only works
with a certain setting of information on the CAN-bus.
Therefore, there was a free
programmable ECU that emulated the necessary vehicle information on the CAN to
enable the ECU and its actuator (see subsection 6.2.2).
The use of steering equipment of a passenger car was possible since it is mounted on
the driver's side of the power steering in the truck, i. e. at the steering column. Here
67
6 Prototypes and experiments for active steering evaluation
the upcoming torque is in the same magnitude as in a passenger car which meant a
safety factor of at least 3 regarding torque. However, since this hardware usually is
used in production vehicles, there is a complex safety structure implemented which is
dependent on the vehicle's information that partially is vehicle specic. Therefore, the
necessary vehicle information was simulated (see section 6.2).
6.1.2 Superposition of steering wheel torque (EPS/MDPS)
The abbreviation MDPS (Motor Driven Power Steering) was established to dier active front steering with superposition of steering wheel torque from active steering
systems with superposition of steering angle. In the eld of passenger vehicles an Electric Power Steering (EPS) which is equivalent to MDPS, enables a free programmable
superposition of torque. An EPS system includes the advantage of not needing an additional actuator to realise active steering. For heavy trucks EPS is not (yet) available,
therefore, both abbreviations, MDPS and EPS, do not describe the system correctly.
However, rst a consequent designation in the eld of road vehicles makes communication easier. Second, in a heavy vehicle using superposition of steering wheel torque,
the actuator will also be used as EPS where the hydraulics will only be a backup for
higher necessary steering torque. Therefore, the designation EPS feels anyhow being
the best designation.
6.2 Implementation and environment
6.2.1 ECU
An active steering system normally has several levels of control. The low level control
regulates the electric motor taking its characteristic into consideration.
This level
controls also the power electronics that supplies the electric motor with current. On
the next level there usually are xed parameterised settings like the basic steering
characteristic.
These levels are highly secured and veried.
On top of these levels
there are the high level controls like steering interventions that are dependent on
external parameters (e. g. LKA). An important dierence to the low level controls is
that there often is an easy fail-safe-mode. A lane keeping assistant system e. g. can be
deactivated without any problems.
The high level control of the active steering system was programmed in Matlab Simulink
[59] (see also section 6.3).
in-the-loop test engine.
The model was tested and parameterised in a software-
Thereafter, it was compiled by means of an in-house rapid
prototyping platform and ashed on a multi purpose ECU (in diagrams labelled as
COO7) which provides several in- and outputs, especially four CAN-controllers.
68
6.3 Functionalities
6.2.2 AFS
For the superposition of angle, CAN output consists of the emulation data for the
AFS controller and the superposed angle. The former is partially an easy emulation
of data that can be constant (even if they in reality are not, e. g. engine speed) and
partially of data that must be generated or calculated from vehicle input data. The
latter is a combination of the calculated steering angle that is to be superposed, and a
compensation of a steering wheel angle dependent superposition of steering angle that
is generated in the AFS controller and cannot be switched o. The generated data
include multiplexing, counter and checksum calculation inside the eight bytes CAN
data eld. For schematic diagrams of hardware and wiring see Appendix B.
6.2.3 EPS
The prototype truck with torque overlay was already existing including certain functionalities like Active (steering wheel) Return and/or LKA. Both of them are not
heavy vehicle specic. Anyway, the heavy vehicle specic functionalities are discussed
in section 6.3.
6.3 Functionalities
The main goal of this second part of the thesis work was to develop and investigate
heavy truck specic functionalities.
Bösch [2] wrote: If the presentation of information for the driver is manipulated or
if manipulated information is oered articially, the manipulation will only be rated
1
as progress if the ordinary driver is led subconsciously to correct acts.
This was
the background for the manipulated steering feel test with rollover prevention as a
truck specic example (see below). Other ideas described in this section, are articial
understeering and yaw rate gain acceleration. These functionalities are based on the
aim for a linearisation of the vehicle behaviour.
6.3.1 Modied steering feel Rollover indication
For a test of driver behaviour manipulation by means of steering feel modication a
rollover prevention function was developed and evaluated.
The driving experiment
and the results are described in detail in Paper F.
1 Original
citation: Wird an der Informationsdarbietung für den Fahrer etwas geändert bzw. werden
manipulierte Informationen künstlich angeboten, ist eine solche Informationsverzerrung nur dann
als Fortschritt zu betrachten, wenn der normale Fahrer ohne Training subbewusst zu richtigem
Handeln verführt wird.
69
6 Prototypes and experiments for active steering evaluation
The general model used for the experiment is shown in Fig. 6.1. However, the input
lateral acceleration was realised with special focus on the experiment as a kind of
model predictive control. Instead of using the quite noisy lateral acceleration sensor
Sign
×
×
×
ay
Factor
Figure 6.1:
Madd
Dead Zone
Matlab Simulink model of MSF functionality for rollover prevention.
signal, an articial signal based on vehicle speed and steering wheel angle was calculated. The advantage of that calculation was next to the lack of noise a signal lead in
phase since the vehicle response time was eliminated. This is an advantage regarding
steering wheel feedback for the driver before rollover but it can be a disadvantage in
certain driving situations like skidding when the calculated lateral acceleration deviates signicantly from the vehicle lateral acceleration.
This means that the system
cannot react on real lateral acceleration. In the test engine environment, this becomes
visible on the one hand when the tyres saturate, and on the other hand when the
vehicle in real life would not react as, for instance, at high frequency random input
4
40
10
2
20
5
0
0
-2
−20
−5
-4
−40
−10
Figure 6.2:
10
20
30
40
50
time [s]
60
70
80
90
MSW,add [Nm]
15
−60
0
70
Steering wheel angle and torque from random test at vx = 70km/h
60
δ SW [°]
ay [m/s2 ]
(see Fig. 6.2).
−15
100
Random test from measurement with torque overlay for steering feel manipulation. The lateral acceleration decreases with increasing steering frequency, the added steering wheel torque remains at high level.
6.3 Functionalities
6.3.2 Articial understeering
Heavy vehicles can, in spite of the fact that they are in principle understeered, become
oversteered at higher acceleration (higher acceleration means here around
3 m/s2 ).
This
origins from tyre characteristics. So the aim of an articial understeering functionality
is to linearise the vehicle behaviour, see Paper G for more detail.
1
delta_SW
-K-
2
v
1
delta_add
PID(s)
Steeringratio
u
PID Controller
2
Product
Square
K_us
K_us_desired
Product1
L
Wheelbase
3
w_z
Figure 6.3:
Matlab Simulink model that provides articial understeering as well as yaw
rate gain acceleration by means of linear calculated yaw rate gain error.
6.3.3 Yaw rate gain acceleration - reducing the vehicle reaction time lag
Heavy vehicles react lethargic. One aim when using active steering is to make their
reaction more quick [60].
The steady-state yaw rate gain
Ψ̇
Ks.s
of passenger cars [61] and heavy trucks (4x2
tractors with trailer) dier by around 50%.
However, extending the view on rigid
trucks, the steady-state yaw rate gain will decrease because of the long wheelbase. The
yaw rate peak gain is for the before named tractors in the same range as the steadystate gain. For passenger cars the yaw rate peak gain can increase up to
Ψ̇
Ks.s,4x2
tractor =
Ψ̇
Ks.s,p.car
=
dωz
dδSW
dωz
dδSW
= [0.13, 0.25]
= [0.22, 0.34]
deg
s
deg
s
0.4
deg
s
/deg.
/deg
/deg
Summarised, passenger cars do react much faster than heavy trucks.
The reasons
for this can be found at the tyres and at the steering ratio. The tyres are especially
designed for the heavy weight. The steering ratio is often ruled indirectly by certication for steering force at servo failure. Moreover, the sporty almost nervous feeling
in certain passenger cars is not really desired in a heavy truck where the driver spends
71
6 Prototypes and experiments for active steering evaluation
◦
ωz [ /s]
δSW [◦ ]
Figure 6.4:
Yaw rate ω over steering wheel angle δ from simulation with enabled/disabled articial understeering.
his whole working day.
z
SW
Anyhow, a more direct or spontaneous vehicle reaction can
be desired. One possibility to perform this is shown in Fig. 6.3. The calculation of
an added steering angle based on a yaw rate error (dierence between measured and
calculated yaw rate) does not only inuence the understeering gradient but of course
also the vehicle yaw rate reaction.
This becomes even more clear when examining
Fig. 6.4 where the hysteresis loop width is a metric for the understeering gradient and
the inclination is a metric for the yaw rate gain. However, the steady-state yaw rate
gain is lower with the algorithm enabled which is a result of the articial understeering,
but the transients are more interesting.
At the top and bottom dead centres as well as at the origin it can be seen that
the vehicle reacts faster avoiding the hysteresis. A steering angle movement will be
transferred directly into a vehicle yaw reaction with only a very short time delay. This
time delay is shown in Paper E to be a metric that correlates to the steering feel
dimensions stability and response which includes the impressions of direct/indirect as
well as quick and precise where shorter time delays used to improve the experienced
steering feel [28]. So by means of decreasing the yaw rate gain time lag the steering
feel of a heavy truck is expected to be improved.
6.3.4 Yaw rate gain acceleration with variable steering ratio
In the model shown in Fig. 6.3 a variable steering ratio can be included as well. In
the model the steering ratio is only a constant factor. By means of an algorithm or
a look-up table a steering ratio characteristic can be achieved. However, the logical
next step is then not to calculate a theoretical yaw rate by means of an simple model
but to preset a certain yaw rate behaviour as function of vehicle speed (see Fig. 6.5).
This would enable an even higher yaw rate gain input at low vehicle speed.
72
st ein Modell,
tion des Fahrzeugs
Lenkeingabe
leunigen soll
6.4 Safety concept
1
delta_SW
delta_SW
2
v_x
v_x
w_z,calc
w_z/delta_SW
w_z,err
Product
PID(s)
1
delta_add
PID Controller
vx over yaw rate gain
3
w_z
w_z,measured
Figure 6.5:
Matlab Simulink model that provides yaw rate gain acceleration according
to a desired yaw rate gain characteristic.
6.4 Safety concept
For active steering systems with a powerful actuator there is need for a safety system
that prevents the overall system from unwanted actuator actions in case of a failure
(see section 5.4). The safety concept has to ensure steer-ability of the vehicle by the
driver even in the case of software or hardware failures. In case of angle overlay the
safety concept has also to prevent rollover caused by an active steering intervention.
This does not include an overall rollover prevention by means of active steering, though,
the driver can provoke a rollover anyway so there is still need for ESC but a failure of
the active steering system must not rollover the vehicle.
6.4.1 Superposition of torque
For the prototype with superposition of steering torque, the actuator was limited on
several layers. The maximum torque by the actuator was at two higher layers limited
to a that low level that the driver always could override the added torque with his own
steering torque input. On a lower layer the torque gradient was limited, too.
6.4.2 Superposition of angle
For the prototype with superposition of steering angle a maximum lateral acceleration
corridor was dened. The idea of the corridor was that the freedom of the actuator
only was within the limits of the corridor. If the corridor in a certain driving situation
δSW = +/−45◦ and the actual steering wheel angle adjusted by the driver
◦
was already δSW = −30 , then the actuator had to be limited to apply as extreme
◦
◦
values δadd = −15 or δadd = 75 . However, the steering wheel angle on its own is
was equal to
not very meaningful, therefore, a maximum lateral acceleration limit was identied by
means of a potential (conservative) rollover risk:
aymax = ±3 m/s2
(6.1)
73
6 Prototypes and experiments for active steering evaluation
To achieve a magnitude for the resulting maximum steering wheel angle, the steadystate driving situation was analysed based on the understeering gradient.
Kus =
L
)
(δ − R
ay
(6.2)
In this equation the Ackermann angle can be substituted with:
ay
L
=L· 2
R
vx
(6.3)
This propagates to:
Kus =
δ
L
− 2
ay
vx
(6.4)
From this, the inverse of the lateral acceleration gain can be read:
δ
L
= 2 + Kus
ay
vx
(6.5)
which can be rewritten for the steering angle:
δ = ay (
L
+ Kus )
vx2
(6.6)
So the maximum steering angle derives from the maximum lateral acceleration limit
according to the following equation:
δmax =
aymax (L + vx2 · Kus )
vx2
(6.7)
The result was a lter that limited the calculated added steering angle (in Fig. 6.6
visualised as Safety).
A higher requested added steering angle saturated at that
maximum value. For the yaw rate gain acceleration this meant that the peak in the
beginning was limited earlier, however, the actuator has physical limits of acceleration
and jerk as well so the safety limitation did hardly inuence the results.
6.5 Virtual test environment
For the development of the functionalities the models were embedded in a test environment that provided the function with the necessary input values (see Fig. 6.6). The
development environment contained a bicycle-model of a two-axle tractor with tyres
modelled according to Magic Formula [62]. The intervention of each functionality was
incorporated in this model via an implemented emulation of the active steering actuator modelled by its dynamic properties. The input values were generated articially
or taken from real vehicle measurements.
74
6.6 Modied steering feel Track test
Functionality
Safety
Vehicle Model
Actuator Model
Driver Input
Measurements
Script-controlled test-engine
Figure 6.6:
Active steering functionality in a simulation environment.
6.6 Modied steering feel Track test
6.6.1 Test vehicle
This study concentrates on two-axle heavy tractors with a dead weight of eight tonnes
and a maximum weight of 18 tonnes. The experiment and the results are described in
detail in Paper F. The test vehicle was a series production Scania tractor from 2010
(Fig. 6.7) with the vehicle data corresponding to Table 6.1.
Table 6.1:
Vehicle data of the test vehicle for the driver behaviour track test.
Manufacturer
Scania
Type
R730
Axle conguration
4x2
Wheelbase
3.7m
Total weight
15.1t (loaded for test case)
Axle load distribution
51% front 49% rear (loaded for test case)
Special feature
Active steering (torque overlay)
These vehicles are normally used in combination with semi-trailers for transportation
of goods and are typically for European road transport. In the test case a semi-trailer
was abandoned because of several reasons.
On the one hand there were practical
constraints, on the other hand there was the important indicator vehicle speed while
cornering which was easier to use for evaluation if exceptional good lateral acceleration
performance was available.
Since the test track was quite hilly, a quick truck was
preferred which meant relative lightweight and high engine power.
This could be
reached by mounting a frame on the tractor's fth wheel carrying extra weight. By
means of this extra weight the tractor's total weight could be increased from
to
15.1 t
8.06 t
and the axle load distribution could be changed from 75:25 to 51:49 which
makes the vehicle behaviour more realistic.
1
75
6 Prototypes and experiments for active steering evaluation
Figure 6.7:
Test vehicle with active steering (superposition of steering torque) for the
driver behaviour track test.
The truck was equipped with a screen as cockpit display. This means that tachometer,
engine revolution meter and all other information was shown as an animated picture
on one screen. This screen could be switched o during testing.
6.6.2 Test drivers
Overall there were 33 test drivers in the age range from 23 to 66 years chosen for
this experiment. The test drivers were according to the driver categories dened in
Paper C from the categories B, C and D (see Table 4.1). All of them were employed by
Scania R & D, 10 of them were professional truck drivers with more than
100 000 km/a
driving experience. Since the experiment was divided into two parts because of the
dierent sign of the intervention (see subsection 6.3.1), the test drivers were divided
into two groups, one driving Setting 1 against the reference, the other driving the
Setting 2 against the reference.
In the end 27 of the drivers were evaluated since
for six of them there occurred problems like failure of the active steering system or
measuring equipment or that the drivers chose a that low cornering speed that they
never came up into the lateral acceleration range of intervention.
6.6.3 Test track
The experiment was performed on a closed test track. The track that the drivers were
asked to drive, was similar to a country road with some curvy parts, some nearly
straight parts and some hairpin bends. Every lap was
7.6 km
long and contained 12
bends that forced the driver to slow down to avoid rollover and therewith could be
evaluated for this investigation. Ten of these bends were qualied for evaluation. Here
76
6.6 Modied steering feel Track test
the drivers could choose the vehicle speed on their own irrespectively of other trac.
Depending on the driver the test took around two hours.
According to Schmidt [20] haptic signals are easiest to detect when the steering activity
is low like on narrow roads or in smooth bends. So the dierent characteristics of bends
on the test track should enable the possibility for the driver to detect the articial
feedback (conscious or unconscious) and adapt their driving style according to their
perception.
6.6.4 Test procedure
The experiment was performed in Sweden in November, December and January and
after 5pm on each test day. This means that the experiment was performed in darkness
and in general on wet or snowy road conditions. These were on the one hand in general
adverse conditions, on the other hand do these conditions represent situations where
the driver relies on the vehicle and predominantly uses the non-visual feedback.
The drivers were instructed to drive the tractor (equipped with extra load) according
to their own style of driving. They knew about active steering being installed in the
test vehicle but according to the instruction the purpose with driving was to collect
representative driving data with dierent drivers. The cockpit including the tachometer was inactive, meaning black, so the drivers had to estimate the vehicle speed by
their own perception.
Each driver completed ten laps at a time.
The steering feel
manipulation system was enabled or disabled randomly from lap to lap but consistent
for each lap. Bends with another vehicle ahead were discarded by help of a protocol.
The test leader was a passenger in the vehicle to operate the measurement equipment
and taking notes for the protocol.
6.6.5 Measurements
For measuring data, the on-board sensors of the test vehicle were used. Vehicle speed,
vehicle driving distance, lateral acceleration and yaw rate were measured with a sampling frequency of 1 to 100 Hz depending on the signal frequency on the CAN bus. The
bends were in advance dened by distance. A pre-test had shown that this method
worked properly.
6.6.6 Data analysis methods Tests on dierent populations
When testing driver assistance systems (DAS), the measurable dierences can be very
small smaller than the normal spreading of the measurement that comes to existence when a driver performs the same manoeuvre several times. Therefore, only a
77
6 Prototypes and experiments for active steering evaluation
statistical analysis can show whether a DAS leads to the desired behaviour.
This
section describes the statistics that was used in this experiment. The statistical methods are used to conrm dierences in driver performance by comparing distributions
of driver performance quantities. The used methods are described and discriminated
from methods nearby.
In Paper F the steering feel of a heavy truck was modied to test the general hypothesis
that this kind of inuence would lead to a change in driver behaviour. The hypothesis
for the experiment was:
Drivers of heavy trucks will perform with more safety margin to the rollover threshold if the steering wheel torque is decreased or additionally increased at high lateral
acceleration.
For the experiment a model was developed and tested that returned an indirect warning
to the driver via the steering wheel about too high vehicle speed. This was realised by
modifying the experienced steering wheel torque as function of the lateral acceleration.
The model was implemented in a test truck with active steering with torque overlay. In
the experiment 33 drivers drove 10 times a lap with 10 evaluable bends, partially with
active system, partially with inactive system.
External constraints lead to dierent
numbers of data for each curve and for each case.
So for each curve there is a set
of data containing a certain number of drivings for each driver with and without the
studied system applied.
The evaluation task is now to verify statistically that the
drivers in general reach a lower lateral acceleration in each bend when the system is
active than when the system is not active.
Kernel density estimation
The kernel density estimation computes the probability density function of e. g. an
observed lateral acceleration measurement.
The kernel density estimation is a non-
parametric method which means that there need not be the assumption that the data
2
are e. g. normally distributed.
The distribution is only assumed to be a continuous
function.
When comparing two sets of data one has to make sure that the kernel density estimation computes the function at the same points (starting point, interval and end point
have to be equal). Otherwise there could exist dierences resulting from the shifted
analysis intervals [63]. According to the
√
n-rule
the fragmentation of the estimation
interval should be below the root of number of values in the sample. This leads here
to a fragmentation of 1/6 on the interval
the mean lateral acceleration and a
fragmentation of 1/5 on the interval
[0, 4] for
[0, 5] for the
maximum lateral acceleration since
every sample consisted of around 750 values inside these intervals.
2 The
parameters for
parametric
and standard deviation
78
σ.
methods that assume normal distribution, are expected value
µ
6.6 Modied steering feel Track test
χ2 test
The
χ2
test of independence can distinguish two populations. However, it always needs
pairs of samples, this means pairwise points from both of the samples. For example,
in the track test (Paper F) this means that every case of a certain curve for a certain
driver with enabled system must be compared to a corresponding case for the same
curve and the same driver with disabled system.
This would only be true if every
driver would have driven once with the enabled system and once with disabled. But
in the present data set there are several drivings with enabled as well as with disabled
system available and they cannot be logically paired. Moreover, for some bends the
data are not representative because of another vehicle ahead. This situation makes it
impossible to decide whether the driver chooses the vehicle speed himself or whether
the vehicle speed is determined by the vehicle driving ahead. This means that there
usually are dierent numbers of cases with or without enabled system. One possibility
was, of course, to calculate the mean values for each driver, however, this reduces the
variance of the data and therewith the content of information. Moreover, the lateral
acceleration had to be divided into classes for this test which also reduces the variance
of data. However, a matrix for the
Table 6.2:
χ2
test of independence could look like in Table 6.2.
Possible distribution matrix of averaged lateral acceleration in all studied
bends for χ test of independence.
2
Lateral acc. without ROP
Lateral acc. with ROP
... 2
2 ... 3
3 ... 4
4 ...
2 m/s2
2 ... 3 m/s2
3 ... 4 m/s2
4 m/s2 ...
...
Student's t-test
The independent two-tailed t-test [64] is used for hypotheses testing about the expected values of two normal distributed sets of data with unknown standard deviation. It compares the dierence of the mean values of populations corresponding to
the t-distribution.
The t-test can only used for interval-scaled data.
The assump-
tion of normal distribution can be tested by means of the Shapiro-Wilk test or the
Kolmogorow-Smirnow test [65]. If these tests show that the populations are not normal distributed, the t-test can be substituted by the Wilcoxon-Mann-Whitney test
or a Wilcoxon-signed-rank-test.
Anyhow, the t-test is quite robust for not fullled
requirements as long as the number of data is large enough (n
> 30)
and the samples
do not dier too much in size [64].
79
6 Prototypes and experiments for active steering evaluation
Wilcoxon-Mann-Whitney test
The Wilcoxon-Mann-Whitney test [66] (also Mann-Whitney-U-Test or U-Test) requires
only continuous distribution functions and is based on the following principal: After
sorting the measured samples of the two distributions by size, the sum of the ranks
are to be compared.
If these dier signicantly, it can be assumed that one of the
distributions diers signicantly from the other.
Wilcoxon-signed-rank-test
The Wilcoxon signed rank test (SR-test) is an alternative for Student's t-test but it can
cope like the Wilcoxon-Mann-Whitney test with not normal distributed populations
[67]. The test can distinguish two populations by means of comparing their medians.
The calculation is based on a pairwise comparison and the sum of the ranked pairs'
dierences' sign. The sum is then compared to a tabled critical value depending on
the numbers of pairs
n.
In this work the SR-test is only limited meaningful since there
are not that clearly dened pairs (see
χ2
test).
F-test
The F-test [68] gives a condence for two samples coming from dierent normal distributed populations by comparing their variance due to the Fisher-distribution. The
F-test requires as well as the t-test normal distributed populations. However, if the
number of observations is situated beyond 50, the requirement of a normal distributed
population becomes more and more weak because of the central limit theorem [69].
So if the number of observations is large enough, the test results will be able to show
signicance anyway.
6.7 Results
The track test was assessed by evaluating the histogram data of lateral acceleration
mean and maximum values. Since the dierences were expected to be small, statistic
tests were performed (see subsection 6.6.6) to detect signicance between the samples.
The populations are visualised by means of the kernel density estimation which outputs the distribution function continuously.
The distribution function is calculated
separately for the laps driven with the reference and those with enabled system (Settings 1 and 2) and for the maximum and the average lateral acceleration. Fig. 6.8(b)
shows the kernel density estimation for the lateral acceleration maxima for the reference and Setting 1.
80
The most important section is the lateral acceleration interval
6.7 Results
Kernel density estimation of Lateral acceleration AVERAGES (10 curves, all drivers)
0.6
0.4
0.2
0.5
1
1.5
2
2.5
3
22]
Lateral acceleration average ay,mean
[m/s
m/s
Lateral acceleration [
]
3.5
4
(a) Absolute lateral acceleration AVERAGES.
Figure 6.8:
between
Probability
density
function
[1]
Number
of entities
[1]
Probability
density function [1]
Number of entities [1]
0.8
0
Reference
Setting 1
0.8
Signicance: No
1
0
0.9
Reference
Setting 1
1.2
Signicance: Yes
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0
0.5
1
1.5
2
2.5
3
3.5
4
22]
Lateral acceleration maxima ay,max
[m/s
m/s
Lateral acceleration [
]
4.5
5
(b) Absolute lateral acceleration MAXIMA.
Setting 1 kernel density estimation and its reference calculated from the
histogram data (10 curves, all drivers). The signicance statement origins
from the tests described in subsection 6.6.6.
2.0 m/s2
and
4.5 m/s2
where a decrease of entities at high acceleration can be
seen. The decrease of entities at high acceleration is obvious and marked as blue area.
Concurrently, an increase of entities from reference to Setting 1 at medium high lateral
acceleration (interval between
2 m/s2
and
3 m/s2 ,
marked as red area) indicates that the
drivers who before drove near the limit, decreased the vehicle speed and therewith the
maximum lateral acceleration.
The average values are more sensitive for driver failures which can come up while cornering especially when driving in the dark and on uncertain road condition. Moreover,
the beginning and the end of the bend with lateral acceleration near zero pull the
average down without increasing the information content.
Summarised for Setting 1 (decreased steering wheel torque) a signicant dierence in
maximum lateral acceleration with enabled system can be detected compared to the
reference (disabled system). For the mean lateral acceleration there is no signicant
dierence. The results from Setting 2 do not either show a similar aect to the drivers.
However, this shows that a there is need for a tight systematic to inuence the drivers
behaviour signicant and in the desirable direction.
81
6 Prototypes and experiments for active steering evaluation
82
7 Results papers in summary
This chapter presents the condensed results of this work and provides the summaries of the appended papers and the main results. Paper A and C describe
the performed experiments.
Paper B focusses on the word pool which was a
result of the lessons learned from the experiment in Paper A. Papers D and
E present further evaluations of the simulator experiment from Paper C. Paper F presents the track tests with modied steering feel by means of active
steering with the aim to inuencing the drivers' behaviour. Paper G presents
a functionality to make the truck more linear in its steering behaviour.
7.1 Word pool (Paper B)
In this work, words for the description of steering feel were dimensioned. This resulted
in up to nine dimensions describing the steering feel of road vehicles.
The dimen-
sions were found by multidimensional scaling and, together with the words which they
contain, they are listed in Table 3.2.
The manual dimensioning showed similar results, but better explained the complex
dimension Response.
The assumption that there are orthogonal dimensions in the
human-perceived space of steering feel could not be validated. However, some of the
dimensions are orthogonal but for the majority of the word pool a circle similar to
Russell's Circumplex Model of Aect [70] is the better mode of explanation.
If one
extracts two dimensions, they can be visualised as a circle (see Paper B). This will
make it easier to explain where certain words are placed even if they are not contained
in one of the dimensions.
Further on, the dimensioning by multidimensional scaling is insucient.
A certain
fuzziness can only be cleared by manual dimensioning or, probably, with the help of a
larger number of subjects. The remaining fuzziness of the presented word pool led to
a certain confusion in the simulator experiment, which produced an uncertain result of
one of the ve dimensions (Paper C). However, seven of nine dimensions were clear to
all test drivers even from dierent levels of experience. This supports the hypothesis of
this paper that there are multiple dimensions of human perception regarding steering
feel since the dimensions that people use to perceive steering feel were found and
the non-instrumental space was dened more understandably and in more detail than
before.
83
7 Results papers in summary
Question 1
easy heavy
Understeer
gradient
Torquemagnitude
Bandwidth 1013
Time delay 35
Torque
11
6
gradient
Response
deadband
0
20
59
56
40
60
1
100
100
76
1
Question 2
steering wheel return
6
41
73
78
100 0
20
40
60
Regression Analysis Relative Coecients
Figure 7.1:
87
19
93
80
Question 4
stable unstable
89
100
93
92
100
97
80 100 0
54
87
86
84
18
15
20
40
60
100
100
86
89
80 100
Neural Network Relative Weights
Comparison between NN relative weights and Regression analysis relative
coecients with normalised input data from the simulator experiment.
7.2 Correlation between handling and steering feel
(Papers A & C)
One of the hypotheses in the present research work is that there is a correlation between
steering feel, as perceived by drivers, and vehicle handling properties. To prove the
hypothesis, test drivers rated dierent steering characteristics of heavy trucks, both on
a test track (Paper A) and in a moving base driving simulator (Paper C). The same
steering characteristics were also tested in accordance with ISO-standards for vehicle
handling, resulting in characteristical instrumental quantities. The non-instrumental
ratings and the instrumental measurements were then analysed on correlation between
each other. The measured data from the track test showed insuciencies but the data
from the simulator could be used for evaluation. Analyses by means of Neural Networks
or Regression Analysis led to quite similar results. Fig. 7.1 shows the comparison. At
the same time, the gure also shows the inuence of each handling variable. Since the
input data is normalised, the dierent handling variables are also comparable to each
other.
The dimension easy heavy (Question 1) correlates most to the handling variables
Torque magnitude, Response deadband and Understeer gradient as can be seen in
Fig. 7.1 by means of the high coecients/weights. This is a comprehensible correlation
since torque will be experienced as force and both Torque magnitude and Response
deadband describe steering wheel torque.
The former is a quantity for torque at a
certain lateral acceleration. The latter is the torque that is necessary to raise the yaw
rate gain from zero and can be understood as the sum of all resistances that have to be
overcome before the vehicle shows a response. The correlation to Understeer gradient
is less obvious and hardly explainable since the Understeer gradient was modied by
a change of the torsion bar stiness which in the simulation model was independent of
the steering servo boost-curve. Therefore this is a result that needs to be veried, par-
84
7.2 Correlation between handling and steering feel
ticularly since this dimension had the largest uncertainty regarding the 95% condence
interval.
The dimension steering wheel return (Question 2) correlates most to the handling variables Bandwidth, Time delay, Torque gradient and Response deadband (see Fig. 7.1).
Torque gradient describes steering wheel torque as a function of lateral acceleration
which equates to the eect of the self-aligning torque. Moreover, Response deadband
actually describes the torque that is necessary to raise the yaw rate gain from zero.
This is, as mentioned above, understood as the sum of all resistances that have to be
overcome before the vehicle shows a response, but from this point of view it consists
of all parameters that restrain the steering wheel from returning. Further, Bandwidth
and Time delay are quantities for how fast a vehicle's response is. This may explain
the correlation.
than
10 %;
However, the dierence of range for these two quantities was lower
therefore, it is recommended that this correlation is validated in a future
test.
The dimension stable unstable (Question 4) correlates to all handling variables
except Torque gradient as visualised in Fig. 7.1. The correlation to Torque magnitude
is also comparatively low.
This indicates that stability, as perceived in this test, is
not a matter of torque but it is a matter of response (Bandwidth, Time delay and
Understeer gradient ) and parasitics (Response deadband ).
However, there were also dimensions that could not be explained by analysing these
measurements and ratings:
The dimension necessary steering wheel angle (Question 3), which was one of the
non-explained non-instrumental dimensions, was probably not covered enough by the
parameter changes. There was only the parameter stiness that inuenced the steering
wheel angle. The quantity Steady-state yaw rate gain, which is expected to correlate,
diers only in a range of less than
10 %.
This seems to be insucient.
Therefore
further investigations should oer a wider range of steering wheel angle modications.
The dimension direct indirect (Question 5), which was the other of the non-explained
non-instrumental dimensions, was probably insuciently exactly described (see section 7.1). The average deviation for Question 5 was up to
other questions.
40 %
higher than for the
This illustrates the higher grade of spreading when rating due to
Question 5. The dimension was quite complex and actually needed a more detailed
denition.
drivers.
This became clear during the experiment when discussing with the test
Unfortunately, this discussion did not arise in the pre-test.
However, the
description (in)direct is quite often used when discussing handling and steering feel;
therefore, this non-instrumental dimension must be claried in future work.
85
7 Results papers in summary
7.3 Correlation between vehicle handling and driver performance
in a moving base driving simulator (Paper D)
This work presents the evaluation of the relationship between driver performance and
vehicle handling quantities in an experiment using a moving base driving simulator.
Both of them are instrumental quantities; however, the former are driver-dependent,
7 Results papers in summary
the latter only vehicle-dependent. A moving base driving simulator was used to examine 16 truck-trailer combinations. The driving of 28 test drivers in a specially developed
manoeuvrecorrelations
resulted in in-between
characteristic
performance
values
evaluatedsteering
from vehicle
evaluating
the driver
instrumental
open- and
closed-loop
feel
system quantities
measured
while
Stationary and
dynamical
ISO-handling
measurements,
vehicle
handling
anddriving.
driver performance.
These
correlations
should
tests whether
resulted one
in handling
quantities.
regression
was
performed
evaluatshow
of them was
completelyAexplained
byanalysis
the others
which
would reduce
ing number
correlations
in-betweensteering
the instrumental
openand
closed-loop
steering
feel meathe
of instrumental
feel quantities
when
mapping
steering
feel. Table
2
surements,
vehicle
handling
and driver
performance.
These correlations
should are
show
in
Paper D shows
the
mathematical
results
from the evaluation.
Some correlations
whether one
of them not
wasall
completely
explained
the others,
wouldequations
reduce the
available.
However,
of them are
obviousbyneither
do allwhich
regression
2
number
of
instrumental
steering
feel
quantities
when
mapping
steering
feel.
represent a comprehensive model (see residuals R ). This can be reasoned byTable
lack of7.1
shows
the
mathematical
results
from
the
evaluation.
Some
correlations
are
available.
measured data or characteristic quantities or by non-linearities that are not represented
However,
not all of them are obvious, nor do all regression equations represent a comby
these models.
2
prehensive model (see residuals R ). This can be reasoned by lack of measured data or
The conclusion of this work is that for a comprehensive mapping of steering feel in
characteristic quantities or by non-linearities that are not represented by these models.
the instrumental domain both open-loop handling quantities and closed-loop driver
Alternatively, some drivers acted atypically since they tried out the steering feel, which
performance quantities are needed.
led to a wider spread-out of data and conceals the results.
Table 7.1:
Table 7.1:
DPQ
Regression
analysis
results
showing the
the averaged
coecientsDPQ
andexplained
R p -values of the
Regression
analysis
results
showing
by handling
analysis
with
averaged
DPQ
explained
by
handling
values.
values. For coecients, residuals and p-values see Table 2 in Paper D.
U nsymmetry3
δSWmax
δSWIF P S,P SD
MSWmean
2
explained by
Bandwidth
Time delay
Torque gradient (o
Time delay
Torque gradient (o
Response deadband
Understeer gradient
Time delay
Torque gradient (o
Understeer gradient
Bandwidth
Time delay
Torque gradient (o
DPQ
center)
center)
center)
δSWIF 2P S,F F T
MSWmax
MSWIF P S,P SD
ayRM S
ωzIF 2P S,F F T
center)
explained by
Understeer gradient
Bandwidth
Time delay
Torque magnitude
Torque gradient
Response deadband
Bandwidth
Time delay
Response deadband
Understeer gradient
Bandwidth
Response deadband
The conclusion of this work is that for a comprehensive mapping of steering feel in
the instrumental domain both open-loop handling quantities and closed-loop driver
7.4
Correlation between steering feel assessment and driver
performance quantities are needed.
performance in a moving base driving simulator (Paper E)
This paper describes how non-instrumental measurements (steering feel assessment
86
made
by human measurement gauges) and instrumental measurements (driver performance measured by instruments) can be distinguished in quantities that are dependent
of the vehicle, the driver's skills or the driver's individual preferences. Moreover, this
paper shows the completed gure illustrating the instrumental, non-instrumental as
7.4 Correlation between steering feel assessment and driver performance
7.4 Correlation between steering feel assessment and driver
performance in a moving base driving simulator (Paper E)
This paper describes how non-instrumental measurements (steering feel assessment
made by human measurement gauges) and instrumental measurements (driver performance measured by instruments) can be distinguished in quantities that are dependent
on the vehicle, the driver's skills or the driver's individual preferences. Moreover, this
paper shows the completed gure illustrating the instrumental, non-instrumental as
well as subjective and objective steering feel values. By means of a regression analysis
the averaged assessments were explained by the averaged driver performance values
(see Table 7.2).
Likewise, in this evaluation, three of the ve dimensions of human
steering feel perception could be explained.
Table 7.2:
Regression analysis results of the top-down stepwise extended analysis showing the averaged DPQ explaining averaged assessments. For details see
Table 3 in Paper E.
Question 1
Question 2
Question 3
Question 4
Question 5
easy - heavy
steering wheel
necessary
(un)stable
(in)direct
δSWChanges
aymean
MSWmax
MSWmean
ayIF P S,F F T
∆t(δSW , ωz )
MSWIF P S,P SD
U nsymmetry2
ayRM S
∆t(δSW , ωz )
aymax
aymean
return
MSWmax
δSWIF 2P S,F F T
δSWChanges
U nsymmetry2
MSWIF P S,P SD
ayRM S
∆t(δSW , ωz )
SW angle
aymean
ayRM S
δSWIF 2P S,F F T
MSWmean
MSWmax
δSWIP S,P SD
aymax
aymean
MSWmean
aymax
U nsymmetry1
The dimension of human perception easy - heavy can be explained with 82% by the
maximum steering wheel torque
MSWmax
and the integrated spectrum (FFT) of the
δSWIF 2P S,F F T . Other
R2 only insignicantly.
steering wheel angle multiplied with the square of the frequency
performance values increase the coecient of determination
The correlation to the steering wheel torque is rather intuitive and conrms the method.
The signicant inuence of
δSWIF 2P S,F F T
points to the fact that an easy or heavy
steering also depends on the steering activity.
The dimension steering wheel return is dominated by lateral acceleration, steering
activity and steering wheel torque. Especially the latter is self-explanatory; however,
the former could also point to the drivers' play with the steering with the aim of nding
the answer to the requested dimension.
87
7 Results papers in summary
The dimension stability can be explained with 84% by the number of changes of
steering wheel rotation direction, lateral acceleration and steering wheel torque. The
value
δSWChanges
was introduced with the hypothesis that it could be a measure for
stability. However, the coecient is quite low so the inuence of
δSWChanges
can be
interpreted as low, too. The lateral acceleration and especially its amplitude spectrum
show the reaction of the vehicle which can be interpreted as disturbing when asking
for stability.
In fact, both decreasing the integrated spectrum (FFT) of the lateral
acceleration multiplied by the frequency
acceleration
aymean
ayIF P S,F F T
and a decreasing mean lateral
led to the assessment more stable.
In addition, the fact that
an increasing steering wheel torque increases the perceived stability coincides with
experiences from earlier driving tests.
Grouping the drivers by means of a cluster analysis due to their assessment and therefore due to their way of perception, was not successful. Nor could an evaluation of a
personal corridor (see subsection 3.6.3) be correlated to other metrics.
7.5 Inuencing driver behaviour by means of steering feel
(Paper F)
This paper presents an investigation into inuencing the driver's behaviour intuitively
by means of modied steering feel. For a rollover indication through haptic feedback, a
model was developed and tested that returned a warning to the driver about too high
vehicle speed. This was realised by modifying the experienced steering wheel torque
as function of the lateral acceleration. The hypothesis for this work was that drivers
of heavy vehicles will perform with more safety margin to the rollover threshold if the
steering feel is manipulated by means of decreased or additionally increased steering
wheel torque at high lateral acceleration. Therefore, the model was implemented in a
test truck with active steering with torque overlay and used for a track test. 33 drivers
took part in the experiment. The investigation showed that a decreasing steering wheel
torque which is experienced as driving on an ice-patch, leads to a signicant decrease
of lateral acceleration while cornering (see Fig. 7.2).
For an increasing steering wheel torque, no dierence to the reference could be found.
The main contribution of this paper is the nding that the driver behaviour can be
inuenced by means of systematic steering feel modication.
7.6 Articial understeer handling test investigation (Paper G)
The main contribution of this paper is that it shows the advantages of a PID controlled
active front steering system based on linear calculated yaw rate error in a heavy truck.
The hypothesis for this function (Fig. 7.3) is that the vehicle handling is improved
88
CT
Probability
[1]
Numberdensity
of entities function
[1]
celeration range just before roll over, heavy trucks tend to oversteering. This is a result
ce focussing on mileage on the front and traction on the rear axle. This behaviour of
t low lateral acceleration and neutral to oversteering
at higher
lateral
acceleration
makes
7.6 Articial
understeer
handling
test investigation
predictable for the driver. The unpredictability astonishes the driver since it happens quite
the vehicle at that high lateral acceleration. Therefore, it is even more important in such
0.9
situations that the vehicle is controllable
meaning consistent andReference
thereby predictable.
0.8
teering system in a truck is characterised
by a quite long distance Setting
from1 the steering wheel
Signicance:
0.7
wheels. Even with a high stiness
of the system, there will
exist aYescertain distorsion.
oints have at least some kind of0.6 play, even if the driver does not experience this as
lly, truck tyres are usually high-prole
tyres (e. g. 315/80R22.5 or 315/70R22.5) which in
0.5
th the above named non-linearities
leads
to a delay of the vehicle reaction relative to the
0.4
nput. This steering wheel angle deadband around yaw rate and lateral acceleration equal
0.3
lly δSWdeadband = ±10 ◦ for heavy trucks.
here meaning the superposition 0.2of steering angle or changing the steering ratio in some
0.1
govern the transfer function between
steering wheel angle as the driver's input to the
0
and the vehicle's response. Inuencing
this
transfer
function
means
to change
the steering
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
2 ]]
Lateral
acceleration
maxima a [m/s
[m/s
Lateral
acceleration
le.
estions in this investigation is then how to test the changes of the transfer function? The
Figure
7.2:
preparation
for further
investigations with track tests.
ents a functionality that changes the steering feel by enabling articial understeering and
test methods for its investigation.
y,max
2
Absolute lateral acceleration maxima kernel density estimation for Setting 1 and its reference calculated from the histogram data (10 curves, all
drivers). The signicance statement origins from several statistic methods
(see Table 2 in Paper F).
Understeering
when the vehicle reacts more linearly, which is expected to be more predictable by the
driver. The
idea of the
function
to shorten that
the vehicle's
al Understeering
functionality
there
is theisassumption
the driverresponse
has sometime
kindbyofmeans
superposition
steering
angle, which
creates
an overshoot
at the
beginning
model inofhis
mind. The of
more
experienced
the driver
is, the
better is the
model
in his of a
for most
of theoperation.
drivers these
models
are only linear.
avoid non-linear
steering
When
the expected
vehicleTo
response
is reached,behaviour
the angle of
overlay
measured
yaw rate In
of this
the vehicle
is compared
a linear are
calculated
reference
yaw rate.
is reversed.
paper the
evaluationtomethods
investigated
in a software-in-thef these two
rates is then
reduced byfor
means
of driving
a PID-controller
which outputs control
loopyaw
simulation
as a pre-study
future
tests.
steering system with superposition of steering angle.
δSW
re are four inputs:
wheel angle (calculated
heel steering angle)
w rate (measured)
Figure 7.3:
δ
vx ωz,Ref
Kus
vx
Kus
eed (speed adaptation)
understeer gradient
i
Reference ωz
+
−
PID
δadd
ωz
Simplied
model Model
of articialofundersteering
For deFigure Simulink
1: Simulink
Articial functionality.
Understeering
tails see Fig. 6.3.
functionality.
The integrated controlled system shows increased performance: Fig. 6.4 shows that
the vehicle reacts more quickly and more linear-like over a wide lateral acceleration
1
range, which increases safety especially in emergency conditions.
89
7 Results papers in summary
90
8 Scientic contribution
This chapter lists the main scientic contributions of the thesis and its appended
papers:
1. The framework for the mapping of steering feel, distinguishing subjective/objective
from instrumental/non-instrumental measurements (Fig. 1.1). The dierence between ratings and descriptive assessments is an important fact there. The results
were input for the design of the presented simulator experiment with appropriate
drivers and a proper manoeuvre. The outcomes support the rst hypothesis.
2. The word pool for the comprehensive description of steering feel classied in
up to nine dimensions (Paper B) including the strategy of Word pool mining,
Word pool election and Word pool clustering with its statistics. This supports
the second hypothesis.
3. Mapping a part of the multidimensional space of steering feel by means of dierent evaluation methods: Regression analysis and neural networks according to
the framework in Fig. 1.1 (Papers C, D and E). Correlations between instrumental and non-instrumental values can then be shown (corresponding to the rst
hypothesis).
4. Inuencing the drivers' behaviour by means of modied steering feel (Paper F).
By means of active steering, the steering wheel torque was modied as a function
of lateral acceleration, which has been shown to lead to decreased maximum
lateral acceleration values. This decreases the risk for rollover and corresponds
to the third hypothesis.
5. A method for articial understeering and acceleration of vehicle reaction (Paper G & patent [60]). By means of superposition of steering angle articial understeering could be realised, which leads to a more linear vehicle behaviour and
faster vehicle response, which was tested by software-in-the-loop (corresponding
to the third hypothesis).
91
8 Scientic contribution
92
9 Conclusions and recommendations to future work
This chapter draws conclusions from the results chapter and provides a perspective for further research work.
9.1 Conclusions
The rst hypothesis for this work is: There exists a correlation between instrumental
handling values and non-instrumental human perceived values regarding steering feel!
The instrumental and non-instrumental values described in chapter 3 were combined
in chapter 4 and resulted in several correlations (see chapter 7).
Even if some of
the described non-instrumental dimensions could not be explained during this work,
certain correlations could be found such as between the non-instrumental Steering
wheel return and the instrumental Response deadband, Torque gradient, Time delay
and Bandwidth. This supports the rst hypothesis.
En route, a method for nding a word pool for the description of steering feel was
developed.
The experience from the simulator test showed that the resulting word
pool worked properly, which supports the second hypothesis.
This is an important
tool that, in addition, also can be used when widening the focus to steering comfort,
for instance, or to an even wider driver experience.
It followed the application orientated hypothesis that the drivers' behaviour can be
inuenced by means of modied steering feel. An experiment was performed where
the drivers' assessment was deceived by a controlled change of steering feel (see section 6.6 and Paper F). Depending on the group of drivers, the steering wheel torque
was increased or decreased discontinuously as a function of lateral acceleration. The
steering feel modication was meant as an intuitive warning for rollover. The results
in Paper F show that the drivers chose to drive with a decreased maximum lateral
acceleration when the function decreased the steering wheel torque, which decreases
the risk for rollover. For future work it can be concluded that the driver behaviour
in general can be inuenced by systematic steering feel modication. Combinations
with other driver assistance systems may increase the collective eectiveness of the
systems. Especially in heavy trucks where the same driver operates the vehicle over
many hours, a higher adaptation (self-learning) and personalisation could be realised.
The change in driver behaviour indicates future potential for this kind of personalised
driver assistance systems.
Thus, on the basis of knowledge about steering feel, an application-oriented hypothesis
was formulated and evaluated. The fundamental part of this thesis work contributed a
93
9 Conclusions and recommendations to future work
piece of the puzzle to the mapping of steering feel while the advanced part established
ties to future applications.
9.2 Future work
9.2.1 Word pool and extending the mapping of steering feel
A suggestion for future work on the word pool tool presented in Paper B is to clarify the
dimension that was used for Question 5 in Paper C. Question 5 was labelled with the
words direct, indirect but contained also the words controlled, delayed, distinct, erratic,
obedient, quick, reactive, precise, (in)exact, slippery, sensitive. This was obviously too
wide and comprehensive.
Furthermore, it is still an uncertainty about whether all necessary instrumental variables are covered.
Hence, if an important variable is missing, it should be added.
Moreover, the parameters should be modied even more and one should ask for more
dimensions to cover a wider range of the non-instrumental space of steering feel, as
described in Paper B.
In this work the mapping of correlations between handling variables and steering feel
has been started (see chapter 7). However, in the simulator experiment only ve of the
nine questions that were extracted in Paper B were analysed. Moreover, only three of
them could be explained. Therefore further experiments - in the simulator and on the
test track - are suggested to map the whole space of steering feel. Furthermore, there
must be a review of the non-instrumental dimension as well as the handling variables
to make sure that all aspects are covered.
For track tests a vehicle with full active
steering, which means superposition of torque and angle, alternatively with steer-bywire will be necessary to cover the full spectrum and to avoid the simulator feeling for
the driver.
The dimensions for the non-instrumental description of steering feel were chosen as a
list of descriptive words. Other descriptions would be possible e. g. a list of metaphoric
expressions such as Steering feel like driving on an icy puddle or Steering feel like
driving over a curb [71]. This could lead to supplementary or an alternative kind of
mapping.
9.2.2 Modied steering feel
The modied steering feel as presented in Paper F is based on the drivers' experience of
driving on slippery road conditions. At high lateral acceleration, the necessary steering
torque applied by the driver is reduced by means of torque overlay which simulates
a lower self-aligning torque.
94
This is interpreted by experienced drivers as icy road,
9.2 Future work
for instance. Depending on the region worldwide where drivers could use a vehicle, it
is possible that the driver has never experienced vehicle behaviour on slippery roads.
Possibly, the driver could not react properly because of lack of this special kind of
experience. This should be taken into consideration when implementing this kind of
modied steering feel. A possible next stage might be to combine modied steering
feel with driver skill detection [72] to develop a system that chooses the best warning
case depending on the driver.
9.2.3 Driver behaviour manipulation by means of superposition of steering angle
In Paper F the inuence on the driver behaviour was presented with a rollover indication system by means of superposed steering torque. The logical next step is to test
the functionality with active steering with superposition of steering angle (see Fig. 9.1)
for dierent drivers in dierent manoeuvres.
Sign
ay
vx
Gain
×
×
÷
×
δadd
Dead Zone
Factor
Figure 9.1:
Matlab Simulink model of rollover indication functionality for superposition of steering angle.
Depending on the sign, the vehicle will become more understeered or more oversteered.
The more understeered variant is expected to be more intuitive.
In contrast to the
torque overlay system, a speed adaptation will be necessary for an added steering
angle!
So, as future work a general extension of knowledge is suggested e. g. with superposition
of angle and possibly with a repetition of the described test. In a wider perspective
one should investigate how a systematic modication of steering feel could be utilised
in other applications.
9.2.4 Triggering the driver's reaction
The very interesting results from Bender [31] (see subsection 5.5.1) about triggering the
driver's reaction for obstacle avoidance manoeuvres should be supplemented. Bender
shows that an advanced driver assistance system (ADAS) can help by triggering the
95
9 Conclusions and recommendations to future work
driver to perform the necessary manoeuvre. However, the experiment was driven with
an obstacle that suddenly appeared from the roadside so the direction or an avoidance
manoeuvre was obvious. Thus, the question still remains whether the trigger function
would work even in cases where the driver had to decide whether to manoeuvre left or
right.
96
Bibliography
[1] M. Rothhämel. Capturing steering feel - A step towards implementation of active
steering in heavy vehicles. Licentiate thesis, KTH Vehicle Dynamics, 2010. trita-
ave 2010:57.
[2] P. Bösch. Der Fahrer als Regler. PhD thesis, Technical University Vienna, 1991.
[3] International Standard ISO/TR 8726:
Road vehicles Transient open-loop re-
sponse test method with pseudo-random steering input, 1998.
[4] International Standard ISO 14792: Heavy commercial vehicles Steady-state cir-
cular test, 2003.
[5] International Standard ISO 13674-2: Road vehicles Test method for the quan-
tication of on-centre handling Part 2: Transition test, 2006.
[6] International Standard ISO 13674-1: Road vehicles Test method for the quan-
tication of on-centre handling Part 1: Weave test, 2003.
[7] M. Juhlin.
Assessment of crosswind performance of buses.
PhD thesis, KTH
Vehicle Dynamics, 2009. trita-ave 2009:25.
[8] P. Pfeer and M. Harrer. Lenkungshandbuch. Vieweg & Teubner, 2011.
[9] Uniform provisions concerning the approval of vehicles with regard to steering
equipment, 2005. UN/ECE Addendum 78, Regulation 79.
[10] P. F. Sweatman and P. N. Joubert. Detection of changes in automobile steering
sensitivity. Human Factors, 16(1), 1974.
[11] Servo
eort
curve
http://i979.photobucket.com/albums/ae275/Paneuropean/
ServotronicEortCurve.jpg.
[12] L. J. K. Setright. The mythology of steering feel. Automotive Engineer, (5), 1999.
[13] L. J. K. Setright. Steering feel is a myth and the steering wheel should be ditched
in favour of the joystick. Automotive Engineer, (6), 1999.
[14] F. Dorsch. Psychologisches Wörterbuch. 10, 1982.
[15] H. Wolf. Untersuchung des Lenkgefühls von Personenkraftwagen unter besonderer
Berücksichtigung ergonomischer Erkenntnisse und Methoden. PhD thesis, Technical University Munich, 2008.
97
Bibliography
[16] A. Neukum and H.-P. Krüger.
Fahrerreaktionen bei Lenkstörungen - Unter-
suchungsmethoden und Bewertungskriterien. VDI Berichte, (1791), 2003.
[17] B. Buschardt.
Synthetische Lenkmomente.
PhD thesis, Technical University
Berlin, 2003.
[18] T. Barthenheier. Subjektive Fahreindrücke sichbar machen III, chapter Potenzial
einer individuellen Lenkmomentgestaltung. Expert Verlag, 2006.
[19] T. Barthenheier. Potenzial einer fahrertyp- und fahrsituationsabhängigen Lenkrad-
momentgestaltung. PhD thesis, Technical University Darmstadt, 2004.
[20] G. Schmidt. Wann spürt der Fahrer überhaupt? VDI-Berichte, (2015), 2007.
[21] K.-H. Deppermann. Fahrversuche und Berechnungen zum Geradeauslauf von Per-
sonenkraftwagen. PhD thesis, Technical University Braunschweig, 1989.
[22] G. T. Fechner. Elemente der Psychophysik Teil 1. Breitkopf & Härtel, 1860.
[23] M. Koide and S. Kawakami. Analysis of steering feel evaluation in vehicles with
power steering. JSAE Review, 9(3), 1988.
[24] M. J. van Randwijk. Correlation of driver judgements and vehicle directional data
to evaluate and predict truck handling. In Proceedings of 3
rd
EAEC International
Conference Vehicle Dynamics and Powertrain Engineering, number EAEC No.
91054, Strasbourg, 1991.
[25] A. Riedel, R. Gnadler, and K. Dibbern. Bewertungskriterien zur Fahrverhaltensanalyse in Versuch und Simulation (Ermittlung neuer Kennwerte für den ISOSpurwechsel). VDI Berichte, (1007):775 799, 1992.
[26] A. Zomotor. Fahrwerktechnik: Fahrverhalten. Vogel Verlag, 1991.
[27] W. Linke, B. Richter, and R. Schmidt. Simulation and measurement of driver vehicle handling performance. Automobile Engineering Meeting Detroit, (SAE730489),
1973.
[28] F. O. Jaksch.
Driver-vehicle interaction with respect to steering controllability.
Passanger Car Meeting Hyatt Regency, Dearborn, (SAE790740), 1979.
[29] D. C. Chen and D. A. Crolla. Vehicle handling behaviour: subjective v. objective
comparisons. (F98T210), 1998.
[30] M. Agebro. Driver Preferences of Steering Characteristics. Licentiate thesis, KTH
Vehicle Dynamics, 2007. trita-ave 2007:66.
[31] E. Bender.
Handlungen und Subjektivurteile von Kraftfahrzeugführern bei au-
tomatischen Brems- und Leinkeingrien eines Unterstützungssystems zur Kollisionsvermeidung. PhD thesis, Technical University Darmstadt, 2008.
98
Bibliography
[32] S. Buld and H.-P. Krüger. Wirkung von Assistenz und Automation auf Fahrerzustand und Fahrsicherheit. Final report of the project EMPHASIS.
[33] W. Götze, C. Deutschmann, and H. Link. Statistik. Oldenbourg Wissenschaftsverlag, 2002.
[34] A. Nilsson, M. Agebro, and A. Stensson Trigell.
Study of path tracking skill
and strategy using a moving base simulator. In FISITA Transactions, Yokohama,
2006. F2006D075T.
[35] A. Neukum, H.-P. Krüger, and J. Schuller. Der Fahrer als Messinstrument für
fahrdynamische Eigenschaften? VDI Berichte, (1613), 2001.
[36] C. Demant, B. Streicher-Abel, and P. Waszkewitz. Industrielle Bildverarbeitung:
Wie optische Qualitätskontrolle wirklich funktioniert. Springer-Verlag, 2001.
[37] G. D. Rey and K. F. Wender. Neuronale Netze. Hans Huber Verlag, 2008.
[38] International Standard ISO 15037-2:
Road vehicles - Dynamic test methods Part 2: General conditions for heavy vehicles and busses, 2002.
[39] E. Sagan. Ansätze zur objektiven Fahrdynamik. Darmstädter Kolloquium Mensch
und Fahrzeug, 2007.
[40] S. Gies. one-to-one interview. FISITA World Automotive congress, Munich, 2008.
[41] VTI Driving Simulator II. www.vti.se.
[42] N. Dela, L. Laine, F. Bruzelius, H. Sehammar, L. Renner, G. Markkula, and A.-S.
Karlsson.
A pilot evaluation of using large movement driving simulator experi-
ments to study driver behaviour inuence on active safety systems for commercial
heavy vehicles. In Proceedings of 21
st International Symposium on Vehicle System
Dynamics, Stockholm, Sweden, 2009.
[43] G. Dreisbach. one-to-one email-interview. Institute for Psychology, University of
Regensburg, 2010.
[44] M. Spreng. Handbuch der Ergonomie. chapter A-2.4.3, page 9. Bundesamt für
Wehrtechnik und Beschaung, 1975-2002.
[45] M. Schuster, M. Grupp, T. Richter, and M. Pischinger.
Die Aktivlenkung des
neuen BMW 3er. Sonderausgabe von ATZ und MTZ BMW 3er, 2005.
[46] M. Reuter and S. Ullmann. Dynamic steering Aerospace technology in vehicle
application. ATZautotechnology, 8, 2008.
www.lexus.com.bh/technology_
explorer/variable_gear_ratio_steering_VGRS.asp?model=All.
[47] Lexus vgrs - variable gear ratio steering.
[48]
www.bmw.com.tw/com/en/insights/technology/technology_guide/
articles/_shared/img/mm_active_steering.jpg.
99
Bibliography
[49] Harmonic
drive
operating
operatingprinciples/.
principles.
www.harmonicdrive.net/reference/
[50] M. Rothhämel (Inventor) and Scania (Assignee).
Patent: Anordning för aktiv
styrning av ett lastfordon och styrinrättning med sådan anordning, 6th September
2011. SE 534 469 C2, Sweden 2011.
[51] U. Wiesel, A. Schwarzhaupt, M. Frey, and F. Gauterin.
reducing fuel consumption of commercial vehicles.
Hybrid steering for
ATZ Worldwide, (1):5258,
2010.
[52] D. E. Williams.
Synthetic torque feedback to improve heavy vehicle drivability.
223(12):15171527, 2009.
[53] M. Rösth.
Hydraulic power steering system design in road vehicles - Analysis,
testing and enhanced functionality. PhD thesis, Linköping University, 2007.
[54] Patent: Anordning för aktiv styrning och servostyrning med sådan anordning.
[55] iHSA
intelligent
Hydraulic
Steering
Assist.
intelligent-hydraulic-steering-assist.html.
www.td-steering.com/
[56] International Standard ISO 26262: Road vehicles Functional safety, 2012.
[57] F. Asp. Scania internal report, 2006.
[58] H.-P. Krüger, A. Neukum, and J. Schuller.
Bewertungen von Fahrzeugeigen-
schaften vom Fahrgefühl zum Fahrergefühl. In Proceedings of 3
rd Berliner Werk-
statt Mensch-Maschine-Systeme, in VDI-Fortschritt-Berichte Reihe 22, 1999.
[59] The mathworks, inc.
www.mathworks.se/index.html.
[60] M. Rothhämel (Inventor) and Scania (Assignee). Patent: Styrreaktionsförstärkning, 24th April 2012. submitted, Sweden, 2012.
[61] E. Schindler. Fahrdynamik: Grundlagen des Lenkverhaltens und ihre Anwendung
für Fahrzeugregelsysteme. Kontakt & Studium. Expert-Verlag, 2007.
[62] H. B. Pacejka. Tyre and Vehicle Dynamics. Butterworth-Heinemann, 2006.
[63] U. Kohler and F. Kreuter.
Datenanalyse mit Stata: allgemeine Konzepte der
Datenanalyse und ihre praktische Anwendung. Oldenbourg Wissenschaftsverlag,
2008.
[64] B. Rasch, M. Friese, W. J. Hofmann, and E Naumann. Quantitative Methoden 1:
Einführung in die Statistik für Psychologen und Sozialwissenschaftler. Springer,
2010.
[65] R. M. Feldman and C. Valdez-Flores. Applied probability and stochastics. Springer,
2nd edition, 1995.
100
Bibliography
[66] M. E. Stokes, C. S. Davis, and G. G. Koch. Categorical data anlysis using the SAS
system. SAS Institute Inc., 2nd edition, 2000.
[67] R Lyman Ott and M. Longnecker. An introduction to statistical methods and data
analysis. Brooks/Cole, 6th edition, 2010.
[68] J. Bortz. Statistik: Für Human- und Sozialwissenschaftler. Springer.
[69] J. Stevens. Intermediate Statistics. A Modern Approach. Erlbaum, 1999.
[70] J. A. Russel.
A circumplex model of aect.
Journal of Personality and Social
Psychology, (39):1161 1178, 1980.
[71] J. IJkema. one-to-one interviews, 2006-2013.
[72] A. Erséus. Driver-vehicle interaction Identication, characterization and mod-
elling of path tracking skill. PhD thesis, KTH Vehicle Dynamics, 2010. trita-ave
2010:29.
101
Bibliography
102
Appendix
Appendix A
Nomenclature
Notation
Parameter
Unit
Description
ay
ajq
f
i
k
n
ns
pkq
ps
t
t0(x,y)
Akj
C
F
I
Itot
Iwheel
K Ψ̇
Kus
KD
L
M
ML,0
ML
ML,stat
ML,dyn
ML,rel
ML,rel,dyn
MSW
m/s2
Lateral acceleration
1 According
Loading factor (Principal Component Analysis)
Hz
Frequency
Ratio (of steering gear)
Linear factor in correlation
Number of teeth (gear boxes)
Number of Setting (Contingency analysis)
Factor in principal component analysis
Hydraulic pressure in steering system
s
Time
s
Time delay between the metrics x and y
Composed Assessment (Principal Component Analysis)
Coecient of contingency
From F-distribution
kg m2
kg m2
kg m2
1/s
◦/ms−2
Inertia
Summed Inertia
Inertia of the front wheels including the steering linkage
Yaw rate gain (referred to steering wheel angle)
Understeer gradient
Characteristic value for double lane change
m
Wheel base
Measurement
Nm
Nm
Nm
Nm
Nm
Nm
Nm
1
Friction torque at steering wheel
Breakaway torque at steering wheel
1
1
Self-aligning torque at steering wheel
1
1
wheel
Dynamic self-aligning torque at steering wheel
Relative self-aligning torque at steering
Dynamic relative self-aligning torque at steering wheel
Steering wheel torque
to Deppermann, see section 3.2.2.
1
Q
R
α
β
δ
δr
δl
δ0
δSW
δSW0
ωz
χ
∆
Ψ
2
Assessment as answer to a question
m
Radius in Ackermann geometry
◦/rad
Roll angle
◦/rad
◦
◦
Body slip angle
Steering angle
Steering angle right front wheel
◦
Steering angle left front wheel
◦
Theoretical steering angle (Ackermann)
◦
Steering wheel angle
◦
Theoretical steering wheel angle (Ackermann)
◦/s
Yaw rate
Parameter dened from
χ2
test
Any dierence
◦
Yaw angle
List of abbreviations
ABS
Antilock Braking System
ADAS
Advanced Driver Assistance System
AFS
Active Front Steering
ASIL
Automotive Safety Integrity Level
3
CoG
Center of Gravity
DAS
Driver Assistance System
EAS
Electrically Assisted Steering
ECU
Electronic Control Unit
ESC
Electronic Stability Control
4
EPS
Electric Power Steering
EHPS
Electro-Hydraulic Power Steering
HD
Harmonic Drive
HPS
Hydraulic Power Steering
iHSA
intelligent Hydraulic Steering Assist
JND
Just Noticable Dierence
NN
(articial) Neural Network
PCA
Principal Component Analysis
RA
Regression Analysis
SbW
Steer-by-Wire
SW
Steering Wheel
TRW
TRW Automotive (Thompson Ramo Wooldridge)
ZF
ZF Friedrichshafen AG (ZahnradFabrik)
ZFLS
ZF Lenksysteme GmbH
2 In Paper C even rating instead of assessment.
3 Meaning active front steering with superposition
4 According to ISO 26262.
106
of steering angle in this thesis work.
Appendix B
Schematic diagram of hardware wiring for angle overlay
107
Scania
red CAN
250kBaud
standard
Scania
Sensor CAN
(Plausibility
check))
ωz_Reference
from Sensor CAN
ESP-Intervention from red CAN
δSC, ay, ωz
from red CAN (20ms)
values are plausibility checked
vx from red CAN (50ms)
Truck
Active Steering Simulink
Model
com
mpile
-
UDC=12V
+
OEM SW angle dependent AFS compensation
OEM emulation d t
data generator
t
Active Steering Simulink Models
COO7
Parameter
‐ wheelbase
‐ mech. ratio
UDC=12V
(δadd, OEM emulation
data part 1)
(δSW)
- AFS Characteristic
- Ignition
- VIN
environment:
Emulation of OEM
COO6
-
User Interface
(portable PC)
+
(Param
meters,
Diagnos CAN
(OEM emulation
data part2)
UDC=12V
+
(δSW)
(δActuator)
(3 Fas)
El t i
Electric
motor
OEM “Smart Actuator”
UDC=12V
-
OEM
controller
OEM Gateway
OEM
comfort-CAN
100kBaud
Standard
DLC variable
i bl
OEM
S-CAN
(sharing D-CAN)
500kBaud
Standard
DLC variable
Emergency Off
‐ short‐circuiting CAN (?)
‐ Switching off voltage to OEM controller
(δadd)
(δActuator)
(δSW)
OEM
D-CAN
(sharing S-CAN)
500kBaud
Standard
DLC variable
-
Output Plaussibility Controll:
Lim
miter as functio
on of vehicle speed
Block output aand report errror
B
-
UAC=220V
SW
Scania
δSC‐Sensor
Harmonic
H
i
drive
OEM δSW‐Sensor
OEM‐ADS
OEM OBD
UDC=12V
UDC=24V
OEM Actuator:
- Mechanical ratio: 1,02
- Blocked to mechanical ratio
without current (fail-safe)
- Blocked to mechanical ratio
if the
th controller
t ll detects
d t t a
failure or implausible signals
(fail-safe)
+
+
+
+
Log-da
ata)
No In
nput Plausibilitty Control neccessary
Appended Papers
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement