10th IFAC Symposium on Intelligent Autonomous Vehicles

10th IFAC Symposium on Intelligent Autonomous Vehicles

Gdansk,
10th IFACPoland,
10th IFAC July 3-5,
Symposium
Symposium on 2019
on Intelligent
Intelligent Autonomous
Autonomous Vehicles
Vehicles
Gdansk, Available online at www.sciencedirect.com
10th IFACPoland,
Gdansk,
Gdansk, Poland,
Poland,
July 3-5,
Symposium
July
July 3-5,
3-5,
2019
on Intelligent
2019
2019 Autonomous Vehicles
Gdansk, Poland, July 3-5, 2019
ScienceDirect
IFAC PapersOnLine 52-8 (2019) 136–141
Tire-Road
Tire-Road Friction Coefficient Estimation
Tire-Road Friction
Friction Coefficient
Coefficient Estimation
Estimation
under
Tire-Road
under Constant Vehicle
Friction Speed
Coefficient Control
Estimation
under Constant Vehicle Speed Control
Constant Vehicle Speed Control
under Constant Vehicle∗∗ Speed Control
Juqi Hu Subhash Rakheja Youmin Zhang ∗∗∗
∗
Juqi Hu ∗∗∗ Subhash Rakheja ∗∗ ∗∗ Youmin Zhang ∗∗∗
∗∗∗
∗∗∗
Juqi Hu
Juqi Hu ∗ Subhash
Subhash Rakheja Rakheja ∗∗ ∗∗
Youmin
Youmin Zhang
Zhang ∗∗∗
∗ Juqi Hu Subhash Rakheja Youmin Zhang
∗ Department of Mechanical, Industrial and Aerospace Engineering,
∗ Department of Mechanical, Industrial and Aerospace Engineering,
∗ Department
∗ Department
of
of Mechanical,
Concordia University,
Mechanical, Industrial
Montreal,
Industrial and
and Aerospace
Canada,
Aerospace Engineering,
(e-mail:
Engineering,
Department Concordia
of University,
Mechanical, Montreal,
Industrial andCanada,
Aerospace (e-mail:
Engineering,
Concordia University, Montreal,
[email protected]).
Concordia University, Canada,
Montreal, Canada, (e-mail: (e-mail:
∗∗ [email protected]).
Concordia University, Montreal,and Canada, (e-mail:
∗∗ Department of Mechanical, [email protected]).
Industrial
∗∗ Department of Mechanical, Industrial and Aerospace Engineering,
[email protected]).
[email protected]).
Aerospace Engineering,
∗∗ Department
Department of
Concordia
of Mechanical,
University,
Mechanical, Industrial
Montreal,
Industrial and Aerospace
Canada,
and Aerospace Engineering,
(e-mail:
Engineering,
∗∗
Department Concordia University,
of Mechanical, Montreal,
Industrial Canada,
and Aerospace (e-mail:
Engineering,
Concordia
Concordia University, Montreal,
[email protected])
University, Montreal, Canada,
Canada, (e-mail:
(e-mail:
∗∗∗ Concordia [email protected])
University, Montreal, Canada, (e-mail:
Department of [email protected])
Mechanical, Industrial
∗∗∗ Department of Mechanical,
∗∗∗ Industrial and Aerospace Engineering,
[email protected])
[email protected])
and Aerospace Engineering,
∗∗∗ Department
DepartmentConcordiaof
of Mechanical,
University,
Mechanical, Industrial
Montreal,
Industrial and Aerospace
Canada,
and Aerospace Engineering,
(e-mail:
Engineering,
∗∗∗
DepartmentConcordia University,
of Mechanical, Montreal, Canada, (e-mail:
Concordia
Concordia University,Industrial
Montreal,and
[email protected])
University, Montreal, Aerospace
Canada,
Canada, Engineering,
(e-mail:
(e-mail:
Concordia [email protected])
University, Montreal,
[email protected])
[email protected]) Canada, (e-mail:
[email protected])
Abstract: Tire-road friction
Abstract: Tire-road friction coefficient (TRFC)coefficient (TRFC) is is of
of critical
critical importance
importance for for advanced
advanced vehicle vehicle
Abstract:
control
Abstract: Tire-road
systems
Tire-roadand friction
autonomous
friction coefficient
vehicle
coefficient (TRFC)
(TRFC) is
is of
applications.
of critical
This
critical importance
paper
importancepresents for
for advanced
a sequential
advanced vehicle
tire
vehicle
control
Abstract: systems
Tire-roadand autonomous
friction vehicle(TRFC)
coefficient applications.
is of for This paper
critical importancepresents a sequential tire
control
longitudinal
control
longitudinal
systems
systems and
and autonomous
stiffness
stiffness and TRFC
autonomous
and TRFC
vehicle
vehicle applications.
estimation
estimation method
applications.
method for
This
roadpaper
This
road vehicles
paper
vehicles byfor
presents
presents
by
advanced
aa sequential
separate
separatesequential vehicle
control
control
tire
of
tire
of
control
longitudinal
the front
longitudinalsystems
and and
stiffness
rear
stiffness autonomous
and TRFC
wheels,
and TRFC
for vehicle
estimation
which only
estimation applications.
method for
standard
method for This
Kalmanroad paper
road vehicles
vehicles
filter presents
(KF)by and
by a
separate
separatesequential
control
recursive
control tire
of
least
of
the front andstiffness
longitudinal rear wheels,
and are for which
TRFC only standard
estimation method Kalman filter (KF) and recursive least
the
the front(RLS)
squares
front
squares
and techniques
and
(RLS)
rear wheels,
rear wheels,
techniques are
for
for which
required.
which
required.
only
only
While braking for
standard
Whilestandard
braking
Kalman
torque
Kalman
torque
roadis vehicles
is
filter
actively
filter
actively
(KF)
(KF)by
applied
applied
separate
and
and to
control
recursive
to rear
recursive
rear wheels, of
least
least
wheels,
the front
squares and
(RLS) rear wheels,
techniques
the proportional-integral-derivative
proportional-integral-derivative
squares are
(RLS) techniques are required. for which
required.(PID) only
While
While standard
braking
speed Kalman
torque
controller
braking is filter
actively
dictates
torquedictates
is actively (KF)applied
the applied and
corresponding recursive
to rear least
wheels,
driving
to rear wheels,
the
squares (PID) speed controller the corresponding driving
the
the to(RLS)
front techniques
proportional-integral-derivative
torque wheels to
proportional-integral-derivative
torque to front wheels
are required.
to compensate
compensate theWhile
(PID)
(PID)
the decreasebraking
speed
speed
decrease thetorque
controller
of the
controller
of vehicle is longitudinal
actively
dictates
dictates
vehicle the applied
the
longitudinal
corresponding
speed.
corresponding
speed.
to Tire
rear force
Tire
wheels,
driving
driving
force of
of
the
rear proportional-integral-derivative
torque
torque to are
tires
to frontestimated
front wheels to
wheels tobased
compensate
on
compensate the (PID)
the decrease
KF,
the speed
decrease
both controller
of the
the vehicle
longitudinal
of dictates
vehicle
stiffness the TRFC
longitudinal
and
longitudinal corresponding
speed.
speed. of rear
Tire driving
Tiretires
force
force of
are
of
rear
torquetires
to are
front estimated
wheels tobased on the KF,
compensate the both longitudinal
decrease of the has stiffness
vehicle and TRFC
longitudinal speed. of rear tires
forceare
Tirepotential of
rear
rear tires
identified
tires are
using
are estimated
estimated based
RLS technique.
technique.
based on The
on the KF,
the KF,
proposedboth longitudinal
both longitudinal
method stiffness
simple
stiffness and TRFC
TRFC
structure
and and of the
of rear
rear tires
tires are
are
identified
rear tires using
are RLS
estimated based on The
the proposed
KF, both method
longitudinal has simple
stiffness structure
and TRFC and of the
rear potential
tires are
identified
of estimating
estimating
identified using
using RLS
theRLS technique.
parameters
technique. The
without proposed method
severely affecting
The proposed affecting
method the has
the simple
vehicle’s
has vehicle’s structure and
desired longitudinal
simple structure the
longitudinal potential
and the potential speed.
of
identified using the parameters without severely desired speed.
of
of estimating
Simulation
estimating
Simulation theRLS
the
results
results
technique.
parameters
from
parameters
from vehicle
vehicle
The proposed
without
dynamics
without
dynamics
severely
severely method
affecting
software
software CarSim
affecting
CarSim
has vehicle’s
the
the simple structure
desired
and Matlab/Simulink
Matlab/Simulink
vehicle’s
and
and the
desired longitudinal
longitudinal potential
validate
validate
speed.
the
speed.
the
of estimating
Simulation
effectiveness
Simulation the
results
of
resultstheparameters
from
proposed
from vehicle
vehiclewithout
dynamics
method.
dynamicsseverely affecting
software
software CarSim
CarSimthe vehicle’s
and desired longitudinal
and Matlab/Simulink
Matlab/Simulink validate
validate speed.
the
the
effectiveness
Simulation of the fromproposed method.
effectivenessresults
effectiveness of the
of the proposed
proposedvehicle dynamics software CarSim and Matlab/Simulink validate the
method.
method.
© 2019, IFAC of
effectiveness (International
the proposed Federation
method. of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: Tire-road
Keywords: Tire-road frictionfriction coefficient,
coefficient, Kalman
Kalman filters,
filters, recursive
recursive least
least squares,
squares, PID PID control.
control.
Keywords: Tire-road
Keywords: Tire-road frictionfriction coefficient,
coefficient, Kalman
Kalman filters,
filters, recursive
recursive least
least squares,
squares, PID PID control.
control.
Keywords: Tire-road friction coefficient, Kalman filters, recursive least squares, PID control.
1. INTRODUCTION
1. INTRODUCTION perature. For
perature. For example,
example, road road conditions
conditions are are distinguished
distinguished
1. INTRODUCTION
1. INTRODUCTION perature.
based on
perature. Forroad
on the
the
For example, road level
luminance
example, road conditions
utilizing
conditions areimage
are distinguished
analysis
distinguished
1. INTRODUCTION based
perature. For road luminance
example, road level utilizing
conditions image
areimage analysis
distinguished
based
technology
based on the
on the byroad
road luminance
Yamada
luminanceet al. level utilizing
(2001).
level image
Although
utilizing analysis
different
analysis
Tires are
are the
the only
only contact
contact of of the
the vehicle
vehicle withwith road
road surfaces.
surfaces. based technology byroad
Yamada et al.level (2001). Although different
Tires
Tires
The are the
maximumthe only
only contact
forces that of tires
the vehicle
vehicle
can with road
generate road
are surfaces.
mostly rough on
technology
technology
rough
theby
road
road
by luminance
Yamada
conditions,
Yamada
conditions,
et
likeal.
et
like
al.‘Dry’,
‘Dry’,
utilizing
(2001).
(2001). ‘Wet’,
image different
Although
‘Wet’, ‘Slushy’,
Although
‘Slushy’,
analysis
different
‘Icy’,
‘Icy’,
Tires
The are
maximum contact
forces that of the
tires can with
generate are surfaces.
mostly technology
rough road by Yamada
conditions, et
like al. (2001).
‘Dry’, Although
‘Wet’, ‘Slushy’,different
‘Icy’,
Tires
The are
maximumthe only contact
forces that of the
tires vehicle
can with
generate road
are surfaces.
mostly and
rough ‘Snowy’,
road
and ‘Snowy’, can be
conditions, discriminated
like
can be discriminated ‘Dry’, even
‘Wet’, without
‘Slushy’,physical
‘Icy’,
determined
The maximum
determined
The maximum
by the
by the
forces
forces
tire-road
that tires
tire-road
that
friction
friction coefficientare
can generate
coefficient (TRFC).
mostly rough
(TRFC). and road these
‘Snowy’,
excitations, conditions,
can methods
be discriminated
discriminated ‘Dry’, even
likeinvariably ‘Wet’,
even
without
‘Slushy’,
without
neglect
physical
‘Icy’,
physical
the influence
influence
determined
Accurate
determined
Accurate
by the
by the
information
information of thetires
tire-road
tire-road
of the TRFC cancan
friction
friction
TRFC
generate
coefficient
can
are
facilitate
coefficient
facilitate
mostly and
(TRFC).
designs
(TRFC).
designs and
‘Snowy’,
excitations,
‘Snowy’,
can
these
can
be
methods
be invariablyeven
discriminated even
without
neglect the
without
physical
physical
determined by the tire-road friction coefficient excitations,
of many
many factors theseincluding
factors methods invariably types,neglect
tire types, tire wear
wearthe influence
condition
Accurate
of advanced
advanced
Accurate
of
information
vehicle
information
vehicle
of the
of the TRFC
control
control
TRFC
and
and
can
safety
can
safety systems(TRFC).
facilitate
facilitate
systems
designs
such
designs
such as excitations,
as of
excitations,
these methods
including
theseincluding
methods
invariably
tire
invariably
neglect
tire
neglect
the condition
influence
the condition
influence
Accurate
of advanced information
vehicle of the
control TRFC can
and traction facilitate
safety systems
systems designs
such as of of
and
of many
tire
many factors
inflation
factors including
pressure tire
on
tire types,
the
types, tire
friction
tire wear
wear condition
estimation. In
adaptive
of advanced
adaptive cruise
cruise control
vehicle
control (ACC),
control
(ACC), and safety
traction control such
control system
system as andmanytire inflation
factors pressure tire
including on the types,friction
tire estimation.
wear condition In
of advanced
adaptive vehicle
cruise controlcontrol
(ACC), and safety
traction systems
control such
system as and
and tire
addition,
tire inflation
vehicle
inflation pressure
control
pressure and on
on the
safety
the friction
systems
friction estimation.
cannot benefit
estimation. In
In
(TCS), autonomous
adaptive autonomous
cruise control emergency braking
(ACC), braking system
tractionsystem (AEBS),
control(AEBS),
system and addition, vehicle control andon safety systems cannot benefit
(TCS),
adaptive
(TCS), cruise
autonomous
and collision
collision control
warning
emergency
(ACC), avoidance
emergency
or collision
collision tractionsystem
braking control
system
systems. system fromtire
addition,
These addition,
(AEBS), from the inflation
the
vehicle
vehicle
output
pressure
control
output control
without
without
and
and the systems
safety
aa numerical
numerical
safety friction
valueestimation.
systems
value
cannot
showing
cannot
showing
benefit
benefitIn
the
the
(TCS),
and autonomous
warning emergency
or braking
avoidance systems.(AEBS),
These addition,
from the vehicle
output control
without and safety systems cannot benefit
(TCS),
and autonomous
collision
systems
and warning
maywarning
collision only be beemergency
orconservatively
or braking
collision avoidance
collision avoidance system
designed in(AEBS),
systems.
systems. ab- from
These
the
These
TRFCthe
TRFC related
output
related to tire
to tire
without
and aa numerical
and vehicle numerical
vehicle value showing
maneuver.
value
maneuver.
showing the the
systems
and may
collision only
warning orconservatively
collision avoidancedesigned in
systems. the ab-
These from
TRFC
TRFC the output
related
related to
to without
tire
tire and
and a numerical
vehicle
vehicle value showing the
maneuver.
maneuver.
systems
sence of
systems may
of may
TRFC, onlywhich
only be conservatively
be conservatively
would lead designed
lead designed
to reduced
reduced in the
in the ab- TRFC
perfor-
ab- Alternatively, some
related some researchers
to tireresearchers
and vehicleutilize utilize steering and
maneuver. and brak-
brak-
sence
systems TRFC, which would to perfor- Alternatively, steering
sence
mances,
sence of may
of TRFC,
see
TRFC, onlywhich
Singh be
which
and conservatively
Taheri
would lead designed
would (2015),
lead to De
to reduced
reduced in the
Castro ab-
perfor-
et
perfor-al. Alternatively,
ing/acceleration
Alternatively, some
some researchers
responses
researchers of theutilize
vehicle
utilize steering
and/or
steering and
and brak-
tires
brak-to
mances,
sence see Singhwhichand Taheri (2015), De Castroperfor-et al. Alternatively,
ing/acceleration responses of the vehicle and/or tires to
(2012) ofand
mances,
mances,
(2012) and TRFC,
see
see Singhet and
Ahn
Singh
Ahn et
and
al.
would (2015),
Taheri
al. (2013).
(2013).
Taheri lead
(2015),
In
In
to De
De
addition,
addition,
reduced
Castro
autonomous
Castro
autonomous
et al.
et al. ing/acceleration
ing/acceleration
identify the some
friction researchers
responses
coefficient.
responses of the
of theutilize
These vehiclesteering
and/or
estimation
vehicle and/or and brak-
tires
strategies
tires to
to
mances,
(2012) andseealso
Singh
Ahn et and
al. Taheri vehicles,
(2013). (2015),
In addition, De
addition, Castro
autonomous identify the friction
et al. ing/acceleration coefficient.
responses of theThese estimation
vehicle strategies
and/orstrategies
tiresandto
vehiclesand
(2012)
vehicles or
or Ahn
also called
et
called intelligent
al. (2013).
intelligent In
vehicles, though
though currently
autonomous
currently identify
tend to
identify the
to the friction
varyfriction coefficient.
considerably
coefficient. These
depending
These estimation
on excitationsstrategies
excitations
estimation
(2012)
vehicles and
or Ahn
also et
called al. (2013).
intelligent In addition,
vehicles, though autonomous
currently tend
identify vary
the considerably
friction coefficient. depending
These on
estimation and
strategies
at an
vehiclesearly
at an earlyor stage
also
stage of
called development,
intelligent
of development, have
vehicles, attracted
though
have attracted a lot
currently of
a lot of tend tend
road
tend to vary
conditions,
to vary considerably
and
considerably they depending
can
dependingbe on
further
on excitations
classified
excitations and
into
and
vehicles or also called intelligent vehicles, though currently road to conditions, and they depending
vary considerably can be further classified into
at
at an early
research
an
research
early stage
attention
stage
attention
offrom
of
from
development,
both academia
development,
both academia have and
have attracted
industry
attracted
and industry aa lot
lot of road
(Gill
of
(Gill road conditions,
three conditions,
three categories.
categories.
and first
The
and
The
they category
they
first
can be
can
category is on
be further
further
is
excitations
based
based
classified
on the
on the com-
classified and
into
com-
into
at
et an
research
al.
researchearly
(2015)).stage
attention
attention of
While
fromdevelopment,
fromthe both
both academia
human
academia have
driver attracted
andinindustry
and industry a
conventionallot of
(Gill road
(Gill three conditions,
categories. and
The they
first can
category be further
is based classified
on the into
com-
bined categories.
longitudinalThe andfirst
lateral dynamics, seeon Choi et al.
et al. (2015)).
research attention While
from theboth human
academia driver inindustry
andin conventional (Gill three
bined
three longitudinal and lateralcategory is based
dynamics, see the et
Choi com-al.
et al. (2015)).
vehicles
et al.
vehicles
(2015)). While
effectively
While
effectively
the
compensates
the
compensates
human
human for
driver
fordriver in conventional
variations
variations
conventional
in friction,
in friction, (2013),categories.
bined
bined
(2013),
longitudinal
Qi et
Qi al. The
et al.
longitudinal andfirst
(2015).
and
(2015).
lateralcategory
Strongly
lateral
Strongly
is based
dynamics,
coupled
dynamics,
coupled
see
see on theofet
Choi
effects
Choi
effects of
com-
et al.
the
al.
the
et
theal.knowledge
(2015)). While the ishuman driver in conventional au- bined
vehicles
the
effectively
knowledge
vehicles effectively
of
compensates
of TRFC
TRFC
compensates is vital
forand
vitalfor variations
necessary
variations
and necessary
in friction,
in friction,
for au-
for lateral longitudinal
(2013),
(2013),
lateral
Qi longitudinal
and
Qi
and
et al.
et
longitudinal
and friction
al. (2015).
(2015). lateral
Strongly
Strongly
friction
dynamics,
coupledsee
coefficients
coupled
coefficients may
may
Choi
effects ofet
poseof
effects
pose
al.
the
com-
the
com-
vehicles
the
tonomous
the effectively
knowledge
knowledge drivingof to
of to compensates
TRFC
TRFC adapt is vital
the for variations
and
control necessary in
strategies friction,
for
to au-
the (2013),
lateral Qi et al.
and longitudinal(2015).
longitudinal Strongly
friction coupled
coefficients effects
maythe of
poseTRFC the
com-
tonomous
the knowledge drivingof toTRFC adaptis vitalcontrol
the and necessary
strategies for to au- plex challenges.
the lateral
plex challenges.
and It has
It has friction
also been
also been shown that
coefficients
shown that
may pose
the TRFCcom-
tonomous
maximum
tonomous driving
grip
driving to
levels adaptis the
available
adapt vital
the androads
control
oncontrol
the necessary
strategies
(Talvala
strategies for
toetau-
to al. lateral
the
the plex
can
plex and longitudinal
challenges.
be estimated
challenges. Iton
It hasthe
has friction
also
also basis been
been coefficients
of shown maythe
that
longitudinal
shown that poseTRFC
the TRFCcom-
dynamics.
maximum
tonomous grip levels
driving available
to available
adapt on the roads (Talvala et al. can be estimatedIton thealso basis of shown
longitudinal dynamics.
maximum
maximum
(2011)).
grip
(2011)). Considerable
grip levels
Considerable
levels efforts the
available
efforts
oncontrol
have
on
have
the
the
been
strategies
roads
beenroads
made
made the to
(Talvala
in the
(Talvala
in
etthe
recent
et
recent
al. plex
al. can challenges.
be
Slip-slope
can be estimated
based method
estimated onhasthe
method
on theisbasis
basis
one been ofthe
ofof thatapproaches,
longitudinal
common
longitudinal the TRFC
dynamics.
dynamics.
maximum
(2011)). grip levels
Considerable available
efforts on
have the
been roads
made (Talvala
in the et
recental. Slip-slope
can be based
estimated on the isbasis
one of ofthe common
longitudinal approaches,
dynamics.
decades to
(2011)).
decades to obtain reliable
Considerable
obtain reliable andhave
efforts
and on-line
on-line beenestimations
made in the
estimations of TRFC
of TRFC Slip-slope
recent Slip-slope
which assumes basedlinear
assumes
based method
method is one
one of
relationship
is of the
the common
between
common approaches,
the longitudinal
longitudinal
approaches,
(2011)). Considerable efforts which linear relationship between the
decades
considering
decades
considering
to
to obtain
obtain reliable
different
reliable
different andhave
and
operating
operating
on-line
on-line been madeand
estimations
conditions
estimations
conditions
in the
and
of
of recent Slip-slope
TRFC
dynamic
TRFC
dynamic which
force
which
force andand based
assumes
the
assumes method
linear
slip within
linear
the slip
isaone
relationship
relationship
within very ofsmall
a very small
the common
between
between slip the approaches,
longitudinal
deformation
the
slipthe longitudinal
deformation of
of
decades
considering
responses
consideringtoofobtain reliable
different
the vehicle.
vehicle.
different and on-line
operating
Some
operating estimations
conditions
researchers
conditions and on
focus
and of TRFC which
dynamic
TRFC
dynamic force assumes
and linear
the slip
slip relationship
within a very
very between
small slip longitudinal
deformation of
responses of the Some researchers focus on TRFC the
force tire
the tire and(Gustafsson
the
(Gustafsson (1997),
within
(1997), a Rajamani
Rajamanismall et
slipal. (2012)).
deformation
et al.deformation
(2012)). LastLast
of
considering
responses
estimationof
responses different
ofconsidering
the vehicle.
the vehicle.operating
Some
the
Some primary conditions
researchers
causes focus
researchers and on
focus
leading ondynamic
TRFC
to
TRFCvari- force
the and
tire the one
slip is
(Gustafsson within
(1997), a veryRajamanismall slipet al.
al. (2012)).
(2012)). of
Last
estimation considering the primary causes leading to vari- but tire
the
but not
not least
(Gustafsson
least one is the lateral
(1997),
the lateral dynamics-based
Rajamani et
dynamics-based methods.
Last
methods.
responses
estimation
ations in of
in TRFC,
estimation TRFC,the vehicle.
considering
such as
considering Some
the
astheroad researchers
primary causes
irregularities,
primary focus
leading on
weather,
causes leading TRFC
to vari-
tem-
to tem-
vari- but the tire (Gustafsson
but not leastleast one
one is (1997),
is the
the lateral Rajamani
lateral dynamics-based et al. (2012)).
dynamics-based methods. Last
methods.
ations
estimation
ations in
ations in TRFC,
TRFC,
such
considering
such as
such asthe
road
road
road
irregularities,
primary
irregularities,
irregularities,
weather,
causes leading
weather,
weather, vari- but not
to tem-
tem- not least one is the lateral dynamics-based methods.
ations in ©
2405-8963 TRFC, such(International
2019, IFAC as road irregularities,
Federation of weather,
Automatic tem- Control) Hosting by Elsevier Ltd. All rights reserved.
Copyright © 2019 IFAC
Copyright
Peer review© under
2019 IFAC
responsibility of International Federation of Automatic Control.
Copyright
Copyright ©
© 2019
2019 IFAC
IFAC
10.1016/j.ifacol.2019.08.061
Copyright © 2019 IFAC
2019 IFAC IAV

Gdansk, Poland, July 3-5, 2019 Juqi Hu et al. / IFAC PapersOnLine 52-8 (2019) 136–141 137

Fig. 2. Wheel dynamic model

Fig. 1. Schematic diagram of the vehicle dynamics
load Fz , longitudinal stiffness Cx , road friction µ and slip
Yoon et al. (2016) simultaneously estimated both the ve- ratio κ, as:
hicle sideslip angle and the TRFC from a sensor fusion
framework based on the Kalman filter (KF). Ahn et al.
(2013) proposed an adaptive state and parameter observer Cx (κ/(1 + κ))
Fx = F
to estimate the TRFC and slip angle by studying the f

lateral acceleration and tire-aligning moment responses. f = Cx2 (κ/(1 + κ))2
Most of these, however, impose relatively large magni-  (2)
tudes of braking/accelerating or steer inputs to achieve  f2 f3
f− + , if f  3µFz
sufficient variations in dynamic responses for identifying F = 3µFz 27µ2 Fz2

the TRFC. These maneuvers may not be practical during µFz , if f > 3µFz
normal driving and may interfere with the desired vehicle Both the normal force and slip ratio can be computed
motion in real driving situations. The effective estimation based on the measured acceleration ax , wheel rotation
of TRFC generally involves a trade-off between sufficient angular speed ω and vehicle forward speed Vx , according
excitation and small disturbance on vehicle motion. It is to:
desirable to develop TRFC estimation methods involv-
ing adequate excitations to warrant convergence of the mglr ∓ max hc
parameter estimation algorithm while ensuring minimal Fzf,r = (3)
interference with the vehicle motion. Motivated by this, l
and
we integrate speed controller with parameter estimation
considering longitudinal dynamic responses in this paper.  Re · ω
By separate control of front and rear wheels, the longitu- 
1 − , if Vx  Re · ω, Vx = 0
Vx
dinal stiffness and the TRFC are identified in sequence by κ= (4)
actively employing different magnitudes of braking torque 1 − Vx , if Vx < Re · ω, ω = 0

on rear wheels without severely influencing vehicle forward Re · ω
speed. where subscript f, r refers to front- and rear-axle, respec-
tively. Re is the tire effective radius, g is the acceleration
2. SYSTEM MODELING due to the gravity, l = lf + lr is the wheel base of the
vehicle.
2.1 Vehicle Model
2.3 Wheel Rotation Dynamics
A bicycle type vehicle model, ignoring the difference be-
tween left and right tires, is considered and shown in Fig. 1. The moment equilibrium equation for an individual wheel
Neglecting road gradient and wind speed, the longitudinal as shown in Fig. 2 is:
dynamics can be expressed as: Iω ω̇ = Td − Tb − Re Fx − Re Fr (5)
mV̇x = Fxf + Fxr − FR (1) where Iω is the moment of inertia of the wheel, Td and
Tb are the driving torque and braking torque, respectively,
where m is the total vehicle mass, Vx is the vehicle longitu-
and Re Fr = Re fr Fz is the rolling resistance moment with
dinal speed, FR represents the total resistance attributed
fr being the rolling resistance coefficient.
to rolling resistance of tires and the aerodynamic drag
force, and Fxf and Fxr are the longitudinal forces devel-
oped at the front- and rear-axle tires, respectively. In Fig. 3. PROPOSED ESTIMATION METHOD
1, hc is the height of the center of gravity (cg), lf and lr
are the distances between the cg and the front- and rear- In this section, the longitudinal stiffness and TRFC es-
axle, Fzf and Fzr are the normal forces of the front- and timation method is proposed. The block diagram of the
rear-axle tires, respectively. proposed method is shown in Fig. 3. Adequate excitations
are generated by actively applying braking torque to the
2.2 Tire Brush Model rear tires, and the required driving torque to the front
tires to compensate for the effects on the vehicle motion is
calculated from the vehicle speed PID controller.
The tire model used in this paper is the tire brush model
as described in Singh and Taheri (2015), Choi et al. (2013). By exerting different types of the braking torque on rear
Assuming the tire slip angle is small, the model describes tires (small constant torque Tb•,s in the first stage, braking
longitudinal force Fx as a nonlinear function of normal pulse torque with large magnitude Tb•,l in the second

137
2019 IFAC IAV
138
Gdansk, Poland, July 3-5, 2019 Juqi Hu et al. / IFAC PapersOnLine 52-8 (2019) 136–141

Fig. 3. Block diagram of the proposed estimation framework

stage), the longitudinal stiffness and the TRFC of rear 3.3 Parameter Identification
tires (Cxr• and µr• ) are estimated in sequence based on
the corresponding estimated tire force (F̂xr•,s , F̂xr•,l ) and Both longitudinal stiffness Cxr• and TRFC µr• are esti-
computed slip ratio (κr•,s , κr•,l ), where the • sign in the mated from rear tires in this paper. Braking forces of the
subscript denotes that the variables are the same for left rear tires are estimated from the KF, and in the meanwhile
and rear wheels. As only straight-line driving scenario with normal forces and slip ratios of the rear tires are computed
constant forward speed is considered in this paper, the according to (3)-(4). The recursive least squares (RLS)
differences between left and right tires can be neglected. parameter identification technique is utilized to identify
the longitudinal stiffness and the TRFC.
3.1 PID Speed Controller Longitudinal Stiffness Estimation In the Stage I, a small
constant braking torque Tb•,s is applied to the rear tires.
Vehicle longitudinal velocity is maintained using a PID The required driving torque of front tires Td•,s is sub-
controller, whose overall control function can be expressed sequently calculated from the PID speed controller and
mathematically as: applied to the front wheels to maintain the desired longitu-
 t dinal speed Vx,des . Based on wheel rotation dynamic equa-
de(t)
Td•,j (t) = Kp e(t) + Ki e(τ )dτ + Kd , j = s, l (6) tion (5), the longitudinal braking force Fxr•,s is estimated
0 dt using the KF technique. Longitudinal slip ratio of rear
where Td•,j (t) is the driving torque of front tires, e(t) = tires κr•,s is computed from available measured signals,
Vx,des (t) − Vx (t), with Vx,des (t) and Vx (t) being the de- namely forward speed Vx and angular speed of the rear
sired and actual vehicle forward velocity, respectively, Kp , wheels ωr• . When the slip ratio is small, it is reasonable to
Ki and Kd denote the coefficients for the proportional, assume linear relationship between the longitudinal force
integral, and derivative terms, respectively. and the slip ratio as Fxr• = Cxr• κr•,s . The RLS technique
is applied for the longitudinal stiffness Cxr• identification
subsequently. For more details about the RLS, see Simon
3.2 Tire Force Estimation (2006).
Rear tires are the excitation wheels and the braking force TRFC Estimation The dependence of tire longitudinal
of rear tires are estimated using Kalman filter (KF). force on the TRFC is not clearly identifiable when the slip
Define the state vector as x(t) = [Fxr• , ωr• ]T and the ratio is very small. For the purpose of TRFC estimation, a
measurement as z(t) = ωr• , where Fxr• and ωr• are relatively large braking input is required to induce notable
braking force and rotation angular speed of individual rear changes in longitudinal dynamics and consequently for the
tire, respectively. Based on (5), the following state-space convergence of the parameter identification algorithm. The
system with process noise w(t) and measurement noise v(t) application of a constant large braking torque to the rear
is built: tires makes it difficult to develop sufficient driving torque
   0 0   0 Td•,l of the front wheels to overcome the effect of braking
Ḟxr• F xr•
= Re + 1 Ttot + w(t) of the rear tires. Alternatively, a braking pulse torque of
ω̇r• − 0 ωr•
Iω Iω (7) sufficient magnitude Tb•,l is exerted on the rear tires in
 the Stage II for estimating TRFC in a reliable manner.
Fxr•
z(t) = [0 1] + v(t) Braking force and slip ratio of rear tires, Fxr•,l and κr•,l ,
ωr•
are estimated or computed in a manner similar to that in
where Ttot = −Tb•,j − Re fr Fzr• , j = s, l is the total torque the stiffness estimation in the Stage I. In order to utilize
applied to the individual rear tire (driving torque of rear the RLS to estimate the TRFC, the nonlinear form of tire
tires is zero in this paper). Equation (7) is discretized using brush model (2) should be firstly linearized. Equations in
zero-order hold for being applicable to the discrete-time (2) could be written as follows:
KF. The algorithm of the discrete-time KF is referred to
y(k) = f (k, θ) + v (8)
Simon (2006). Therefore, braking forces of the rear tires
are estimated based on the active braking torque and the where k is the index for time step, y(k) = [F̂xr•,k ] is the
measured rotation angular velocity. observed tire force from the KF, f (k, θ) is the expression

138
2019 IFAC IAV

Gdansk, Poland, July 3-5, 2019 Juqi Hu et al. / IFAC PapersOnLine 52-8 (2019) 136–141 139

Torque (N-m) V x =100 km/h Fxr.,s (N)

100
V x=100 km/h 100.5 0
-50
80 100 -100 V =100 km/h
V x=54 km/h x
99.5 -150
60 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Desired True
40 V x=100 & 54 km/h V x =54 km/h Estimated
Actual
54.5 0
20 T b.,s -50
54 -100 V =54 km/h
T d.,s x
0 53.5 -150
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time (s) Time (s) Time (s)
(a) (b) (c)

Fzr.,s (N) 104

10-3 r.,s 6
2900 4

Stiffness [N/unit slip]

V x=100 km/h 2 V x=100 km/h V x=54 km/h
2800
0 5
2700 -2
2600 -4
2 4 6 8 10 12 14 2 4 6 8 10 12 14 4 V x=100 km/h
Actual Actual
Calculated 10-3 Calculated 3
2900 4
2800 2 V x=54 km/h True
0 2 Estimated
2700 V x=54 km/h -2
2600 -4
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time (s) Time (s) Time (s)
(d) (e) (f)

Fig. 4. Longitudinal stiffness estimation at two different speeds (µ = 0.6)

Fxr.,l (N)
600 T b.,l (100 & 54 km/h) V x =100 km/h
1000
T d.,l (100 km/h)
500 T d.,l (54 km/h) 100 0

-1000 V x=100 km/h

400
Torque [N-m]

98
5 10 15 5 10 15
300 True
Desired
V x =54 km/h Actual Estimated
200 55 1000
54 0
100
53 -1000 V x=54 km/h

0 52
5 10 15 5 10 15 5 10 15
Time (s) Time (s) Time (s)
(a) (b) (c)

Fzr.,s (N) r.,s 0.8

3400 0.1 V x=100 km/h
3200 V x=100 km/h V x=100 km/h
3000 V x=54 km/h
0
2800 0.6
2600
-0.1
2400
TRFC

5 10 15 5 10 15
Actual Actual
0.4
Calculated Calculated
3400 0.2
3200 V x=54 km/h V x=54 km/h
3000 0.2
2800 0
2600
2400
-0.2 0
5 10 15 5 10 15 5 10 15
Time (s) Time (s) Time (s)
(d) (e) (f)

Fig. 5. TRFC estimation at two different speeds (µ = 0.6)

of tire brush model, θ = [µr• ] is the TRFC, and v is the y(k) ≈ F (k)(θ̂k − θ̂k−1 ) + f (k, θ̂k−1 ) (9)
observed noise. y(k) can be approximated as: where

139
2019 IFAC IAV
140
Gdansk, Poland, July 3-5, 2019 Juqi Hu et al. / IFAC PapersOnLine 52-8 (2019) 136–141

Torque (N-m) V x =100 km/h Fxr.,s (N)

100
V x=100 km/h 100.5 0
-50
80 100 -100 V =100 km/h
V x=54 km/h x
99.5 -150
60 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Desired True
40 V x=100 & 54 km/h V x =54 km/h Estimated
Actual
54.5 0
20 T b.,s -50
54 -100 V =54 km/h
T d.,s
x
0 53.5 -150
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time (s) Time (s) Time (s)
(a) (b) (c)

Fzr.,s (N) 104

10-3 r.,s 6
2900 4
V x=100 km/h

Stiffness [N/unit slip]

2 V x=100 km/h V x=54 km/h
2800
0 5
2700 -2
2600 -4
2 4 6 8 10 12 14 2 4 6 8 10 12 14 4 V x=100 km/h
Actual Actual
Calculated 10-3 Calculated 3
2900 4
2 V x=54 km/h True
2800
0 2 Estimated
2700 V x=54 km/h -2
2600 -4
2 4 6 8 10 12 14 2 4 6 8 10 12 14 2 4 6 8 10 12 14
Time (s) Time (s) Time (s)
(d) (e) (f)

Fig. 6. Longitudinal stiffness estimation at two different speeds (µ = 0.4)

∂f (k, θ) estimation, normal force and slip ratio calculation at Stage

F (k) = |θ=θ̂k−1 (10) I and II. It is reasonable to recognize that the accuracy
∂θ
Define hk as follows: of tire braking force estimation deteriorates when a large
braking torque is applied, as shown in Fig. 5(c). Overall,
h(k) = y(k) + F (k)θ̂k−1 − f (k, θ̂k−1 ) (11) acceptable estimation results of the longitudinal stiffness
Substituting (9) into (11), the following linearized form is and TRFC from the proposed algorithm under two forward
obtained: speeds are obtained, as presented in Figs. 4(f) and 5(f).
h(k) ≈ F (k)θ̂k (12) Although high rotating speed of tire reduces longitudinal
At each time step, F (k) is built with the newly estimated stiffness, it has little influence on the TRFC estimation.
values θ̂k−1 , Fzr•,k , and κr•,k . Based on this linear obser-
vation model (12), the RLS method is applicable for the 4.2 Verification on Road Surface with µ = 0.4
TRFC estimation.
In this simulation, the same constant braking torque is
4. SIMULATION RESULTS applied to rear tires for longitudinal stiffness estimation.
As can be seen from Fig. 6(c), the braking forces under
The proposed two-stage TRFC algorithm is implemented two speeds are very close to that on road (µ = 0.6).
in the CarSim and Matlab/Simulink environment consid- As a result, the required driving torques on front tires
ering straight driving maneuvers (braking/acceleration) on are as well almost the same with their counterparts in
two different road surfaces (µ = 0.6 and 0.4). The effect of the first simulation since the vehicle is keeping the same
forward speed on the TRFC estimation is also investigated constant speed. Again, the actual forward speed under
with two different speeds (Vx,des = 100 and 54 km/h). TRFC estimation experiences larger oscillations due to
large magnitude of braking pulse exerted, as shown in
4.1 Verification on Road Surface with µ = 0.6 Figs. 6(b) and 7(b). Rear tire braking force estimation,
normal force and slip ratio calculation results at Stage I
In this simulation, active braking torque applied to rear and II are presented in Figs. 6(c)-6(e), and Figs. 7(c)-7(e),
tires and the corresponding driving torque employed to respectively. The overall results are very similar to the first
front tires for longitudinal stiffness estimation are shown simulation. As one can see from Fig. 6(f), the longitudinal
in Fig. 4(a). Larger driving torque is required to maintain stiffness under two different forward speeds maintains the
a higher desired speed (Vx,des = 100 km/h) due to larger same value with that on the first road surface (µ = 0.6),
aerodynamic drag force generated. For TRFC estimation, respectively. Compared to the first road condition, smaller
sufficiently large braking torque is necessary to induce braking torque is required and therefore smaller tire force
adequate excitations, as shown in Fig. 5(a). As a result, is generated for TRFC estimation, as shown in Figs. 7(a)
the actual forward speed under TRFC estimation has more and 7(c). The TRFC is finally accurately estimated as
obvious oscillations than that under stiffness estimation, as depicted in Fig. 7(f). It should be noted that the TRFC can
can be seen from Figs. 4(b) and 5(b). Figs. 4(c)-4(e), and be estimated with very small time-delay once the braking
Figs. 5(c)-5(e) illustrate good accuracy of tire braking force pulse torque is employed.

140
2019 IFAC IAV

Gdansk, Poland, July 3-5, 2019 Juqi Hu et al. / IFAC PapersOnLine 52-8 (2019) 136–141 141

Fxr.,l (N)
400 Tb.,l (100 & 54 km/h) V x =100 km/h
Td.,l (100 km/h)
100.5 500
Td.,l (54 km/h) 100 0
300 99.5
-500
99
Torque [N-m]

-1000 V x=100 km/h

98.5
5 10 15 5 10 15
200 True
Desired
V x =54 km/h Actual Estimated
1000
55
100
54 0

53 -1000 V x=54 km/h

0
5 10 15 5 10 15 5 10 15
Time (s) Time (s) Time (s)
(a) (b) (c)

Fzr.,s (N) r.,s 0.5

0.1 Vx =100 km/h
V x=100 km/h V x=100 km/h
3000 Vx =54 km/h

0 0.4

2500
-0.1 0.3

TRFC
5 10 15 5 10 15
Actual Actual
Calculated Calculated 0.2
V x=54 km/h 0.1 V x=54 km/h
3000
0 0.1

2500 -0.1
0
5 10 15 5 10 15 5 10 15
Time (s) Time (s) Time (s)
(d) (e) (f)

Fig. 7. TRFC estimation at two different speeds (µ = 0.4)

5. CONCLUSION Gustafsson, F. (1997). Slip-based tire-road friction esti-

mation. Automatica, 33(6), 1087–1099.
Estimation of longitudinal stiffness and the TRFC is con- Qi, Z., Taheri, S., Wang, B., and Yu, H. (2015). Estimation
sidered in this paper and it is divided into two sequential of the tyre-road maximum friction coefficient and slip
stages. Only conventional KF and RLS techniques are slope based on a novel tyre model. Vehicle System
required. By separate control of front and rear wheels, the Dynamics, 53(4), 506–525.
estimation method yields reliable estimates of both the Rajamani, R., Phanomchoeng, G., Piyabongkarn, D.,
stiffness and the TRFC without severely affecting vehicle’s and Lew, J.Y. (2012). Algorithms for real-time es-
desired motion. timation of individual wheel tire-road friction coeffi-
cients. IEEE/ASME Transactions on Mechatronics,
ACKNOWLEDGEMENTS 17(6), 1183–1195.
Simon, D. (2006). Optimal state estimation: Kalman, H
This work was supported in part by the Natural Sciences infinity, and nonlinear approaches. John Wiley & Sons.
and Engineering Research Council of Canada (NSERC) Singh, K.B. and Taheri, S. (2015). Estimation of tire-road
and a Concordia University VPRGS Team Grant. friction coefficient and its application in chassis control
systems. Systems Science & Control Engineering, 3(1),
REFERENCES 39–61.
Ahn, C., Peng, H., and Tseng, H.E. (2013). Robust esti- Talvala, K.L., Kritayakirana, K., and Gerdes, J.C. (2011).
mation of road frictional coefficient. IEEE Transactions Pushing the limits: From lanekeeping to autonomous
on Control Systems Technology, 21(1), 1–13. racing. Annual Reviews in Control, 35(1), 137–148.
Choi, M., Oh, J.J., and Choi, S.B. (2013). Linearized re- Yamada, M., Ueda, K., Horiba, I., and Sugie, N. (2001).
cursive least squares methods for real-time identification Discrimination of the road condition toward under-
of tire-road friction coefficient. IEEE Transactions on standing of vehicle driving environments. IEEE Trans-
Vehicular Technology, 62(7), 2906–2918. actions on Intelligent Transportation Systems, 2(1), 26–
De Castro, R., Araújo, R.E., and Freitas, D. (2012). 31.
Real-time estimation of tyre-road friction peak with Yoon, J.H., Li, S.E., and Ahn, C. (2016). Estimation of
optimal linear parameterisation. IET Control Theory vehicle sideslip angle and tire-road friction coefficient
& Applications, 6(14), 2257–2268. based on magnetometer with GPS. International Jour-
Gill, V., Kirk, B., Godsmark, P., and Flemming, B. (2015). nal of Automotive Technology, 17(3), 427–435.
Automated vehicles: The coming of the next disruptive
technology. In Report, Conf. Board of Canada, Ottawa.

141