Available online at www.sciencedirect.

com

ScienceDirect
IFAC PapersOnLine 53-2 (2020) 15332–15337
On
On Nonlinear
Nonlinear Control
Control for
for Lane
Lane Keeping
Keeping
On
On Nonlinear
Assist System
Nonlinear Control
in for Lane
Steer-By-Wire
Control for Lane Keeping
Road
Keeping
Assist
On
Assist System
Nonlinear
System in
in Steer-By-Wire
Control for Lane
Steer-By-Wire Road
Keeping
Road
Assist Wheeled Vehicles 

Assist System
System in
Wheeled
in
Wheeled
Steer-By-Wire
Vehicles
Steer-By-Wire
Vehicles  Road
Road
Wheeled Vehicles 

Wheeled Vehicles
Gabriele
Gabriele Perozzi, Perozzi, Chouki Chouki Sentouh,
Sentouh, Jérôme Jérôme Floris, Floris,
Gabriele Perozzi, Jean-Christophe
Chouki Sentouh,
Jean-Christophe Popieul
Popieul Jérôme Floris,
Gabriele
Gabriele Perozzi, Perozzi, Chouki
Jean-Christophe Sentouh,
Chouki Sentouh, Popieul Jérôme
Jérôme Floris, Floris,
Department of Jean-Christophe
Jean-Christophe
Automatic Control, Popieul
Popieul
LAMIH-UMR CNRS 8201,
Department of Automatic Control, LAMIH-UMR CNRS 8201,
Université
Department
Université Polytechnique
of Automatic
Polytechnique Hauts-De-France,
Control,
Hauts-De-France, LAMIH-UMR Valenciennes,
Valenciennes,CNRS France.
8201,
France.
Department
Université
Department of
of Automatic
Polytechnique
Automatic Control,
Control, LAMIH-UMR
Hauts-De-France, LAMIH-UMR Valenciennes,CNRS
CNRS France. 8201,
8201,
Université
Université Polytechnique
Polytechnique Hauts-De-France,
Hauts-De-France, Valenciennes,
Valenciennes, France.
France.
Abstract: This
Abstract: This paper
paper deals deals with
with thethe design
design of of aa lane
lane keeping
keeping system system in in steer-by-wire
steer-by-wire road road four-
four-
wheeled
Abstract:
wheeled vehicles
This paper
vehicles using
using a
a nonlinear
deals with thecontroller.
nonlinear design of aThe
controller. lanecontroller
The keeping system
controller is
is designed
designed to
in steer-by-wire
to take
take into
into account
road four-
account
Abstract:
the driver’s
wheeled
Abstract:driver’s This
vehicles
This paper
behavior usingand
paper deals
and with
availability
a nonlinear
deals the design
and to
with thecontroller.
design of a lane
toofcontrol
control
aThe keeping
the wheels’
lanecontroller
keeping system
wheels’ angle
is designed
system in steer-by-wire
of tothetake
in steer-by-wire steer road
system.
into four-
account
road four- A
the
wheeled behavior
vehicles using a availability
nonlinear and
controller. The the
controller is angle
designed of the
to steer
take system.
into account A
high
the
wheeled
high order
driver’s sliding
behavior
vehicles
order sliding mode
using
mode anda control
nonlinear
control is
availability chosen
and
controller.
is chosen as
to feedback
control
The
as control the controller,
controller
feedbackthecontroller, wheels’
is having
angle
designed
having of time-varying
the
to steer
take
time-varying upper-
system.
into account
upper- A
the
bound
high
the driver’s
order
driver’s behavior
disturbance
sliding
behavior mode and
strictly availability
depended
andcontrol and
by
andthe
is chosen
availability to
asroad
to curvature
feedback
control wheels’
thecontroller,and
wheels’ angle
by the
having
angle of the
the steer
ofinfluence
time-varying
steerof system.
the wind,
upper-
system. A
A
bound
high disturbance
order sliding strictly
mode depended
control is by
chosen the
as road curvature
feedback and
controller, by the
having influence
time-varying of the wind,
upper-
which
bound
high
which allows
disturbance
order
allows an
sliding
an easier
strictlytuning
modetuning
easier process
depended
controlprocess by
is chosen of
of the
the parameters
road
as feedback
the curvature
parameters based
and
controller,
based on
by
on athe
having
a physical
influence meaning
time-varying
physical of
meaning the and
wind,
upper-
and aa
bound
consequent
which
bound disturbance
allows attenuation
disturbancean strictly
easier
strictlyof
tuningdepended
the chattering
process
depended by
byof the
the
the road
effect. The curvature
parameters
road controller
curvature and
based
and mustby
on
by a the
be
the influence
robust
physical
influence to of
meaning
of the
minimize
the wind,
and
wind, the a
consequent
which attenuation of the chattering effect. The controller must bephysical
robust to minimize the
lateral allows
consequent
which
lateral deviation
allows
deviation
an
an andeasier
and
attenuation
easier thetuning
the heading
ofheading process
errors,of
the chattering
tuning process
errors,
the
and itparameters
effect.
ofand
theit must be sensitive
The controller
parameters
must be
based
sensitive
basedmust on
on
to
aabe
to assist
assist the driver
robust
physical
the
meaning
driver
to during
minimize
meaningduring
and
andthethe
aa
consequent
shared deviation
lateral
consequent control attenuation
mode,and the
attenuation of the
i.e.ofthe
the
heading chattering
assist
theassist systemand
errors,
chattering effect.
must The
give
it must
effect. The thecontroller
the authority
be sensitive
controller must
toto
must be
toassist robust
therobust
be driver
the driverto
when
to minimize
the driver
during
minimize driverthe
the
shared
lateral control
deviation mode,and i.e. system must give authority the driver when the
is available.
shared
lateral
is control mode,
deviation
available. and the heading
i.e. the
the assisterrors,
heading systemand
errors, must
and it
it must
give the
must be
be sensitive
authorityto
sensitive totoassist the
the driver
the driver
assist whenduring
driver the driver
during the
the
shared
is control
available.
shared control mode,
mode, i.e. the assist
i.e. the assist system
system must
must give
give the
theunderauthority
authority to the driver
the driverlicense
toBY-NC-ND when the
when the driverdriver
Copyright
is available. © 2020 The Authors. This is an open access article the CC
(http://creativecommons.org/
is available.Sliding mode control,
Keywords:
Keywords: Sliding mode control, licenses/by-nc-nd/4.0)
Automotive control,
Automotive control, Autonomous
Autonomous vehicles, vehicles, Interactive
Interactive vehicle
vehicle
control,
Keywords:
control, Co-operative
Sliding
Co-operative mode control.
control,
control. Automotive control, Autonomous vehicles, Interactive vehicle
Keywords:
control,
Keywords: Sliding mode
Co-operative
Sliding mode control, Automotive
control.
control, Automotive control, control, Autonomous
Autonomous vehicles, vehicles, Interactive
Interactive vehicle
vehicle
control,
control, Co-operative
Co-operative control.
control.
1. INTRODUCTION
1. INTRODUCTION are
are adaptively
adaptively weighted weighted to to dynamically
dynamically change change the the con-
con-
1. INTRODUCTION trol
are
trol authority
adaptively
authority share
weighted
share between
between to driver
dynamically
driver and
and controller.
change
controller. theIn
In Dai
con-
Dai
Advanced driver 1. INTRODUCTION
1. INTRODUCTION
assistance systems (ADASs) are a hot are
et
trol
are adaptively
al. (2018),
authority
adaptively the weighted
weighted between to
driver-vehicle-road
share to dynamically
driver
dynamically model
and change
to
controller.
change follow the
theIn con-
large-
Dai
con-
Advanced driver assistance systems (ADASs) are a hot et trolal.authority
curvature
(2018), the
path
driver-vehicle-road
share
is between
built under driver
the
model
and to followIn
controller.
assumption that
large-
Dai
the
topic
Advancednowadays
driver in both
assistanceeconomicssystems and research,
(ADASs) and
are a they
hot et
trolal.authority
curvature (2018), the
path driver-vehicle-road
share
is between
built under driver
the model
and
assumption to follow
controller. thatlarge-
In Dai
the
topic
Advancednowadays
driver inassistance
both economics systems and(ADASs)
research,are andatheyhot et
nearal. (2018),
and far the
vision driver-vehicle-road
information of themodel
road to
for follow
guidance large-
is
will
topic become
Advanced
will nowadays
become even
driver
even inmore
both popular
assistance
more economics
popular in
systems
in the
and
the near
(ADASs)
near future
research,
future a hot near and far vision information of the road for guidancethe
because
and they
arebecause curvature
et al. (2018), path
the is built under
driver-vehicle-road the assumption
model to follow thatlarge-
is
topic
the nowadays
incorporation inmore
of both
ADASseconomics
in modern and research,
vehicles mayand they
reduce curvature
considered
near
curvatureand path
farby the
vision
path is
is built
human
information
built under
driver,
under the
of
theand
the assumption
the
road
assumption parameters
for that
guidance
that the
de-
is
the
will
topic
the become
nowadays even
incorporation inof both popular
ADASseconomics
in in
modern the
and near
vehiclesfuture
research, may because
and they
reduce considered by the human driver, and the parameters de-
will become
traffic evenand
accidents more popular
make life in the
easier thefornear
near
the future
human because
driver. near
scribing and
considered
near and far
the
farbyvision
driver’s
the
vision information
steering
human
information driver, of
of the
and
the road
characteristicsthe
road for
and guidance
parameters
for behaviors
guidance is
de-
is
the
will incorporation
become
traffic even
accidents and of
more ADASs
popular
make life in modern
in
easier vehicles
forvehiclesfuture
the human may reduce
because
driver. scribing thebydriver’s steering characteristics and behaviors
the incorporation
Different types ofof ADASs
ADAS in
can modern
be found in the may reduce
literature considered
are considered
scribing
considered the the
as
driver’s
by the human
uncertainties
steering
human driver, of and
the
characteristics
driver, and the
model.
the parameters
and In
parameters de-
Sentouh
behaviors
de-
traffic
the
Differentaccidents
incorporation
types of and
of make
ADASs life
in easier
modern for the
vehicles humanmay driver.
reduce are considered as uncertainties of the model. In Sentouh
traffic
such accidents andADAS can easier
make life
life be found for the in human
the the literature
driver. scribing
et
are al.
scribing (2018)
considered the
the driver’s
aa ascontrol
driver’s steering
supervisor
uncertainties
steering characteristics
allows
of the
characteristics aa and
model. Inbehaviors
smooth
and tran-
Sentouh
behaviors
such as
Different
traffic
Different
lane
lane keeping
types
asaccidents
keeping
types
of
and
of
ADAS
ADAS
assist,
make
assist, collision
can be found
easier
collision
can be
avoidance,
for
avoidance,
found in
theautomated
in human
the
literature
driver.
automated
literature et
are al. (2018)
considered control
as supervisor
uncertainties of allows
the model. smooth
In tran-
Sentouh
trajectory
such as
Different planning,
lane keeping
types of adaptive
ADASassist, can cruise
collision
be control,
avoidance,
found in etc.
the The
automatedwork
literature sition
et
are al.
sition between
(2018)
considered
between a two
control
as
two local
uncertainties
local optimal-based
supervisor of
optimal-based allows
the controllers
model.a smooth
controllers In with
tran-
Sentouh
with
trajectory
such as laneplanning,
lane keeping adaptive
assist, cruise
collision control,
avoidance, etc. The
automatedwork et
two al. (2018)
predefined aa two
control
objectives, supervisor
i.e. lane allows
keeping aa smooth
and tran-
conflict
on this
trajectory
such as paper is
planning,
keeping based on
adaptive
assist, the lane
cruise
collision keeping
control,
avoidance, assist,
etc. The where
automatedwork sition
et
two al. between
(2018)
predefined control local
objectives, optimal-based
supervisor
i.e. lane allows
keeping controllers
smooth
and with
tran-
conflict
on this
trajectory paper is
planning, based on
adaptive the lane
cruise keeping
control, assist,
etc. where
Thedriver
work sition sition between two local optimal-based controllers with
the
on vehicle
this
trajectory papermust is
planning, follow
based onthe
adaptive the lane,
lane
cruise but the
keeping
control, human
assist,
etc. The where
work management
two predefined
management between between
objectives,
two
between the
local
the controller
i.e. lane and
optimal-based
controller and the
keeping
the driver.
controllers
driver. Wang
and conflict with
Wang
the
on vehicle
this papermust is followon
based thethelane,lane but the human
keeping assist, driver
where two
et al. predefined
(2018) objectives,
propose a methodi.e. lane keeping
where different and conflict
drivers’
must
the
on supervise
vehicle
this papermust isthe road
follow
based and
onthethe he has
lane,
lane the
but authority
the
keeping human
assist, over the
driver
where management
two predefined between
objectives,the controller
i.e. lane and the
keeping driver.
and Wang
conflict
mustvehicle
supervise thefollow
road and he has the the authority over the et al. (2018) between
management propose the a method
controller where
and different drivers’
the
controller.
must
the must
supervise
vehicle must thefollow
road and the he
the lane,
hasbut
lane, the the
but human
authority
humanover driver
driver characteristics
et
the management al. (2018) between
characteristics are
propose
are considered
a method
the
considered and
controller
and and the
handled
where
handled the by
different
by
driver.
driver. Wang
aa polytope
drivers’
Wang
polytope
controller.
must supervise the road and he has the authority over the et
and al.
a (2018)
regional propose
pole a
placement method is where
applied, different
then, a drivers’
method to
must supervise the road and he has the authority over the et
controller. al. (2018)
characteristics
and a regional pole propose
are a
considered method
placement isand and where
handled
applied, different
then, by a drivers’
polytope
aamethod to
Many papers
Many papers can
controller. can be be found
found in in the
the literature
literature dealing
dealing with with characteristics
convert
and a the
regional
characteristics are
pole
are considered
multi-objective
placement
considered H is handled
robust
applied,
and control
handled then, by
by is
a a polytope
converted
method
polytopeto
controller. convert the multi-objective H∞ robust control ismethod
converted
lane
Many
lane keeping/shared
papers
keeping/sharedcan be lateral
found
lateral incontrol
the
control and
literature
and considering
dealing
considering the
with
the and
into a
convert
and aa regional
singlethe
regional pole
objective placement
multi-objective
pole control.
placement H
∞is applied,
Boehm
is robust
applied, et then,
al.
control
then, (2016)a
is
a develop
converted
method to
to
Many
human
lane
Many papers
action
keeping/shared
papers can
with
can beorlateral
be found
without
found incontrol
in the
the
the literature
monitoring
and
literature dealing
of
considering
dealing the with
hu-
the
with into a single
convert the objective control.
multi-objective H
∞ Boehm
robust etcontrol
al. (2016) is develop
converted
human
lane action
keeping/shared with or without
lateral the
control monitoring
and considering of
considering the hu-
the ainto a system
into
converta
system single model
the
model which
objective
multi-objective considers
control.
whichcontrol. H
considers
∞
∞ that
Boehm
robust
that et a
et driver
al.
control
a driver (2016) can
is tighten
develop
converted
can develop
tighten
man driver’s
human
lane driver’s status.
actionstatus.
keeping/shared with or For example,
without
lateral Benloucif
theBenloucif
control monitoring
and et al.
al. (2019),
of the hu-
the grip
into aaonsingle objective
aa steering wheel, contract Boehmarm al.
al. (2016)
muscles, and change
man
human
Benloucif
man
human action
et
driver’s
action with
al.
status.
with (2017)
For
or
For
or
example,
without
take
example,
without the
into
the monitoring
account
Benloucif
monitoring
et
ettheof
al.
of
(2019),
the hu-
driver’s
(2019),
the hu-
a system
grip
a systemonsingle objective
model
steering
model
which
wheel,
which
control.
considers
contract
considers
Boehmarm et
that
that
a driver
muscles,
a driver
(2016) anddevelop
can
can
tighten
change
tighten
Benloucif
man et
driver’s al.
status. (2017)For take
example, into account
Benloucif ettheal. driver’s
(2019), posture
grip
a system
posture on a to
to have
steering
model
have the
which
the authority
wheel, contract
considers
authority over
over an
arm
that
an automation
muscles,
a driver
automation and
can system.
change
tighten
system.
steering
Benloucif
man
steering torque
et status.
driver’s
torque in(2017)
al.in the
the trajectory planning
take intoplanning
Fortrajectory
example, accountlevel
Benloucif level
etthe in order grip
driver’s
al.in(2019),
order All
grip the
posture on
on aaprevious
tosteering
have
steering wheel,
papers
thewheel, contract
deals
authority
contract with
over armthe
an
arm muscles,
human
automation
muscles, and change
interaction
and system.
change
Benloucif
to adjust
steering
Benloucif et al.
the
torque
et al. in (2017)
system’sthe
(2017) take into
desired
trajectory
take into accountinlevel
trajectory
planning
account the
a
the waydriver’s
in that
order
driver’s All the previous
posture to have papers
the deals over
authority with an theautomation
human interaction system.
to adjust
steering the
torque system’s
indriver desired
the trajectory
trajectory trajectory
planning in a
level way that
in(2017),
order using
All the
posture linear or
previous
to have nonlinear
papers
the controllers.
deals
authority with
over the
an Among
human
automation of
of all
all the
interaction
system.
better
to adjust
steering
better suits
thethe
torque
suits thesystem’s
in the
driver intention.
desired
intention. In
In Guo
trajectoryGuo in
planning et aal.
level
et al.way
in that using
order
(2017), All the
controllers,
using
All
linear
thelinear
or nonlinear
previous
nonlinearpapers
papersSMCs
or nonlinear
previous
controllers.
deals with
are the
withbecoming
controllers.
deals
Among interaction
the human
Among
human very
the
of popular
all the
interaction
to
the adjust
shared the
controlsystem’s desired
problem, with trajectory
or without in a
theway that
driver’s controllers,
using linear nonlinear
or nonlinear SMCs are
controllers. becomingAmong very of popular
all
better
to
the adjust
better
suits
shared the the
suitscontrol
the
driver
system’s problem,
driver
intention.
desired with or
intention.
In
trajectoryGuo
without
In without
Guo et
et
in aal.
theway
al.
(2017),
that
driver’s
(2017), tools
using
tools in
controllers, the
the last
inlinear or years
nonlinear
last nonlinear
years because
SMCs
because of
aretheir
controllers.
of excellent
becoming
their Among
excellent very all the
robustness.
of popular
robustness. the
intervention,
the
bettershared
suits
intervention, is
control
the
is formulated
problem,
driver
formulated as
as a
with
intention. a constrained
or
In Guo
constrained optimization
et the
al. driver’s
(2017),
optimization controllers,
A type
tools in
controllers, of
the nonlinear
SMCs
last
nonlinear is
years SMCs
super-twisting,
because
SMCs are
of
are becoming
their which
excellent
becoming is very popular
preferred
very robustness.
popularby
the sharedwhich
problem
intervention,
the shared control problem,
is problem,
solved online
is formulated
control with
online
aswith or without
by without
a constrained
or a model
model the driver’s A
predictive
optimization
the driver’s type
tools in of SMCs
the last is super-twisting,
years because of their which
excellent is preferred
robustness. by
problem
intervention,which is is solved
formulated as a by a
constrained predictive
optimization researchers
A type
tools in
researchers of
the because
SMCs
last
because is
years of
of its
its unique
super-twisting,
because of
unique theirproperty
which
excellent
property is that
preferredcan
robustness.
that can be
by
be
controller.
problem
intervention, In
which
controller.which Nguyen is
is formulated
In Nguyen et
solved al. (2017),
online the
by
as a constrained
et al. online
(2017), road-vehicle
a model system
predictive
optimization
the aroad-vehicle A type
applied
system researchers
A type of of
to SMCs
systems
because
SMCs is super-twisting,
is having
of its relative
unique degree
super-twisting, which
which one,
property is preferred
thatas
is preferred in
can by
Rath
be
by
problem
is transformed
controller.
problem In
which Nguyen is
into
is solved
a polytopic
et
solved al. (2017),
online LPV by
the
by form model
which
road-vehicle
a model predictive
allows
system
predictive to applied
researchers to systems
because having
of its relative
unique degree
property one, thatas in
canRath
be
is transformed
controller. Inlarge into
Nguyen a polytopic
et al. range
(2017), LPV the form which
road-vehicle allows
system to et
et al.
applied
researchers (2019).
al. (2019). to However,
systems
because
However, having
of the
its tuning
relative
unique
therelative of
degree
tuning degree the
property
of the one, parameters
one, that
parametersas in
can for
Rath
be
for
deal
is with
controller. a
transformedIn into
Nguyen variation
a polytopic
et al. (2017), LPVof vehicle
the form speed.
which
road-vehicle Borroni
allows
system to applied
nonlinear to systems
high order having
sliding mode controls still as in Rath
deal
is
and
with a
transformed
Tanelli
large into
(2018)
variation
aapropose range
polytopic LPVof vehicle
form
form which
speed. Borroni
allows to et
to nonlinear
al.
applied (2019).
to However,
systems
highHowever, having
order sliding the tuning
relative of
degree
mode controls the still remainsfora
parameters
one, remains
as in for
Rath a
deal
is
and
deal
with a large
transformed
Tanelli
with a
into
(2018)
large
variation
polytopic
propose
variation aa shared-control
range
range
shared-control
LPVof vehicle
of vehicle
formulation
speed.
which Borroni
allows
formulation
speed. Borroni
et
non
et
non
al.
al. (2019).
trivial
nonlinear (2019).
trivial task.
high
task. order
However, the
sliding
the tuning
mode
tuning of the
controls
of the parameters
still remainsfora
parameters
in
and
dealwhich the
Tanelli
with athe
in which assistance
(2018)
largeassistance propose
variationand and a
range the driver
shared-control
the ofdriver steering
vehiclesteering torques
formulation
speed. Borroni
torques nonlinearnonlinear
non trivialhigh high
task.order sliding mode controls still remains a
order sliding mode controls still remains a
and
in
and Tanelli
which the
Tanelli (2018)
assistance
(2018) propose
proposeand a the
a shared-control
driver
shared-control steeringformulation
torques
formulation In
non this paper
trivial task. the design of aa co-operative nonlinear quasi

inThis
which the was
research assistance
funded by andthe the
EuropeandriverUnionsteering
with the torques
Euro- In
non this paper
trivial task. the design of co-operative nonlinear quasi
 continuous sliding mode control in
in
 which
This the
research assistance
was funded by andthe the
Europeandriver steering
Union with the torques
Euro- In this paper
continuous
In this paper
the design
sliding
the mode of
design control in aa steer-by-wire
a co-operative nonlinearvehicle
steer-by-wire quasi
vehicle
pean
This
pean
Regional
researchDevelopment
Regional was funded by
Development
Fund,
the the
Fund, the
FrenchUnion
European
French
 This research was funded by the European Union with the Euro-
State with
State
and the
and the region
the Euro-
region as
In
as lane
continuous
this
lane keeping
paper
keeping mode of
theassistant
sliding design
assistant ofis
is
aa described.
control co-operative The nonlinear
in a steer-by-wire
co-operative
described. The time-varying
nonlinear quasi
vehicle
time-varying quasi
Hauts-de-France.

pean
ThisRegional
research Development
was funded Fund,
by the the French
European State
Union and
with the
the region
Euro- continuous sliding mode control in a steer-by-wire vehicle
Hauts-de-France.
pean Regional Development Fund, the French State and the region as lane
continuous keeping
sliding assistant
mode is
control described.
in a The
steer-by-wire time-varying
vehicle
Hauts-de-France.
pean Regional Development Fund, the French State and the region as
as lane
lane keeping keeping assistant
assistant is is described.
described. The The time-varying
time-varying
Hauts-de-France.
2405-8963 Copyright © 2020 The Authors. This is an open access article under the CC BY-NC-ND license.
Hauts-de-France.
Peer review under responsibility of International Federation of Automatic Control.
10.1016/j.ifacol.2020.12.2343
 Gabriele Perozzi et al. / IFAC PapersOnLine 53-2 (2020) 15332–15337 15333

Fig. 1. Scheme of the bicycle model.

gains of the controller, which are based on the computed
upper-bound disturbance and which depend by physical
variables, help the tuning process. The interaction between
the human driver and the controller is based on a vari- Fig. 2. Visual representation for lane and heading errors.
able parameter representing the driver’s availability. The
controller must be robust to minimize the lane and the
heading errors, and it must be sensitive to help the driver
where ρr is the road curvature, yl the lateral deviation
during the shared control mode, i.e. the assist system
error, ψl the heading error both at the look ahead distance
must give the authority to the driver when the driver is
lp from the vehicle’s center of gravity.
available.
The paper outline is as follows. In Section 2 the vehicle’s
lateral dynamics and the errors’ dynamics are introduced.
The control system and the disturbance upper-bounds are
computed in Section 3. The experimental workspace, the 3. CONTROL SYSTEM FOR LANE KEEPING
validation and the chattering suppression are described
in Section 4. The remarks and discussions on future
improvements conclude the paper in Section 5. The focus of the controller is to take into account the
interaction with the human driver. The controller must be
2. VEHICLE’S AND ERRORS’ DYNAMICS robust to minimize the lane and the heading errors, but
it must be sensitive to assist the driver during the shared
The lane keeping assist systems are controllers useful control mode based on the driver’s availability. For this
to keep the lane in an autonomous driving, taking into reason, the tire steer angle is divided in two parts as
account the human driver interaction. For this reason,
these systems are based on lateral dynamics of the vehicles, δf = (1 − ω)δf a + ωδf m , (3)
on lateral displacement from the center of the lane and on
heading error of the vehicle. where δf a comes from the control system,
The lateral dynamics of the vehicle can be computed from δd
a well established bicycle model as in Fig. 1 (Rajamani δf m =
Rs
(2012)), and it is given by
 comes directly from the driver, δd is the driver’s wheel
mv̇y = Fr + Ff − mvx ψ̇ + Fw
, (1) angle, and ω is a weighted constant reflecting the driver’s
Iz ψ̈ = −lr Fr + lf Ff + lw Fw
availability such as ω = 1 when the driver is available and
where m denotes the vehicle mass, vy = vx φ the lateral the driving is completely manual, ω = 0 when the driver is
velocity, vx the longitudinal driving velocity supposed to not available and the driving is completely automatic, 0 <
be constant, φ the vehicle side sleep angle, ψ the yaw ω < 1 when it is present the shared controller proportional
angle, Iz the yaw moment of inertia, lf , lr , lw respectively to the driver’s availability. Using this decomposition in a
the distances from the center of gravity to the front, steer-by-wire architecture, δf m and δf a are called virtual
to the rear wheels and to the point where the wind angles, and δf is the real angle of the steer system.
disturbance is supposed to influence the vehicle, Fw the
lateral aerodynamic force caused by the wind, In this subsection for synthesis of a control law, a kind
  of HOSMC called quasi-continuous SMC will be applied.
lr ψ̇ − vy lf ψ̇ + vy The first step to design the controller is to compute the
Fr = 2Cr , Ff = 2Cf δf − ,
vx vx error, which has to be minimized, as
are the lateral forces applied on the rear and front wheels
e = k 1 lp ψ l + k 2 yl , (4)
and where Cr , Cf are the front and rear cornering stiffness,
δf denotes the tire steer angle. where k1 , k2 > 0 are tuned parameters, then substituting
For lane keeping systems, additional dynamics must be eqs. (2), (1) in its second derivative, we have
introduced as in Fig. 2, and the equations are
 ë = f + U + ∆, (5)
ψ̇l = ψ̇ − ρr vx
, (2)
ẏl = vy + lp ψ̇ + ψl vx where
15334 Gabriele Perozzi et al. / IFAC PapersOnLine 53-2 (2020) 15332–15337

 
k2 lr l p
f = f (X, δf m ) = Fr − (k1 + k2 )
m I
 z 
l p lf k2
+ δf m ωCf (k1 + k2 ) + ,
Iz m
(6)
 
lp lw k2
∆ = ∆(Fw , ρr , ρ̇r ) = Fw (k1 + k2 ) +
Iz m
2
− ρr k2 v − ρ̇r k1 vx lp , (7)
 
lp lf k2
U = U (δf a ) = δf a (1 − ω) Cf (k1 + k2 ) + ,
Iz m
(8)
are respectively the parts of the dynamics where known
and disturbance unknown terms appear, and the control.
The control can be selected as Fig. 3. Interaction between the controller, the driver and
U = −f + ũ, (9) the vehicle’s prototype.
and substituting eq. (9) in eq. (5), the second derivative
case, when β = 0 and there is no measurement or digital
of the error becomes
noises. In our case, since these restrictions are not satisfied,
ë = ũ + ∆, (10) the convergence is assured with respect to a compact set
where the auxiliary controller ũ must be synthesized. To around the desired trajectory. Then β is tuned accordingly
design a control, which is able to compensate the distur- to achieve a trade-off between control oscillations and
bances, we have to evaluate the upper bounds for them. convergence error.
In the literature, it is often assumed that component of
the disturbance admits a fixed upper bound, which means The interaction between the controller, the human driver,
|∆| ≤ ∆¯ for some known constant ∆ ¯ ≥ 0. Unfortunately, it and the vehicle’s prototype is illustrated in Fig. 3. The
is a conservative hypothesis, and that is why time-varying peculiarities of the proposed controller are: First, the
bounds will be considered in our case for ∆. However, we quasi-continuous SMC (Levant (2005), Ding et al. (2016))
will assume boundedness of the lateral aerodynamic force is tested for co-operative control in autonomous vehicles
by the wind |Fw | ≤ F̄w , for some known F̄w ≥ 0, and with a slightly modification. Second, the upper-bounds
boundedness of road curvature |ρr | ≤ ρ̄r , |ρ̇r | ≤ ρ̇¯r for of Fw , ρr , ρ̇r are considered as input to the time-varying
some known ρ̄r ≥ 0, ρ̇¯r ≥ 0. For this reason, the upper- gain. Their estimation allows a tuning process based on
bound of the reasonable physical meaning, which helps the tuning pro-
 disturbance becomes  cess of the nonlinear controller. Third, the shared author-
l p lw k2
|∆| ≤ |Fw | (k1 + k2 ) + + |ρr | k2 v 2 + |ρ̇r | k1 vx lp ity is assured thanks to a parameter which comes directly
Iz m from the monitoring system and continuously changes over
(11) time based on the driver’s availability. In other words,
In contrast to standard chattering reduction technique the controller must be sensitive to the trajectory changes
based on smoothing of the discontinuous control, the high based on the availability of the human driver (ω). Three
order sliding mode approach suggests to smooth the slid- cases are taken into account: ω = 0 when the driving is
ing motion providing the finite-time convergence to zero completely automatic and the driver is supposed to be not
for the sliding variable e together with the derivatives available; ω = 1 when the driving is completely manual
ė, ë, . . . e(r−1) where r is the relative degree of e with and the driver is available; 0 < ω < 1 when the driving
respect to the control input (r = 2 in our case). In Ding is shared between manual and automatic. Fourth, the
et al. (2016), it is stated that if the second order control chattering effect reduction in the implementation of the
2
¯ |ė| sign(e)+αe high order sliding mode control in the vehicle’s prototype.
is selected as ũ = −∆ |ė|2 +α|e| , where α > 0 is a
There are two possibilities to reduce the chattering of
tuned parameter, then e = ė = 0 is reached in a finite the proposed SMC: implicit discrete-time design, which
time. The proposed quasi continuous SMC can be modi- is preferable but first works are for twisting and super-
fied to counteract the chattering avoiding the saturation twisting controllers as in Brogliato et al. (2018), Huber
functions (mostly used for first order SMCs), and using et al. (2016); or adapt the HOSMC as an approximation
the quasi-continuous function itself as an approximation of the sign on the hyper plane as described in this paper.
of the sign on the plane, with a modification by adding
a small constant β > 0 in the denominator, where β is
4. EXPERIMENTAL VALIDATION
strictly related with accuracy, as in
2
¯ |ė| sign(e) + αe .
ũ = −∆ (12) The experimental validation deals with the implemen-
|ė|2 + α|e| + β tation in a vehicle’s prototype running in discrete-time
The smaller is β the higher is the effort on the steer with sampling time of 0.01 s, the consequent reduction
system, which results in a more accentuated oscillation of chattering effect, and with the evaluation of practical
of the control, but with a smaller convergence error, and performances of the proposed lane keeping assist system
vice-versa. According to Ding et al. (2016), a finite-time with a human driver. The controller must be robust to
convergence of the system can be achieved in the ideal minimize the lateral position and the heading errors, and
 Gabriele Perozzi et al. / IFAC PapersOnLine 53-2 (2020) 15332–15337 15335

it must be sensitive to help the driver during the shared Table 1. Vehicle - Controller parameters
mode, i.e. the assist system must give the authority to the
driver when the driver is available. Description Par. Value Unit
distance CoG , front wheels lf 1.3 m
The dynamic road vehicle simulator SHERPA (Simulateur distance CoG , rear wheels lr 1.6 m
Hybride d’Etude et de Recherche Pour l’Automobile) in distance CoG , wind force lw 0.4 m
LAMIH is used to demonstrate the effectiveness of the look ahead distance lp 5 m
proposed controller. The dynamic simulator uses a modi- mass of vehicle m 2024.86 kg
yaw moment of inertia Iz 2800 kg.m2
fied Peugeot 206 vehicle fixed on a Stewart platform and front wheels cornering stiffness Cf 57000 N/rad
it is structured around a SCANeR network connecting 15 rear wheels cornering stiffness Cr 59000 N/rad
PC-type workstations. The SHERPA simulator is devel- long. velocity in full auto mode vx 20 m/s
oped with RTMaps software composed by several modules steer gear ratio Rs 16 -
which are in charge of different tasks: perception, plan- sliding surface k1 , k 2 1 -
ning, driver monitoring, human–machine interface. It is quasi-continuous controller α, β 1 -
also equipped with a FaceLab driver monitoring system
that indicates the driver state. The driver monitoring 10-4

road curvature (m-1 )

system continuously provides the variable ω in accordance
with the real-time availability of the driver. The tested 5

track comes from the project CoCoVeA (Coopération 0

Conducteur-Véhicule Automatisé), and it is composed by
-5
different road sections, which are used to test the robust-
ness of the controller in a circuit under different road -10
0 50 100 150
curvature conditions. Time (s)
The shared authority is assured thanks to a parameter
which comes directly from the monitoring system and Fig. 4. Profile of the road curvature over time.
continuously changes over time based on the driver’s
availability. For a better presentation of the paper, four
tests, carried out in the same circuit, are illustrated: in
manual mode with the support of a visual line in the center
of the road’s lane as reference for the driver (called manual-
line in the graphics); in automatic mode (called auto in
the graphics); in shared mode with ω = 0.5 (called shared
in the graphics); in shared mode with ω = 0.5 and the
visual line in the center of the road’s lane (called shared-
line in the graphics). The chosen scenarios represent the
three possible cases for shared controllers: the driver is
fully aware of the road (ω = 1); the automatic control
must have the authority because the driver is not available
(ω = 0); the controller must help the driver to follow
correctly the lane (ω = 0.5), for example the controller
must correct the trajectory when a high lateral error
from the center of the lane is present. Values for the
vehicle’s mathematical model (simplified with respect to
the real vehicle in the SHERPA simulator) and for the
controller are in Table 1. The controlled system must be
bound by |yl | ≤ 1.75 m, |ψl | ≤ 5 deg, |vy | ≤ 1.5 m/s,
|v̇y | ≤ 4 m/s2 . The particularity of the proposed controller
is that the gain is time-varying and it changes based
on the driving conditions and the driver’s availability. In
other words, the stability of the vehicle-controller-driver Fig. 5. Errors in manual (manual-line), automatic (auto),
system is always achieved because the gain of the controller shared control (shared ), and shared control with line
changes over time based on the necessity. The tuning reference (shared-line) modes.
process is facilitated and k1 , k2 , α are all chosen equal
to 1. The performance of the controller is affected by the angles δf a and δf m have opposite sign, and to obtain
a correct tuning of these parameters. The road presents a desired smooth transition from the actual to the desired
different sections with varying curvatures as in Fig. 4. driver’s angle. In steer-by-wire architecture, drivers cannot
It’s worth noting that in the steer-by-wire architecture feel the road information because the system does not
the driver’s wheel is not linked mechanically to the steer provide the steering reactive torque reflecting the external
system. For this reason, the angle of the driver’s wheel environment. Then, the steering reactive torque to inform
is set to 0 deg in automatic mode and it is set to δf the driver of the road condition can be designed as in
in shared-control mode, using PID controllers. The use Hiraoka et al. (2008).
of the PID (tuned based on the driver’s preference) is
twofold: to give an haptic feedback to the driver when The errors and the angles δf a , δf m , and δf in the four
modes are illustrated and compared in Figures 5, 6.
15336 Gabriele Perozzi et al. / IFAC PapersOnLine 53-2 (2020) 15332–15337

=0 =1
0.04 0.4

0.02 0.2

yl (m)

yl (m)
0 0
-0.02 -0.2
-0.04 -0.4
50 100 150 50 100 150

0.4
0.2
0.2

(deg)

(deg)
0 0

l
-0.2 -0.2
-0.4
50 100 150 50 100 150
Time (s) Time (s)

Fig. 8. Errors using different β in automatic mode.

Fig. 6. Angles in manual, automatic and shared control
modes.

0.2 Manual
=1
Transition Auto
=0
Transition Manual
=1
of the driver, i.e. it presents less driver’s wheel vibration
0.1 and acceptable errors.
angles (deg)

0 The shared controllers must be evaluated also during

-0.1
-0.2
fa
60 60.5 61 61.5 62
Time (s)
Fig. 7. Transitions between manual and automatic modes.
The manual-line mode presents an unpredictable behavior
based on the personal driver ability to follow the line in the
center of the lane, the driving presents many corrections,
and the steer angle presents many solicitations. In shared-
line mode the driver tested the shared control with the
presence of the reference line in the center of the lane.
The errors are similar to the manual mode, but less
corrections are needed because the driver is helped by
the controller. In this case, the conflict between the driver
operation and the assistance can be seen, and it comes
from the two different references that the driver and the
controller have. The introduction of the novel parameter
β allows the controller to follow as reference a trajectory
which presents a lateral deviation from the center of the
lane, while the driver follows the line which is visibly
at the center of the lane. The auto mode presents a
smoother behavior than the previous cases as expected
since the vehicle follows automatically the center of the
lane with some acceptable lateral deviations coming from
the novel parameter β that it was introduced in this
paper, modifying the original quasi-continuous controller
proposed in Ding et al. (2016). The corresponding steer
angle presents an optimal behavior. In the shared mode
the driver tested the shared control without any auxiliary
visual help to follow the center of the lane. In this scenario
the driver is helped only by the haptic feedback from the
controller. In other words, the driver is free to follow the
lane, and the controller helps to compensate the errors
and provides an haptic feedback to the driver’s wheel
with a small vibration proportional to the errors. This last
mode has the best performance in terms of the compromise
between the lateral and angle position errors and comfort
the transitions phases between manual and automatic
drivings. In this paper, the transition law is governed by
fm f a smooth switch. Fig. 7 shows the performance of the
62.5 63 63.5 64 proposed controller from manual to automatic and vice
versa. The transitions are smooth in both cases to ensure
the best comfort for the driver.
A big shortage of SMC is the chattering (high frequency
oscillations of a discontinuous control signal in the steady-
state mode caused by imprecision of values of model
parameters, digital and measurement noises), which can
decrease drastically the driver’s comfort. In the literature
the problem of chattering reduction is a well-known issue
discussed in many articles. One way to reduce the chat-
tering is to use a high order sliding mode (HOSM) control
(Bernuau et al. (2014)). A kind of HOSM control called
quasi-continuous SMC is chosen in this paper, which can
be considered as an approximation of the sign on the plane.
However, the main difficulty of the experimental validation
is that the controller is designed in continuous-time do-
main, while the implementation is always in discrete-time.
The explicit discretization made by the SCANeR software
does not allow to obtain a smooth sliding motion and it
can destabilize the vehicle-controller system (differences
between explicit and implicit discretizations in SMCs are
explained in Wang et al. (2014)). For this reason, the small
constant β is added, which affects directly the accuracy.
The trade-off between the error and the chattering comes
from the tuning parameter β, which is chosen to have small
lane and heading errors and to avoid the chattering in
the discretized controller. Figures 8, 9 show the influence
of β on the errors and on the control signals in fully
autonomous driving. When β = 0 the controller presents
undesired oscillations with high pitches, but the errors are
minimized. In this case, the chattering produced an unde-
sired vibration in the vehicle and it decreased the comfort
for the drivers who made the tests in the experimental
validation. In opposite, when β = 1 the controller has a
smooth behavior, but with higher lane errors depending
linearly by the amplitude of the road curvature.
 Gabriele Perozzi et al. / IFAC PapersOnLine 53-2 (2020) 15332–15337 15337

1
=0 =1 Borroni, F. and Tanelli, M. (2018). A weighting approach
0.2
to the shared-control of lateral vehicle dynamics. IFAC-
0.5 0.1 PapersOnLine, 51(9), 305–310.
Brogliato, B., Polyakov, A., and Efimov, D. (2018). The
(deg)

(deg)
0 0
implicit discretization of the super-twisting sliding-
fa

fa
-0.5
-0.1 mode control algorithm. In 2018 15th International
-0.2
Workshop on Variable Structure Systems, 349–353.
-1
50 100 150 50 100 150
Dai, M., Wang, J., Chen, N., and Yin, G. (2018). Fuzzy
Time (s) Time (s) steering assistance control for path following of the steer-
by-wire vehicle considering characteristics of human
Fig. 9. Control using different β in automatic mode. driver. In 2018 IEEE Intelligent Vehicles Symposium
5. CONCLUSION (IV), 892–897.
Ding, S., Levant, A., and Li, S. (2016). Simple homoge-
neous sliding-mode controller. Automatica, 67, 22–32.
In this paper, a robust shared SMC approach between
doi:10.1016/j.automatica.2016.01.017.
human driver and automated system is introduced as
Guo, C., Sentouh, C., Popieul, J., and Haué, J. (2017).
lane keeping assistant for road four-wheeled vehicles. The
MPC-based shared steering control for automated driv-
shared authority is assured thanks to a parameter which
ing systems. In 2017 IEEE International Conference on
comes directly from the monitoring system and continu-
Systems, Man, and Cybernetics (SMC), 129–134.
ously changes over time based on the driver’s availability.
Hiraoka, T., Nishihara, O., and Kumamoto, H. (2008).
Upper-bounds of wind force and road curvatures are con-
Steering reactive torque presentation method for a steer-
sidered in the control design, and their estimation allows
by-wire vehicle -Feedback of a lateral acceleration at a
a tuning process based on reasonable physical meaning.
center of percussion with respect to rear wheels-. Review
The implementation of the SMC in the vehicle, which runs
of Automotive Engineering, 29(2), 287–294.
in discrete-time domain, and the consequent chattering
Huber, O., Acary, V., Brogliato, B., and Plestan, F. (2016).
suppression are discussed. The proposed real-time exper-
Implicit discrete-time twisting controller without nu-
iments are used to show the robustness of the automatic
merical chattering: Analysis and experimental results.
controller in a road with different curvatures, to highlight
Control Engineering Practice, 46, 129–141.
the co-operation between the driver and the controller
Levant, A. (2005). Quasi-continuous high-order sliding-
based on the driver’s availability, and to evaluate the
mode controllers. IEEE Transactions on Automatic
transitions between autonomous and manual drivings. As
Control, 50(11), 1812–1816.
results, the controller is robust to minimize the lateral
Nguyen, A.T., Chevrel, P., and Claveau, F. (2017). On
deviation and the heading errors, it is sensitive to the
the Effective Use of Vehicle Sensors for Automatic
trajectory’s change from the human driver when needed, it
Lane Keeping via LPV Static Output Feedback Control.
allows smooth transitions between manual and automatic
IFAC-PapersOnLine, 50(1), 13808–13815.
drivings.
Rajamani, R. (2012). Vehicle Dynamics and Control.
As further improvement the following solutions can be Mechanical Engineering Series. Springer US, 2 edition.
considered. The controller can be designed using an im- Rath, J.J., Senouth, C., and Popieul, J.C. (2019). Person-
plicit discretization. The switching variable ω, which in alised lane keeping assist strategy: adaptation to driving
this work considers only the driver’s availability, can be style. IET Control Theory Applications, 13(1), 106–115.
improved considering the driver’s comfort, the personal Sentouh, C., Nguyen, A., Benloucif, M.A., and Popieul,
driver’s characteristics, the lateral deviation error from the J. (2018). Driver-Automation Cooperation Oriented
center of the lane, and the difference between the angle of Approach for Shared Control of Lane Keeping Assist
the driver’s wheel and angle of the steering system. Systems. IEEE Transactions on Control Systems Tech-
nology, 1–17.
REFERENCES Wang, B., Brogliato, B., Acary, V., Boubakir, A., and
Plestan, F. (2014). Experimental comparisons between
Benloucif, A.M., Nguyen, A., Sentouh, C., and Popieul, implicit and explicit implementations of discrete-time
J. (2019). Cooperative Trajectory Planning for Haptic sliding mode controllers: Towards chattering suppres-
Shared Control between Driver and Automation in sion in output and input signals. In 2014 13th Interna-
Highway Driving. IEEE Transactions on Industrial tional Workshop on Variable Structure Systems, 1–6.
Electronics, 1–1. Wang, J., Dai, M., Yin, G., and Chen, N. (2018). Output-
Benloucif, M.A., Nguyen, A.T., Sentouh, C., and Popieul, feedback robust control for vehicle path tracking consid-
J.C. (2017). A New Scheme for Haptic Shared Lateral ering different human drivers’ characteristics. Mecha-
Control in Highway Driving Using Trajectory Planning. tronics, 50, 402–412.
IFAC-PapersOnLine, 50(1), 13834–13840.
Bernuau, E., Efimov, D., Perruquetti, W., and Polyakov,
A. (2014). On homogeneity and its application in sliding
mode control. Journal of the Franklin Institute, 351(4),
1866–1901.
Boehm, P., Ghasemi, A.H., O’Modhrain, S., Jayakumar,
P., and Gillespie, R.B. (2016). Architectures for Shared
Control of Vehicle Steering. IFAC-PapersOnLine,
49(19), 639–644.