Proceedings of the 20th World Congress

Proceedings
The of the
International 20th World
Federation of Congress
Automatic Control
Proceedings of the 20th World Congress
Proceedings
The of the
International 20th World
Federation of Congress
Automatic Control
Toulouse, France, July 9-14, 2017
The International Federation of Automatic Control
Available online at www.sciencedirect.com
The International
Toulouse, France,Federation of Automatic
July 9-14, 2017 Control
Toulouse, France, July 9-14, 2017
Toulouse, France, July 9-14, 2017

ScienceDirect
IFAC PapersOnLine 50-1 (2017) 5774–5781
Adaptive
Adaptive Cruise
Cruise Control
Control with
with Safety
Safety
Adaptive Cruise
Adaptive Cruise
Guarantees for Control with
Control with
Autonomous Safety
Safety
Vehicles
Guarantees
Guarantees for
for Autonomous
Autonomous Vehicles
Vehicles
Guarantees for Autonomous Vehicles
Silvia Magdici and Matthias Althoff
Silvia
Silvia Magdici and Matthias Althoff
Silvia Magdici
Magdici and and Matthias
Matthias Althoff Althoff
Department of Robotics and Embedded Systems, Technical University
ofDepartment
Department
Munich, Germany
Department
of
of Robotics
of Robotics
Robotics
and
and Embedded
and Embedded
(e-mails: Systems,
Systems, Technical
Systems, Technical
silvia.magdici@tum.de,
Embedded Technical
University
University
althoff@tum.de)
University
of
of Munich,
Munich, Germany
Germany (e-mails:
(e-mails: silvia.magdici@tum.de,
silvia.magdici@tum.de, althoff@tum.de)
althoff@tum.de)
of Munich, Germany (e-mails: silvia.magdici@tum.de, althoff@tum.de)
Abstract: This paper addresses the problem of following a vehicle with varying acceleration in
Abstract:
aAbstract: This
This paper
safe addresses
paper addresses the
the problem of following
consists aaaofvehicle with varying acceleration in
comfortable
Abstract:
apredictive
comfortable Thisand
andpaper
safe
manner. Our
addresses
manner. the problem
Our
architecture
problem
architecture
of
of following
following
consists of
vehicle
a nominal
vehicle
a nominal
with varying
varying acceleration
controller
with controller (here: model
acceleration
(here: model
in
in
a comfortable
a comfortable and
control) safe
and
and safe manner.
a safety
manner. Our architecture
controller.
Our architecture Although consists
model
consists of a nominal
predictive
ofpredictive
a nominalcontrol controller
control attempts
controller (here: tomodel
(here:tomodelkeep
predictive
apredictive control)
control) and
and a safety controller. Although model attempts keep
safe distance,
predictive
aleading
safe control)
distance, it and a
it cannot
cannot a safety
formally
safety
formally
controller.
guarantee
controller.
guarantee
Although
it, due to
Although
it, due
model
model
to the
predictive
predictive control
the assumptions
assumptions
on theattempts
control
on the
behaviorto
attempts
behavior
ofkeep
toof the
keep
the
a safe distance,
vehicle.
a safe distance, it
We cannot
address
it cannot formally
this
formally guarantee
problem
guaranteeby it,
holding due a to the
formally
it, duea formally assumptions
verified
to the assumptions safetyon the behavior
controller
on controller
the behavior of
of the
available.the
leading
leading
Our vehicle.
vehicle.
novel We
We
mechanism address
address this
this
gradually problem
problem
engages by
by holding
holding
the safetya formally
maneuver verified
verified
since safety
safety
most controller
critical available.
available.
situations
leading
Our vehicle.
novel We address
mechanism this problem
gradually engages by theholding
safetya formally
maneuver verified
since safety
most controller
critical available.
situations
Our
Our novel
resolve novel mechanism
quickly. The overall
mechanism gradually
approach
gradually engages
engages the
the safety
is evaluated safety maneuver
against
maneuver since
since most
real traffic data. critical
most The results
critical situations
show
situations
resolve
resolve quickly.
quickly. The
The overall
overall approach
approach is
is evaluated
evaluated against
against real
real traffic
traffic data.
data. The
The results
results show
good
resolve
good
position
position
and
quickly.and The velocity
velocity
tracking
overalltracking
approach performance,
is evaluatedwhile
performance,
while safety
against
safety real and
and
comfort
traffic
comfort data. are
areThe results show
guaranteed.
guaranteed. show
good
good position
position and
and velocity
velocity tracking
tracking performance,
performance, while
while safety
safety and
and comfort
comfort are
are guaranteed.
guaranteed.
© 2017, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: Autonomous vehicles, Intelligent cruise control, Safety guarantees.
Keywords:
Keywords: Autonomous vehicles, Intelligent cruise control,
control, Safety
Safety guarantees.
Keywords: Autonomous
Autonomous vehicles, vehicles, Intelligent
Intelligent cruise cruise control, Safety guarantees.
guarantees.
1. INTRODUCTION (Marzbanrad and Karimian, 2011; Swaroop et al., 1994;
1.
1. INTRODUCTION (Marzbanrad and Karimian, 2011; Swaroop et al., 1994;
1. INTRODUCTION
INTRODUCTION (Marzbanrad
Santhanakrishnan
(Marzbanrad
Santhanakrishnan
and
and Karimian,
and Rajamani,
Karimian,
and
2011;
2011; Swaroop
Rajamani,
2003; Yanakiev
Swaroop
2003;
et
et al.,
al., 1994;
Yanakiev
and
1994;
and
In recent years, one of the most important goals in the Kanellakopoulos, Santhanakrishnan
Santhanakrishnan and
1998). Rajamani,
A
and Rajamani,variety 2003;
of speed
2003; Yanakiev
and
Yanakiev and
deceler-
and
In
In recent
recent years,
years, one
one of
of the
the most
most important
important goals
goals in
in the
the Kanellakopoulos,
Kanellakopoulos,
ation profiles are 1998).
1998).
proposed A
A variety
variety
(Németh of
of speed
speed
and and
and
Gáspár,deceler-
deceler-
2015;
automotive
In recent industry
years, one has
of been
the to
most offer passengers
important goalsthe in high-
the Kanellakopoulos, 1998). A variety of speed and deceler-
automotive ation profiles are proposed (Németh and Gáspár, 2015;
est level of industry
automotive
automotive industry
industry
has
has beenand
has been
safety, comfort, beenand
to
to offer
to offer
offer
passengers
passengers
efficiency
passengers
the
the high-
the high-
by partially high- ation
or Wilson,
ation profiles
2001) are
profiles are proposed
to maintain
proposedthe (Németh
inter-vehicle
(Németh and
and Gáspár,
distance.2015;
Gáspár, Fur-
2015;
est
est level
level of
of safety,
safety, comfort,
comfort, and efficiency
efficiency by
by partially
partially or
or Wilson,
Wilson,
thermore, 2001)
mostto
2001) to ofmaintain
maintain
the work the
theininter-vehicle
inter-vehicle
this direction distance.
distance.
considers Fur-
Fur-a
completely
est level of removing
safety, driving
comfort, duties
and from
efficiency humans.
by Studies
partially or Wilson, 2001) to maintain the inter-vehicle distance. Fur-
completely
completely removing
removing driving
driving duties
duties from
from humans.
humans. Studies
Studies thermore,
thermore,
trade-off most
most
between of
of the
the
safety work
work
and in
in this
this
comfort, direction
direction
yet does considers
considers
not guar-a
a
have shown
completely that
removing active safety
driving systems,
duties from such as
humans. (adaptive)
Studies thermore, most of safety the work in this direction considers a
have
have shown
shown that
that active
active safety
safety systems,
systems, such
such as
as (adaptive)
(adaptive) trade-off
trade-off
antee between
between
safety. safety and
and comfort,
comfort, yet
yet does
does not
not guar-
guar-
cruise
have control,
shown thatelectronic
active stability
safety control
systems, suchor lane
as keeping,
(adaptive) trade-off between safety and comfort, yet does not guar-
cruise control, electronic antee safety.
cruise
which control,
are already on the stability
electronic stability
automotive control
control
market,or
or lane
can keeping,
or lane improve antee
keeping, antee safety.
cruise are
which control,
already electronic
on the stability
automotive control
market, lane
can keeping,
improve Recentsafety.approaches to safe adaptive cruise control use
which
safety
which by are
by already
decreasing
aredecreasing
already onthe on the
the automotive
number
thenumber
automotive of market,
traffic can
accidents
market, improve
(Rieger
can improve Recent approaches to safe adaptive cruise control use
safety of traffic accidents (Rieger Recent
Recent approaches
approaches to
correct-by-construction to safe
control
safe adaptive
software
adaptive cruise control
synthesis,
cruise control use
where
use
safety
et al., by
safety by decreasing
2005). More the
decreasing the number
specifically,
number of
of traffic
Adaptive
traffic accidents
accidents (Rieger
Cruise Control
(Rieger correct-by-construction control software synthesis, where
et al., 2005). More specifically, Adaptive Cruise Control correct-by-construction
the system specifications
correct-by-construction control
are
control software
given
softwarein synthesis,
Linear
synthesis, where
Temporal
where
et al.,
(ACC) 2005).
as
et al., 2005). More
described specifically,
in (ISO,
More specifically, Adaptive
2010),
Adaptive can Cruise
improve Control
Cruise Control traffic the system specifications are given in Linear Temporal
(ACC) the
the system
Logic (LTL)specifications
(Nilsson et al., are given
are 2016, in
in Linear
given 2014). The Temporal
designed
(ACC)
flow
(ACC) andas
as described
described
driving
asdriving
described
in
in
comfort (ISO,
in (ISO,
(Ioannou
(ISO,
2010),
2010),
2010),et etcan
al., improve
can
can improve
1993; Ioannou
improve
traffic
traffic
traffic Logic
system
(LTL) specifications
(Nilsson et al., 2016, 2014).Linear
The Temporal
designed
flow and comfort (Ioannou al., 1993; Ioannou Logic
controller(LTL) which(Nilssonsatisfieset al.,
the 2016,
desired 2014). The
The designed
behavior is based
flow
and
flow and
Chien,
and driving
1993); comfort
driving in addition
comfort (Ioannou
to improving
(Ioannou et
et al.,
al., 1993;
traffic
1993; Ioannou
flow
Ioannou and Logic controller(LTL) which(Nilssonsatisfieset al.,
the 2016,
desired 2014).
behavior designed
is based
and Chien, 1993); in addition to improving traffic flow and controller
on a discrete
controller which
which satisfies
abstraction
satisfies the the
of desired
the
desiredsystem. behavior
behavior is
is based
However, the
based
and
and Chien,
comfort,
Chien, ACC1993);
1993); in
systems
in addition
can also
addition to
to improving
reduce fueltraffic
improving traffic flow
consumption
flow and
and on aa discrete abstraction of the system. However, the
comfort, ACC systems can also reduce fuel consumption on
on a discrete
computation
discrete of abstraction
the finite
abstraction of
of the
abstraction
the system.
is
system. However,
expensive,
However,and the
the
comfort,
(Alam
comfort, et ACC
al.,
ACC systems
2010) and
systems can
trip also
time
can time reduce
(Asadi
also (Asadi
reduce and fuel
and consumption
Vadihi,
fuelVadihi,
consumption2011). computation of the finite abstraction is expensive, and the
(Alam et al., 2010) and trip 2011). computation
size of
computation the of
final
of the
graph
the finite
finite(onabstraction
which
abstraction the is
is expensive,
controller
expensive, and
synthesis
and the
the
(Alam
An
(Alam et al.,
extensive 2010)
et al., 2010) survey and
andon trip
on
trip time
ACC (Asadi
systems
timesystems
(Asadi and and
can Vadihi,
be
Vadihi, found2011).
2011). in size of the final graph (on which the controller synthesis
An extensive survey ACC can be found in size
is
size of
based)
of the
theis final
final graph
exponential
graph (on
in
(on which
the
whichlengththe
the controller
of the
controllerLTL synthesis
formula
synthesis
An extensive
(Vahidi and survey
Eskandarian, on ACC
2003) systems
and (Xiao can
and be
Gao,found2010). in
in is based)
An extensive
(Vahidi and survey on 2003)
Eskandarian, ACC and systems(Xiao can
and be
Gao,found2010). is
and in theis
is based)
based) is exponential
exponential
isdimension
exponential of in
thethe
in
in the
the
length
length
length of
system. of the
of the
LTL formula
the LTL
LTL formula
formula
(Vahidi
(Vahidi and
and Eskandarian,
Eskandarian, 2003)
2003) and
and (Xiao
(Xiao and
and Gao,
Gao, 2010).
2010). and in the dimension of the system.
However, most of the time, the goals of maintaining a safe and in the dimension of the system. and in the dimension of the system.
However, most of the time, the goals of maintaining a safe The ACC problem can also be addressed using control
However,
distance most
and of
of the
the time,
improving the
traffic
theflowgoals byof maintaining
decreasing the aainter-
safe
safe The ACC problem can also be addressed
However,and
distance most improving time,
traffic goals
flow by of maintaining
decreasing the inter- The
The ACC
barrier ACC problem
functions
problem can
can also
(Mehra be
et al.,
also be2015; Ames using
addressed
addressed et al.,control
using
using control
2014).
control
distance
vehicle
distance and
distance
and improving
are
improving traffic
conflicting
traffic flow
flow by
by decreasing
requirements.
decreasing the
the inter-
Therefore,
inter- barrier functions (Mehra et al., 2015; Ames et al., 2014).
vehicle distance are conflicting requirements. Therefore, 1 barrier
These
barrier functions
functions
functions (Mehra
are used
(Mehra et
to
et al.,
al., 2015;
penalize
2015; Ames
the
Ames et
et al.,
violation
al., 2014).
of the
2014).
vehicle
almost
vehicle 50%distance
50%
distanceof are
two-vehicleconflicting
crashes
are conflicting requirements.
are rear-end
requirements. Therefore,
collisions
Therefore, 1 . These functions are used to penalize the violation of the
almost of two-vehicle crashes are rear-end collisions 1 . These
These functions
constraints that
functions are
arise used
are usedfrom toACC
toACCpenalize
penalize the
specifications.violation
the violation of
of the
Therefore,the
almost
If the
almost 50%
leading
50% of two-vehicle
vehicle
of two-vehicle crashes
suddenly
crashes are are rear-end
decelerates collisions
(e.g.
rear-end collisions after1.
. the constraints that arise from specifications. Therefore,
If the leading vehicle suddenly decelerates (e.g. after constraints
propertythat
constraints thatthat arise
thefrom
arise valueACC
from ACCof aspecifications.
control barrier
specifications. Therefore,
function
Therefore,
If the
another leading
vehicle vehicle
cuts-in insuddenly
front of decelerates
the leading (e.g.
vehicle) after
and the propertyinfinity,
that the
If the leading
another vehicle vehicle insuddenly
cuts-in front of decelerates
the leading (e.g. after the
the property
approaches that
that the as value
pointsof
value of aa control
a control
barrier
control barrier
ofapproaching
function
function
the boundaries
another
an
another vehicle
emergency
vehicle cuts-in
situation
cuts-in in
in front
occurs,
front of
of the
the
the leading
ACC
leading is vehicle)
vehicle)
deactivated
vehicle)
and
and
and
propertyinfinity,
approaches the
as value
points approaching barrier
the function
boundaries
an emergency situation occurs, the ACC is deactivated approaches
of the
approachessafe infinity,
region
infinity, as
(i.e.
as points
safe
points approaching
distance becomes
approaching the
thetooboundaries
short),
boundaries is
an
and
an emergency
the driver
emergency situation
becomes occurs,
responsible
situation responsible
occurs, theforthe ACC
for
ACC is
(fully) deactivated
braking.
is deactivated In of the safe region (i.e. safe distance becomes too short), is
and the driver becomes (fully) braking. In of
of the
exploited.safe region
Finding
the safe Finding
region (i.e. (i.e.
a safe
control distance
barrier
safe distance becomes
function,
becomes too short),
however,
too short), is is
is
and
the
and the
near driver
future,
the driver becomes
it
becomesis responsible
assumed
responsible that for (fully)
autonomous
for braking.
(fully) braking. vehicles In
In not exploited. a control barrier function, however,
the near future, it is assumed that autonomous vehicles exploited.
a trivial
exploited. Finding
task. aa control
Finding control barrier
barrier function,
function, however,
however, is is
the
will
the near
take
near future,
over
future, the it is assumed
driving
itdriving duties
is assumed that
from
that autonomous
humans
autonomous vehicles
completely, not a trivial
vehicles not a trivial task. task.
will take over the duties from humans completely,
situations. not a trivial
will
will take
including
take over
safely
over the
the driving
reacting
driving duties
in these
duties from
from humans
emergency
humans completely,
completely, Game theorytask. techniques can also be applied to au-
including
including
Thus, a safely
safely
new ACC reacting
reacting
concept in
in these
these
which emergency
emergency
can always situations.
situations.
guarantee Game
Game theory
theory techniques
in order can
techniques can also
also besafety
appliedandto au-
including
Thus, a safely
new ACC reacting
concept in which
these emergency
can always situations.
guarantee
tonomous
Game theory
tonomous
vehicles
vehicles techniques
in order can
to also be
to increase
increase
applied
besafety
appliedand
to au-
totraffic
au-
traffic
Thus,
safety
Thus, a
and
a new
new ACC
comfort
ACC concept
is required.
concept which
which can
can always
always guarantee
guarantee tonomous
throughput
tonomous vehicles
(Lygeros
vehicles in order
et
in order al., to increase
1998;
to1998;
increaseTomlin safety
safetyet and
al.,
and traffic
2000).
traffic
safety
safety and comfort is required. throughput (Lygeros et al., Tomlin et al., 2000).
safety and
and comfort
comfort is is required.
required. throughput
Each vehicle(Lygeros
throughput
Each vehicle
is considered
(Lygeros
is considered
et
et al.,
al.,an
an
1998;
1998;agent,
agent,
Tomlin
and the
Tomlin
and
et
et al.,
the
2000).
controller
al., 2000).
controller
1.1 Related Work Each
design
Each vehicle
is seen
vehicle is
as
is considered
a game
considered an
between
an agent,
the
agent, and
actions
and the
of
the controller
each agent
controller
1.1 Related Work design
design is
is seen
seen as
as aa game
game between
between the
the actions
actions of
of each agent
1.1 Related
1.1 Related Work Work and
design
and
the
the
disturbances
isdisturbances
seen as a game introduced
betweenby
introduced
by the
thethe
actions of each agent
each
environment.
environment.
Nev-
agent
Nev-
Different space control policies have been presented in ertheless, and
and the disturbances
this
the disturbances approach introduced
has
introduced by
exponential the environment.
complexity.
by the environment. Nev-
Nev-
Different space control policies have been presented in ertheless,
ertheless, this
this approach
approach has
has exponential
exponential complexity.
complexity.
Different
previous
Different space
work
space tocontrol
control policies
check whether
policies ahave
have been
collision
been canpresented
be avoided
presented in
in ertheless,
The MPC this approachishas
framework exponential
widely used to complexity.
tackle adaptive
previous work to check whether a collision can be avoided The MPC framework
previous
previous
1
work
work to to check
check whether
http://www.ntsb.gov/safety/safety-whether a a collision
collision can can bebe avoided
avoided cruise The
The MPC
MPCcontrol problemsis
framework
framework is widely
usingused
widely
is by
widely its to
used
used to
to
tackle
tackle adaptive
tackle adaptive
capability of han-
adaptive
1 http://www.ntsb.gov/safety/safety- cruise
cruise
dling control
control
multiple problems
problems
constraints by
by in using
using
a its
its
receding capability
capability
horizon of
of han-
han-
fashion
1
studies/Documents/SIR1501.pdf
1 http://www.ntsb.gov/safety/safety-
http://www.ntsb.gov/safety/safety-
studies/Documents/SIR1501.pdf
cruise
dling control problems
multiple constraints byinusing
a its capability
receding horizon of han-
fashion
studies/Documents/SIR1501.pdf dling
dling multiple
multiple constraints
constraints in
in a
a receding
receding horizon
horizon fashion
fashion
studies/Documents/SIR1501.pdf
Copyright
2405-8963 ©© 2017,
2017 IFAC 5954Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright © 2017 IFAC 5954
Copyright
Peer review©under
2017 responsibility
IFAC of International Federation of 5954Control.
Automatic
Copyright © 2017 IFAC 5954
10.1016/j.ifacol.2017.08.418
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781 5775
(Bageshwar et al., 2004; Mayne et al., 2000; Corona
et al., 2006; Stanger and del Re, 2013; Li et al.,
2011; Naus et al., 2010). A benchmark setup is pro-
posed in (Corona and Schutter, 2008) to assess differ-
ent model predictive control methods used for ACC. An
overview on constraint MPC can be found in (Maciejowski,
2002); for a comprehensive survey on MPC with con-
straints, the reader is referred to (Mayne et al., 2000).
In the following, we mainly focus on previous work on
ACC which uses MPC, since this work is most closely
related to our approach. In (Bageshwar et al., 2004), a two-
mode ACC is developed using MPC, in which controllers
shift between speed control (transitional operation) and
distance control (steady-state operation). The optimiza-
tion problem is solved subject to desired inter-vehicle dis-
tance and acceleration limitation, which are incorporated
as constraints. In (Li et al., 2011), an optimal control law
is applied in order to increase tracking capabilities and fuel
economy. In order to keep a safe distance between vehicles,
the authors use a constant time headway spacing policy.
The aim of the control problem addressed in (Corona
et al., 2006) is to ensure a minimum distance between
two vehicles. It is assumed that at each sample time,
the host vehicle receives the future reference state of the
leading vehicle. However, if the leading vehicle suddenly
brakes, the host vehicle might not stop within the given
safe distance.
Recently, the idea of cooperative adaptive cruise control
was developed (Stanger and del Re, 2013; Öncü et al.,
2014). String stability, i.e. the capacity to minimize the
tracking errors in the upstream direction of convoys, which
is one of the most important properties of a platoon, is
addressed in (Ploeg et al., 2014; Cook, 2007; Yanakiev and
Kanellakopoulos, 1998). A key component in a cooperative
architecture (platoon) is inter-vehicle communication, i.e.
all entities within the cooperative team know the future
trajectory of the others. However, if the communication is
lost and one of the vehicles performs an unexpected ma-
neuver (e.g. fully braking), a collision might be inevitable.
1.2 Contributions
Designing an ACC concept which simultaneously considers
keeping a safe distance between vehicles and avoiding jerky
maneuvers is not a trivial task, as good tracking capabil-
ities can lead to frequent emergency braking. However, if
the braking is too smooth, a collision might be imminent.
The main contribution of this paper is to design a con-
trol scheme which consists of a nominal controller, which
is supervised by an emergency controller; together they
guarantee safety and comfort at all times. The safety is
achieved by computing a correct safe distance and ensuring
that the inter-vehicle distance is always larger than the
safe distance. First, an optimal control input is generated
utilizing MPC, which minimizes the position error and
the jerk of the host vehicle, guaranteeing performance
and comfort. An emergency maneuver following the op-
timal maneuver is kept available, which is only active
as long as MPC does not provide a safe solution, due
to unexpected disturbances. While switching between an
intelligent cruise control and emergency control is also
considered in a previous work (Mayr and Bauer, 1999),
5955
our approach also ensures limited jerk maneuvers for both
nominal and emergency controllers. Additionally, string
stability is guaranteed during the nominal behavior. By
utilizing our framework, the system can track the com-
puted distance while considering the worst-case scenario
when the leading vehicle suddenly fully brakes. The ride
comfort is guaranteed both during the nominal control,
and during emergency maneuvers, which have a gradual
braking policy while ensuring safety in all circumstances.
1.3 Organization
The remainder of this paper is structured as follows: The
problem description and the assumptions made through-
out the paper are presented in Sec. 1.4. The vehicle model
in addition to safety and comfort constraints is provided
in Sec. 2. In Sec. 3, the general architecture of the ACC
setup is presented. Sec. 4 first analyzes different deceler-
ation profiles and their corresponding braking distances.
Then the emergency and nominal controllers are described.
Numerical evaluations are presented in Sec. 5, followed
by a comparison with a state-of-the art ACC used in the
automotive industry. Finally, the conclusions and future
work are presented in Sec. 6.
1.4 Problem Statement
A typical ACC does not consider an emergency brake
situation where the leading vehicle can suddenly fully
brake. In this situation, a collision might be imminent if
the inter-vehicle distance is not large enough. The goal of
this paper is to design a control scheme that (i) guarantees
safety for all possible scenarios, i.e. a safe distance must
be kept between vehicles, and (ii) ensures comfort at all
times, i.e. there are no jerky maneuvers.
No available communication between vehicles is assumed.
That is, the host vehicle does not know the future veloc-
ity/acceleration profile of the leading vehicle. However, if
communication between vehicles exists, the performance
of the proposed framework would be even better. If there
is no preceding vehicle, ACC behaves like a typical cruise
control system. Note that sensor (e.g. lidars, cameras,
lasers) performance analysis is beyond the work presented
in this paper.
sH sL
aL
aH
δ = sL − sH
dsafe
Host ∆s Lead
Fig. 1. Adaptive cruise control setup.
2. MODELING
In this section, we derive the mathematical model for
vehicle following, as illustrated in Fig. 1. Each vehicle is
described by its absolute position (sH and sL ), velocity
(vH and vL ), and its absolute acceleration (aH and aL ).
The measured distance δ between the host and the leading
vehicle is δ = sL −sH . In the worst-case scenario, where the
leading vehicle fully brakes with the minimum acceleration
Proceedings of the 20th IFAC World Congress
5776
Toulouse, France, July 9-14, 2017 Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781

braking trajectory that brings the ego vehicle to a safe

Initialization stop even when the preceding vehicle would suddenly fully
t ← t + ∆T
brake. As long as the standard controller, which we refer
to as the nominal controller from now on, provides a
safe distance δ ≥ dsafe as shown in Fig. 1, the nominal
Receive new measurements for the leading vehicle;
controller stays in action. Details on how to compute the
update state for the host and the leading vehicle.
safe distance dsafe are presented subsequently in Sec. 4.
In the event that there exists no input u(t) s.t. δ ≥ dsafe ,
Compute: dsafe (t), u(t), asafe (t) the braking trajectory is engaged, which we refer to as
asafe (t). In our work, we do not only consider full braking,
but also discuss several braking profiles with respect to the
yes no
length of dsafe and the jerk values of the braking trajectory.
δ(t) < dsafe (t) A more gradual engagement of brakes decreases jerk and
thus increases comfort, while enlarging the required safe
Apply asafe (t) Apply u(t)
distance dsafe . Controllers for tracking the pre-computed
braking trajectory are not discussed in this work to focus
on the novel aspect of guaranteeing collision avoidance. If,
during the braking maneuver, the inter-vehicle distance
again becomes δ > dsafe (since dsafe has shortened due
Fig. 2. Control scheme of our proposed ACC concept. to the fact that the preceding vehicle has not engaged
brakes to the expected extent), the control is taken back
amin , the braking distance of the leading vehicle can be to the nominal controller. Since (i) we choose our braking
computed by substituting the final velocity with 0 (stand- profiles such that they initially only engage mildly and
still) in the equation of motion, 0 = vL2 − 2|amin |dlead , so (ii) in almost all cases, control quickly goes back to the
that nominal controller, passengers would not realize in most
vL2 cases that the braking trajectory is engaged as discussed
dlead = . (1)
2 |amin | in Sec. 5.
The model used to design the nominal controller for the In this work, we use model predictive control (MPC) as
ACC-equipped vehicle, considering constant velocity aL = a nominal controller. Any controller can be used just as
0, is described as follows: well as MPC in the proposed framework. MPC is used
01 0 0
   
because it provides optimal solutions while attempting to
ẋ = Ax + Bu, A = 0 0 1 , B = 0 , meet constraints—this, however, is not always achieved
00 0 1 in this work, due to assumptions about the behavior of

∆s
 the leading vehicle, which will not exactly materialize.
x = ∆v , u = jH , Our MPC is computed based on the assumption that the
ahost leading vehicle moves with constant velocity, which is a
∆s = δ − dsafe , ∆v = vL − vH , (2) reasonable assumption for optimizing ride comfort, but
safety cannot be ensured since the leading vehicle might
where the control variable u is the jerk of the host vehicle brake.
jH (jerk is the time derivative of acceleration). The state
and control inputs are only allowed to take values within Therefore, an emergency controller has to be applied when
the following intervals: a critical situation occurs. The control output of the MPC
0 ≤ ∆s ≤ ∆smax , is denoted by u(t). Our cost function for MPC is rather
standard and can be formulated as a quadratic program-
amin ≤ aH ≤ amax ,
ming (QP) problem; all matrices have appropriately cho-
jmin ≤ jH ≤ jmax , (3) sen dimensions:
where ∆smax , amin , amax , jmin , and jmax are user-specified min J(x(k), u(k)) = xTN |k P xN |k +
parameters. An additional constraint is considered for the u
acceleration of the host vehicles, in order to achieve string N
 −1  
stability (Kianfar et al., 2011): + xTi|k Qxi|k + uTi|k Rui|k , subject to: (3)-(4),
aH,k ≤ max |aL,τ |, k ∈ {1, . . . , N }, (4) i=0
τ ∈[k−H,k] where:
where k is the current time instant. Parameter N is the
• xi|k and ui|k are the state and input at time instant
prediction horizon, and H is the size of the time window,
i, i ∈ N, i ≤ N , based on the state measurement at
which must be long enough to account for delays arising
time instant k,
in the platoon. Both N and H are hand-tuned.
• J(·, ·) is the cost function,
• matrix Q ≥ 0 is weighting the state vector,
3. ACC WITH SAFETY GUARANTEES
• matrix R > 0 penalizes the control input,
• terminal cost P is chosen to guarantee stability.
The main objective of this work is to embed standard
controllers for adaptive cruise control into a framework Let us denote with Xv (·), v ∈ {L, H} the resulting position
that guarantees collision avoidance. The main idea for of the host vehicle H and the lead vehicle L, by applying
achieving this objective is to always hold available a safe the control input u or the acceleration a. To summarize

5956
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781 5777

x
x
XL (a = 0)

))
XL (amin )

δ̃(t2 )
dsafe (t2 )
XL (amin )

(t
δ(t2 )

H
(u
dsafe (t1 )
Lead: Lead:

X
t))
XH (u(

δ(t1 )

δ(t1 )
(t)) ))
XH (asafe (t
δ(t0 )

δ(t0 )

✗
XH (asafe )
t 1)
(u( (t 1))✓
XH fe
XH (u(t0 ))✓ XH
(a sa
Host: Host:

t0 t1 t t0 t1 t2 t
(a) Case where ∃u(tk ) s.t. δ(tk+1 ) ≥ dsafe (tk+1 ); apply (b) Case where ∄u(tk ) s.t. δ(tk+1 ) ≥ dsafe (tk+1 ); apply
u(tk ). asafe (tk ).

Fig. 3. Switching the control based on the current measurement.

the proposed control scheme, first, an optimal control
output u(tk ) is generated, under the assumption that the
leading vehicle is driving with constant velocity vL . At each
sample time, we verify if by applying u(tk ) for one time
step, the safety distance dsafe is satisfied, i.e. there exists
an emergency maneuver asafe (tk ) that can bring the host
vehicle to a standstill while avoiding any collision, even in
the worst case scenario when the leading vehicle brakes
with full deceleration amin (see Fig. 3(a)). If the verified
control output u(tk ) yields a safe distance, then u(tk ) is
applied to the system.
Let us introduce δ̃(t) as the intervehicle distance at time t,
if the ego vehicle applies asafe (t). If no control output u(tk )
is found such that the safe distance dsafe (tk+1 ) ≤ δ(tk+1 )
(as illustrated in Fig. 3(b)) is met, a gradual emergency
maneuver asafe (tk ) is applied, which guarantees safety
δ̃(tk+1 ) ≥ dsafe (tk+1 ), for any pre-defined deceleration
profile of the leading vehicle. The emergency maneuver is
applied until a new verified control output u(tk ) is found.
4. BRAKING DISTANCE AND DECELERATION
PROFILE
A formal analysis to compute a safe distance is presented
in (Rizaldi et al., 2016), but only the case where constant
acceleration is applied is considered. Here, four different
deceleration profiles are analyzed. Out of those, we select
the solution which guarantees safety at all times and ad-
ditionally, which assures comfort, by generating minimum
jerk. Based on these criteria, we propose a mixed decel-
eration profile asafe (t), and we compute the safe braking
distance dhost , so that a collision is avoided while keeping
low jerk values.
Full deceleration. The most straightforward approach
is to apply constant full deceleration: asafe (t) = amin ,
∀t ≥ 0. This profile provides the smallest safe distance
possible. However, applying full deceleration leads to un-
comfortable driving. Moreover, due to the jerky behavior,
traffic flow might not be improved with this profile.
amin ≤ asafe (t) < 0. The jerk value is low since the acceler-
ation is linearly decreasing. However, the braking distance
of this profile is larger than the full deceleration profile
described previously.
Exponential deceleration. The exponential decelera-
tion profile is defined as follows:
asafe (t) = 1 − ct , ∀t > 0, c > 1, amin ≤ asafe (t) < 0. (5)
When applying the exponential deceleration, the jerk value
is even less compared to linear deceleration. Therefore, if
the leading vehicle fully brakes for only one time step,
the host vehicle will smoothly brake, making this decel-
eration profile suitable for systems whose measurements
are affected by noise. In the following, the computation
of the braking distance is derived: Let s be the solution
of the differential equation s̈ = asafe (t). We define the
braking time of the host vehicle tstop,H as the time when
the velocity reaches 0, where the initial velocity is v0 . The
braking distance dhost is the exact solution of s̈ = asafe (t),
computed for the braking time tstop,H , by double integrat-
ing the acceleration:
1 t2 ctH t stop,H
dhost = 2 + H − 2 + + v0 tstop,H . (6)
ln c 2 ln c lnc
Mixed deceleration. The main drawbacks of the pre-
vious deceleration profiles are (i) if constant maximum
deceleration is applied, the jerk is a Dirac function, so
the value goes to infinity, and (ii) by applying linear or
exponential deceleration, the jerk is comfortable, but the
braking distance is too long. To overcome these disadvan-
tages, a mixed deceleration profile is proposed. Parameter
tH represents the time when the maximum deceleration
is reached during exponential deceleration so that we
continue with full braking. In the following, the mixed
deceleration profile is utilized.
1 − ct if t ≤ tH ,
asafe (t) =
amin if tH < t ≤ tstop,H ,
tH = logc (1 − amin ),
t > 0, c > 1, amin ≤ asafe (t) < 0. (7)

Linear deceleration. Another possible profile is lin- 1

amin t vH Let sL (t) = sL (t0 ) + vL (t0 )t − |amin |t2 be the position
ear deceleration: asafe (t) = , ∀t ≥ 0, c ≥ , 2
c amin and vL (t) = vL (t0 ) − |amin |t the velocity of the leading

5957
Proceedings of the 20th IFAC World Congress
5778
Toulouse, France, July 9-14, 2017 Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781

dL a=0 if ∆v > 0 then ∆s(t) is increasing on ∆τi ⇒

dH amin di = sL (tmin ) − sH (tmin );
in amin
am in if ∆v < 0 then ∆s(t) is decreasing on ∆τi ⇒
am di+1
sL di = sL (tmax ) − sH (tmax ).
• Case (b): ∆v = 0 − vH (t) < 0 then ∆s(t) is decreasing
Distance[m]

di
dsafe
t
c on ∆τi ⇒ di = sL (tmax ) − sH (tmax ).
− di−1
1 • Case (c): ∆v = 0 − vH (t) < 0 then ∆s(t) is decreasing
on ∆τi ⇒ di = sL (tmax ) − sH (tmax ).
s̃H
Lead
Host-min{∆s}
• Case (d): ∆v is computed by integrating the corre-
dmin
∆τi−1 ∆τi+1
Host sponding acceleration difference,
∆τi
sH 1 − ct
t0 tH tstop,L tstop,H ∆v = amin t + vL − t − vH − ; to find if ∆s is increas-
Time[s] ln(c)
Fig. 4. Safe distance computation. ing or decreasing, the solution of ∆v = 0 is computed,
vehicle at time t, when full brake is applied; let sH (t) be which provides the critical point of ∆s as:
 
the position of the host vehicle. The distance dsafe which ln(c)
guarantees safety is computed as: q ln(c) + p LambertW 0, √
∗
p p cq
dsafe = sL (t0 ) − sH (t0 ) − dmin , t =− ,
where the distance between sL (t) and sH (t) over a time p ln(c)
interval ∆τi is where p = (amin − 1)ln(c), q = 1 + (vL − vH )ln(c), and
dmin = min di , di = min (sL (t) − sH (t)). the LambertW function is the inverse function of f (W ) =
i t∈∆τi
W eW . To check if ∆s(t) has a minimum or a maximum
To compute dmin , we exploit the monotonicity of di : Both value at time t∗ , we compute the second derivative of ∆s,
sL (t) and sH (t) are monotonically increasing over time i.e. ∆a(t) = amin − (1 − ct ). Since ∆a(t) < 0 ⇒ ∆s(t∗ )
intervals ∆τi = [tmin , tmax ], tmin , tmax ∈ {tstop,L , tH , has a maximum at t∗ . Therefore, the minimum of ∆s(t)
tstop,H }, where can be at either tmin or tmax . Three further cases can be
vL ln(1 − amin ) distinguished:
tstop,L = , tH = ,
|amin | ln(c)
  (d.1): t∗ < tmin ⇒ ∆s(t) is decreasing on ∆τ ⇒ di =
1 1 − ctH sL (tmax ) − sH (tmax );
tstop,H = + vH ,
|amin | ln(c) (d.2): t∗ > tmax ⇒ ∆s(t) is increasing on ∆τ ⇒ di =
sL (tmin ) − sH (tmin );
tstop,L is the braking time of the leading vehicle.
(d.3): t∗ ∈ [tmin , tmax ] ⇒
Next, to analyze the monotonicity of di , we first compute 
all possible permutations between tH , tstop,H , and tstop,L , ∆s(tmin ), if ∆s(tmin ) < ∆s(tmax )
di = .
since the acceleration mode changes at these points in ∆s(tmax ), if ∆s(tmin ) ≥ ∆s(tmax )
time. There are six possible scenarios, based on the applied
accelerations (alead ∈ {0, amin} and ahost ∈ {0, amin, 1 − ct }). To summarize, first the time intervals
The cases when both vehicles are standing still and when [tmin , tmax ], tmin , tmax ∈ {tH , tstop,H , tstop,L } are selected
the host vehicle is standing still and the leading vehicle is depending on the scenario. Then, dmin is computed accord-
braking are not considered because they already represent ingly, based on the applied deceleration profiles. Finally,
safe situations; therefore, only the remaining four combi- the safe distance dsafe is computed such that any collision
nations of acceleration are analyzed, as shown in Tab. 1. is avoided by applying the proposed deceleration profile
asafe (t).
Table 1. Possible combinations of applied de-
celeration. 5. NUMERICAL EXPERIMENTS

case (a) case (b) case (c) case (d) The presented approach is evaluated with real traffic
Lead amin standstill standstill amin
data for more than 300 vehicles. The data is collected
on a segment of US highway 101 (Hollywood Freeway)
Host amin amin 1-ct 1-ct
located in Los Angeles, California, on June 15th, 2005,
as part of the Next Generation SIMulation (NGSIM) 2
In the following, we compute each di , i ∈ {1, 2, 3, 4} project. In the simulations, the vehicles from the dataset
(for each aforementioned case) considering all possible are considered as leading vehicles in the ACC setup. For
combinations of the applied deceleration of the host each vehicle, the following information is available at
and leading vehicle for each time interval ∆τi . For ex- each sampled time: position, velocity, and acceleration.
ample, in Fig. 4 three different cases can be distin- Additionally, the time step ∆T is introduced. In the
guished: t ∈ [t0 , tH ] : case(d), t ∈ [tH , tstop,L ] : case(a); typical scenarios, the lead vehicle is driving with variable
t ∈ [tstop,L , tstop,H ] : case(b); the case(c) would occur if t ∈ acceleration; however, in order to make the scenarios
[tstop,L , tH ], tstop,L ≤ tH . even more difficult, sudden brakes are added. The host

Each case is analyzed, and di is computed. 2 https://www.fhwa.dot.gov/publications/

• Case (a): ∆v = vL (t) − vH (t), t ∈ ∆τi ; research/operations/its/06135/

5958
 Proceedings of the 20th IFAC World Congress
Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781 5779
Toulouse, France, July 9-14, 2017
vehicle is positioned behind the leading vehicle, with initial
randomly generated velocity and acceleration.
Table 2. Parameters.
Parameter N [-] ∆T [s] v[m/s] a[m/s2 ] j[m/s3 ]
Value 3 0.1 [0,60] [-10,10] [-2,2]
The parameter values used for all considered scenarios are
shown in Tab. 2.
Safe MPC-based ACC. We analyze the simulation
results by using two different deceleration profiles: full
deceleration and mixed deceleration. For both cases, we
evaluate the arithmetic mean j, ∆s, d, and the stan-
dard deviation σj , σ∆s , σd associated with the variables
j, ∆s, and d for all considered vehicles, which are pre-
sented in Tab. 3.
Table 3. Simulation results.
Brake j[m/s3 ] σj [m/s3 ] ∆s[m] σ∆s [m] d[m]
Full -0.005 0.883 3.369 3.557 22.073
Mixed -0.006 0.298 0.287 1.071 23.773
Although the mean jerk value generated by applying full
deceleration is small, the standard deviation shows that
there is a broader range of jerk values, as can be seen in
Tab. 3. Moreover, because of the frequent full braking, the
safe distance tracking parameter, determined by ∆s, shows
less performance, compared with the case when mixed
deceleration is applied. For mixed deceleration, it can be
seen that j is small, which indicates comfortable driving
without jerky maneuvers. The standard deviation σj is
also small; thus most of the jerk values are close to the
mean value. The results show good tracking performance,
as the mean value of ∆s is small. The average distance
between vehicles is around 23m, which is comparable to
the distance provided by the two-seconds distance rule 3
(Martinez and de Wit, 2007), considering that the average
velocity is 10.72m/s.
For illustration purposes, we only present the detailed
simulation results for one considered scenario, whose dura-
tion is more than 2 minutes. The simulation results when
applying the full deceleration profile are depicted in Fig. 5.
It can be seen that big variations in the host vehicle’s
acceleration lead to big variations in velocity (see Fig. 5).
Moreover, the jerk caused by often applying full braking
results in uncomfortable driving.
Platooning using safe MPC-based ACC. To vali-
date the string stability, a four-vehicle setup is considered
as follows: The trajectory of vehicle #1 is taken from
the US101 dataset; the other vehicles are placed behind
one another, and they are controlled by our proposed
algorithm. The task is that vehicle #2 safely follows vehicle
#1, vehicle #3 follows vehicle #2, and vehicle #4 follows
vehicle #3.
3 http://www.rotr.ie/rules-for-driving/speed-limits/
speed-limits 2-second-rule.html
Velocity [m/s] Acceleration [m/s2 ]
5
0
-5
-10
20 40 60 80 100 120 140
∆s[m]
Time [s]
[0,10] 20
10
Lead
Host
0
0 50 100 140
Time [s]
Fig. 5. Safe MPC-based ACC: Full deceleration when
applying the emergency controller.
Here, we choose the mixed deceleration profile when apply-
ing the emergency controller, and it can be seen that the
ACC-equipped vehicles (#2, #3, #4) smoothly follow the
leading vehicle for the entire considered time; the velocity
of the ACC-equipped vehicles also smoothly follows the
velocity of the leading vehicle (see Fig. 6).
In order to not violate the safe distance, the safety mech-
σd [m] anism is engaged for considered vehicles #2, #3, and #4
in 10.14%, 8.84%, and 8.63% of the time, respectively.
11.244 However, the jerk value is kept between the specified
6.080 comfortable value range (Hoberock, 1976). While the lead
vehicle suddenly performs full braking, the ACC-equipped
vehicles smoothly decelerate. Additionally, the position
error ∆s introduced by the leading vehicle braking is
attenuated in the upstream direction, as illustrated in
Fig. 6.
The mean jerk values j and the standard deviation σj are
small (see Tab. 3), which implies comfortable driving with-
out jerky maneuvers. Keeping the inter-vehicle distance as
close as possible to the safe distance dsafe by minimizing
∆s shows good tracking performance. In this way, both
safety and comfort are achieved by utilizing the proposed
ACC concept.
The simulations were performed on a machine with 2.2
GHz, Intel i7 processor, and 16 GB 1600 MHz DDR3
memory, in Matlab R2015a. For solving the QP problem,
the quadprog function from the Optimization Toolbox 4 is
used. The mean value of the computation time is 0.08s;
therefore, the approach is real-time capable.
PI-based ACC. Finally, we compare our method with
a state-of-the-art ACC approach applied in the auto-
motive industry (Corona and Schutter, 2008; Yanakiev
and Kanellakopoulos, 1998), which utilizes proportional-
integral control (PI). Here, we use an implementation
based on (Yanakiev and Kanellakopoulos, 1998), where the
desired inter-vehicle distance is a function of a constant
spacing, a constant time headway, and the velocity of
the leading vehicle. Of course, other spacing policies can
be used, as proposed in the aforementioned papers (e.g.:
variable time headway).
Although the algorithm performs well with respect to
position and velocity tracking, the PI controller itself
cannot guarantee safety. Therefore, the controller fails
to safely track the desired inter-vehicle distance (i.e. the
4 https://www.mathworks.com/help/pdf doc/optim/optim tb.pdf
Proceedings of the 20th IFAC World Congress
5780
Toulouse, France, July 9-14, 2017 Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781

20 6. CONCLUSION AND FUTURE WORK

Velocity [m/s]

15
In this paper an ACC architecture consisting of an emer-
10 gency and a nominal controller is designed in order to
vehicle#1
vehicle#2 ensure safety and comfort. Safety is ensured by switch-
5
vehicle#3 ing between the nominal controller and the emergency
vehicle#4 controller. The nominal controller is based on MPC, and
0
0 50 100 140 it computes optimal inputs such that the safe distance
Time [s]
is intended to be kept. In the emergency controller, the
5 braking distance is solved analytically and computed based
Acceleration [m/s2 ]

on the deceleration profile of the host vehicle, considering

0 at each time step that the leading vehicle can fully brake.
Moreover, the emergency deceleration profile is computed
vehicle#1 such that jerk values remain in the specified comfortable
-5 vehicle#2
vehicle#3 range. The proposed algorithm is evaluated using real
-10
vehicle#4 traffic data, and it shows good performance on position
0 50 100 140 and velocity tracking for all considered vehicles. Thus,
Time [s]
we can conclude that our approach guarantees safety and
comfort for ACC-equipped vehicles, which can take over
5
Position error ∆s[m]

∆s, veh.#1,#2 the driving duties completely.

4 ∆s, veh.#2,#3
∆s, veh.#3,#4
3 ACKNOWLEDGEMENTS
2
1
The authors gratefully acknowledge financial support by
the German Research Foundation (DFG) Graduiertenkol-
0
0 50 100 140 leg 1480 (PUMA) and Grant AL 1185/3-1.
Time [s]

REFERENCES
Fig. 6. Platooning using safe MPC-based ACC: Mixed
Alam, A.A., Gattami, A., and Johansson, K.H. (2010).
deceleration when applying the emergency controller.
An experimental study on the fuel reduction potential
of heavy duty vehicle platooning. In Proc. of the IEEE
position errors ∆s have negative values) as can be seen in International Conference on Intelligent Transportation
Fig. 7. Systems, 306 – 311.
Ames, A., Grizzle, J.W., and Tabuada, P. (2014). Control
barrier function based quadratic programs with appli-
cation to adaptive cruise control. In Proc. of the IEEE
Velocity [m/s]

20
Conference on Decision and Control, 6271 – 6278.
Asadi, B. and Vadihi, A. (2011). Predictive cruise control:
10 Utilizing upcoming traffic signal information for improv-
vehicle#1 ing fuel economy and reducing trip time. IEEE Trans-
vehicle#2
0
actions on Control Systems Technology, 19(3), 707–714.
0 50 100 140 Bageshwar, V.L., Garrard, W.L., and Rajamani, R.
Time [s] (2004). Model predictive control of transitional maneu-
5 vers for adaptive cruise control vehicles. IEEE Transac-
Acceleration [m/s2 ]

tions on Vehicular Technology, 53(5), 1573–1585.

0 Cook, P.A. (2007). Stable control of vehicle convoys for
safety and comfort. IEEE Transactions on Automatic
-5 Control, 52(3), 526 – 531.
vehicle#1
vehicle#2 Corona, D., Lazar, M., Schutter, B.D., and Heemels, M.
-10 (2006). A hybrid MPC approach to the design of a
0 50 100 140 smart adaptive cruise controller. In Proc. of the IEEE
Time [s]
International Conference on Control Applications, 231–
Position error ∆s [m]

8 236.
6 Corona, D. and Schutter, B.D. (2008). Adaptive cruise
4 control for a SMART car: A benchmark for MPC-PWA
2 control methods. IEEE Transactions on Control Sytems
0 Technology, 16(2), 365–372.
-2 Hoberock, L.L. (1976). A survey of longitudinal accelera-
-4 tion comfort studies in ground transportation vehicles.
0 50 100 140
Time [s] Technical report, University of Washington, Department
of Transportation.
Fig. 7. Simulation results. PI-based ACC. Ioannou, P., Xu, Z., Eckert, S., Clemons, D., and Sieja,
T. (1993). Intelligent cruise control: Theory and experi-

5960
 Proceedings of the 20th IFAC World Congress
Toulouse, France, July 9-14, 2017 Silvia Magdici et al. / IFAC PapersOnLine 50-1 (2017) 5774–5781 5781
ment. In Proc. of the IEEE Conference on Decision and
Control, volume 2, 1885 – 1890.
Ioannou, P.A. and Chien, C.C. (1993). Autonomous intel-
ligent cruise control. IEEE Transactions on Vehicular
Technology, 42(4), 657 – 672.
ISO (2010). ISO 15622:2010, Inteligent transport sys-
tems – Adaptive cruise control systems – Performance
requirements and test procedures. Technical report,
International Organization for Standardization.
Kianfar, R., Falcone, P., and Fredriksson, J. (2011). A
receding horizon approach to string stable cooperative
adaptive cruise control. In Proc. of the IEEE Inter-
national Conference on Intelligent Transportation Sys-
tems, 734 – 739.
Li, S., Li, K., Rajamani, R., and Wang, J. (2011). Model
predictive multi-objective vehicular adaptive cruise con-
trol. IEEE Transactions on Control Systems Technol-
ogy, 19(3), 556 – 566.
Lygeros, J., Godbole, D.N., and Sastry, S. (1998). Ver-
ified hybrid controllers for automated vehicles. IEEE
Transactions on Automatic Control, 43(4), 522–539.
Maciejowski, J.M. (2002). Predictive control with con-
straints. Prentice Hall.
Martinez, J.J. and de Wit, C.C. (2007). A safe longitu-
dinal control for adaptive cruise control and stop-and-
go scenarios. IEEE Transactions on Control Systems
Technology, 15(2), 246–258.
Marzbanrad, J. and Karimian, N. (2011). Space control
law design in adaptive cruise control vehicles using
model predictive control. Journal of Automobile En-
gineering, 225(7), 870–884.
Mayne, D.Q., Rawlings, J.B., Rao, C.V., and Scokaert,
P.O.M. (2000). Constrained model predictive control:
Stability and optimality. Automatica, 36(6), 789–814.
Mayr, R. and Bauer, O. (1999). Safety issues in intelligent
cruise control. In Proc. of the IEEE International
Conference on Intelligent Transportation Systems, 970–
975.
Mehra, A., Ma, W.L., Berg, F., Tabuada, P., Grizzle, J.,
and Ames, A. (2015). Adaptive cruise control: Ex-
perimental validation of advanced controllers on scale-
model cars. In Proc. of the IEEE American Control
Conference, 1411 – 1418.
Naus, G., Ploeg, J., de Molengraft, M.V., Heemels, W.,
and Steinbuch, M. (2010). Design and implementation
of parametrized adaptive cruise control: An explicit
model predictive control approach. Control Engineering
Practice, 18, 882–892.
Németh, B. and Gáspár, P. (2015). Design of low conflict
cruise control for safety critical vehicle interactions . In
Proc. of the 19th IFAC Symposium on Fault Detection,
Supervision and Safety of Technical Processes, 1186–
1191.
Nilsson, P., Hussien, O., Chen, Y., Balkan, A., Rungger,
M., Ames, A., Grizzle, J., Ozay, N., Peng, H., and
Tabuada, P. (2014). Preliminary results on correct-
5961
by-construction control software synthesis for adaptive
cruise control. In Proc. of the IEEE Conference on
Decision and Control, 816 – 823.
Nilsson, P., Hussien, O., Balkan, A., Chen, Y., Ames, A.,
Grizzle, J., Ozay, N., Peng, H., and Tabuada, P. (2016).
Correct-by-construction adaptive cruise control: Two
approaches. IEEE Transactions on Control Systems
Technology, 24(4), 1–14.
Öncü, S., Ploeg, J., van de Wouw, N., and Nijmeijer, H.
(2014). Cooperative adaptive cruise control: Network-
aware analysis of string stability. IEEE Transactions on
Intelligent Transportation Systems, 15(4), 1527–1537.
Ploeg, J., van de Wouw, N., and Nijmeijer, H. (2014).
Lp string stability of cascaded systems: Application
to vehicle platooning. IEEE Transactions on Control
Systems Technology, 22(2), 786–793.
Rieger, G., Scheef, J., Becker, H., Stanzel, M., and Zo-
bel, R. (2005). Active safety systems change accident
environment of vehicles significantly - A challange for
vehicle design. In Proc. of the International Conference
on Enhanced Safety of Vehicles.
Rizaldi, A., Immler, F., and Althoff, M. (2016). A
formally verified checker of the safe distance traffic
rules for autonomous vehicles. In Proc. of the NASA
Formal Methods: 8th International Symposium, 175–
190. Springer.
Santhanakrishnan, K. and Rajamani, R. (2003). On
spacing policies for highway vehicle automation. IEEE
Transactions on Intelligent Transportation Systems,
4(4), 198 – 204.
Stanger, T. and del Re, L. (2013). A model predictive
cooperative adaptive cruise control approach. In Proc.
of the IEEE American Control Conference, 1374 – 1379.
Swaroop, D., Hedrick, J.K., Chien, C.C., and Ioannou,
P.A. (1994). A comparison of spacing and headway con-
trol laws for automatically controlled vehicles. Journal
of Vehicle System Dynamics, 23(8), 597–625.
Tomlin, C., Lygeros, J., and Sastry, S. (2000). A game the-
oretic approach to controller design for hybrid systems.
Proc. of the IEEE, 88(7), 949 – 970.
Vahidi, A. and Eskandarian, A. (2003). Research advances
in intelligent collision avoidance and adaptive cruise con-
trol. IEEE Transactions on Intelligent Transportation
Systems, 4(3), 143–153.
Wilson, B.H. (2001). How soon to brake and how hard to
brake: Unified analysis of the envelope of opportunity
for rear-end collsion warnings. Technical report, Inter-
national Society of Automotive Engineers.
Xiao, L. and Gao, F. (2010). A comprehensive review
of the development of adaptive cruise control systems.
Vehicle System Dynamics, 48(10), 1167 – 1192.
Yanakiev, D. and Kanellakopoulos, I. (1998). Nonlin-
ear spacing policies for automated heavy-duty vehi-
cles. IEEE Transactions on Vehicular Technology, 47(4),
1365–1377.