See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/306073071

Decentralized management of intersections of automated guided vehicles

Article · December 2016

DOI: 10.1016/j.ifacol.2016.07.669

CITATIONS READS

6 103

4 authors, including:

Alexandre Lombard Florent Perronnet

Université de Technologie de Belfort-Montbéliard Université de Technologie de Belfort-Montbéliard
6 PUBLICATIONS   15 CITATIONS    16 PUBLICATIONS   135 CITATIONS   

SEE PROFILE SEE PROFILE

Abdeljalil Abbas-Turki
Université de Technologie de Belfort-Montbéliard
56 PUBLICATIONS   399 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Cooperative intelligence for complex traffic environment View project

All content following this page was uploaded by Abdeljalil Abbas-Turki on 02 October 2018.

The user has requested enhancement of the downloaded file.

IFAC Conference on Manufacturing Modelling,
IFAC
IFAC Conference
and on
Conference on Manufacturing
Manufacturing Modelling,
Management
IFAC Conference
Management and on Manufacturing Modelling,
Control
Control Modelling,
Management
June 28-30,
Management and
2016. Control
Troyes,
and Control France Available online at www.sciencedirect.com
June
June 28-30, 2016. Troyes, France
June 28-30,
28-30, 2016.
2016. Troyes,
Troyes, France
France

ScienceDirect
IFAC-PapersOnLine 49-12 (2016) 497–502
Decentralized
Decentralized management
management of
of intersections
intersections
Decentralized
Decentralized management
management of
of intersections
intersections
of
of automated
automated guided
guided vehicles
vehicles
of automated guided vehicles
of automated guided vehicles
Alexandre Lombard ∗∗ Florent Perronnet ∗∗ ∗∗
Alexandre
Alexandre Lombard
Lombard ∗∗Florent Perronnet ∗∗ ∗
∗ Florent Perronnet
Abdeljalil
Alexandre Abbas-Turki
Lombard Abdellah
∗Florent El Moudni
Perronnet ∗∗
∗
Abdeljalil
Abdeljalil Abbas-Turki
Abbas-Turki ∗ Abdellah El Moudni ∗
∗ Abdellah El Moudni ∗
Abdeljalil Abbas-Turki Abdellah El Moudni
∗
∗ IRTES-SeT, Université de Technologie de Belfort-Montbéliard
∗ IRTES-SeT, Université de Technologie de Belfort-Montbéliard
∗ IRTES-SeT, UniversitéBelfort, de
IRTES-SeT, UniversitéBelfort, de Technologie
France de
Technologie Belfort-Montbéliard
France de Belfort-Montbéliard
∗∗ Belfort,
∗∗ Voxelia,
Belfort, France
Belfort,
France France
∗∗ Voxelia, Belfort, France
∗∗ Voxelia, Belfort, France
Voxelia, Belfort, France
Abstract: The emerging topic of cooperative intersection management for vehicles has raised
Abstract:
Abstract: The
The emerging topic of cooperative intersection management for vehicles has raised
up new solutions
Abstract:
up new The emerging
solutions
for traffictopic
emerging
for traffic
of
of cooperative
control
topic
control
in order tointersection
cooperative
in order to
avoid collisions,
intersection
avoid
management
deadlocks,
management
collisions, deadlocks,
for
forand vehicles
and improve
vehicles
improve
has
has raised
traffic
raised
traffic
up
up new
efficiency. solutions
new solutionsThe for
solutions traffic
for traffic control
developed
control for in
for order
road
in order to avoid
traffic
to avoid can collisions,
easily be deadlocks,
applied to and improve
automated traffic
guided
efficiency.
efficiency. The solutions
solutions developed road traffic
traffic cancollisions,
easily be bedeadlocks,
applied to toand improve guided
automated traffic
vehicles toThe
efficiency. overcome
The solutions developed
the common for
for road
drawbacks,
developeddrawbacks, road traffic can
can easily
and improve easilythe be applied
number
applied to automated
of vehicles
automated guided
in a network.
guided
vehicles
vehicles to
to overcome the common and improve the number of vehicles in a network.
In the context
vehicles
In the to overcome
context of a
the
the common
of a network
overcome network common
of
drawbacks,
of vehicles
drawbacks,
vehicles only
and
and improve
only regulated
regulated improve
at
the
the number
at intersections,
number we
intersections,
of
of vehicles
we propose in
vehicles
propose
anaaalgorithm
in
an
network.
network.
algorithm
In
in
In the
order
the context
to
context of
prevent
of a
a network
deadlock
network of
of vehicles
at
vehicles only
intersections
only regulated
in a
regulatednetworkat
at intersections,
of automated
intersections, we
we propose
guided
propose an
vehicles.
an algorithm
algorithm
in
in order to prevent deadlock at intersections in aa network of automated guided vehicles.
in order
order to to prevent
prevent deadlock
deadlock at at intersections
intersections in in a network
network of of automated
automated guided guided vehicles.
vehicles.
© 2016, IFAC (International Federation of Automatic Control) Hosting by Elsevier Ltd. All rights reserved.
Keywords: automated guided vehicles; cooperative intersection management; wireless
Keywords:
Keywords: automated automated guided
guided vehicles;
vehicles; cooperative
cooperative intersection
intersection management;
management; wireless wireless
communication;
Keywords: automated deadlock guided vehicles; cooperative intersection management; wireless
communication;
communication; deadlock
deadlock
communication; deadlock
1. INTRODUCTION server in Dresner and Stone (2004) and de La Fortelle
1.
1. INTRODUCTION
INTRODUCTION server
server in Dresner
inOther
Dresner and
and Stone
Stone (2004)
(2004) and
and de de La Fortelle
LaAdaptive
Fortelle
1. INTRODUCTION (2010).in
server Dresner works and areStonebased(2004)on Cooperative
and de La Fortelle
(2010).
(2010). Other
Other works
works are
are based
based on
on Cooperative
Cooperative Adaptive
Adaptive
In a network with a large number of automated guided (2010). Cruise Control
Other works at Intersections
are based on (CACCI).
Cooperative The server
Adaptive de-
In a network with a large number of automated guided Cruise
Cruise Control
Control at
at Intersections
Intersections (CACCI).
(CACCI). The
The server
server de-
In tects
Cruise conflicts
Control and
at accordingly
Intersections sends
(CACCI).acceleration
The and
server de-
aa network
vehicles
In
vehicles
network
(AGV),
(AGV),
with
with aa large
traffic
traffic
large number
control
number
control is
of automated
is crucial
of automated
crucial to
guided tects conflicts and accordingly sends acceleration and de-
to the system
guided tects conflicts and accordingly
accordingly sends acceleration and de- de-
vehicles
performance.
vehicles (AGV),
(AGV), traffic
Thetraffic control
trafficcontrolcontrolis is must
crucialavoid
crucial to the
to the
the
system
system tects
collisions,
system
celeration
conflicts
celeration
messages
and
messages
to vehicles sends
to vehicles
in order
in order
to avoid collisions
acceleration
to avoid and
collisions
performance. The traffic control must avoid collisions, celeration
in Zohdy messages
and Rakha to vehicles
(2012) and in order
Zohdy to
et avoid
al. collisions
(2012). Fur-
performance.
deadlocks/gridlocks The traffic
traffic
and ensure control thatmusteveryavoid can reach celeration
AGV collisions, messages to vehicles in order et to avoid collisions
performance.
deadlocks/gridlocks
deadlocks/gridlocks
The and ensure
and
control
ensure thatmust
that everyavoid
every AGV collisions,
AGV reach in
can reach
can in Zohdy
Zohdy aand
thermore
in Zohdy and
and
Rakha
Rakha (2012)
Sequence-Based
Rakha (2012)
and
and Zohdy
(2012) Protocol
and Zohdy
Zohdy et al.
(SBP)
et al.is (2012).
al. (2012).
proposed
(2012).
Fur-
Fur-
in
Fur-
its destination,
deadlocks/gridlocks and for
and this purpose
ensure that different
every AGV solutions
can are
reach thermore a Sequence-Based Protocol (SBP) is proposed in
its
its destination,
destination, and
and for
for this
this purpose
purpose different
different solutions
solutions are
are thermore
Perronnet
thermore aet
a Sequence-Based
al. (2013).
Sequence-Based It Protocol that
assumes
Protocol (SBP)
(SBP) theis proposed
is proposed
intersection in
in
proposed:
destination, and for this purpose different solutions are Perronnet et al. (2013).
its
proposed:
proposed: Perronnet
is controlled
Perronnet et as
et al. (2013).: It
al. follows
(2013). It
assumes
Iteither
assumes
assumes
that
that the
the intersection
that the intersection
the intersection
manager
intersection
proposed:
• To avoid collisions, zone control with wireless trans- is is
is controlled
controlled as
as follows
follows :
: either
either the
the intersection
intersection manager
manager
•• To avoid or a decentralized
controlled as negotiation
follows : either explicitly
the determines
intersection manager the
• To To avoidiscollisions,
mission
avoid collisions,
the favorite
collisions,
zone
zone
zone
control
control
traffic control
control
with
withofwireless
with wireless
most environ-
wireless
trans-
trans- or
trans- or aa decentralized
sequence decentralized
of vehicles
negotiation
negotiation
in each
explicitly
explicitly
conflict area.
determines
determines
The sequence
the
the
mission
mission is
isitthe
the favorite traffic control
favoritetotraffic control of most environ- or
of most a decentralized
sequence of vehicles negotiation
in
in each
explicitly
conflict area. determines
The sequence the
ments asis
mission
ments as it
is simple
the
is favorite
simple to
install
traffic
install and of most environ-
and expand.
control expand. environ- sequence sequence
determinesof
determines
of which
vehicles
vehicles
which
vehicle
in
vehicle
each
each is
conflict
is conflict
the first,
area.
the first,area. which
which
Theone
The sequence
is the
sequence
one is the
• ments
To avoid as deadlocks/gridlocks
it is
is simple
simple to to install
installinand
and expand.
the expand.
network of AGV, determines determines
second and so which vehicle is
on. vehicle is the
the first,
first, which
which one one is is the
the
•• ments
To
To avoid
avoid
as it
deadlocks/gridlocks
deadlocks/gridlocks in
in the
the network
network of
of AGV,
AGV, second
second and
and
which
so
so on.
on.
• To the mainly
avoid used solutions
deadlocks/gridlocks are improved
in the network banker’s
of AGV, al-
the
the mainly
mainly used
used solutions
solutions are
are improved
improved banker’s
banker’s al- second
al- Thoughand theseso protocols
on. are designed for isolated intersec-
gorithms (Lawley et al., 1998; Ezpeleta et al., 2002;al- Though these protocols are
the mainly
gorithms used
(Lawley solutions
et al., are
1998; improved
Ezpeleta banker’s
et al., 2002; Though
tion, they these
can protocols
be integratedare designed
designed
in a more
for isolated
forglobal
isolated intersec-
intersec-
solution in
gorithms
Bobanac (Lawley
gorithms (Lawley
and Bogdan, et
et al.,
al., 1998;
2008),
1998; Ezpeleta
and
Ezpeletathe useet
et al., Petri tion, they can be integrateddesigned
2002;
of 2002;
al., Though these protocols are in a more for isolated
global intersec-
solution in
Bobanac
Bobanac and
and Bogdan,
Bogdan, 2008),
2008), and
and the
the use
use of
of Petri
Petri tion,
charge
tion, they
theyof can
managing
can be
be integrated
(routing
integrated in
inanda
a more
gridlock
more global
global solution
prevention)
solution ina
in
nets (Wu and Zhou, 2007), but they both suffer
Petri charge of
Bobanac
nets
nets (Wu
(Wu
andand Bogdan, 2007),
and Zhou,Zhou,
2008), and they
2007), but but
the use of suffer
they bothboth suffer chargecharge of managing
fleet of vehicles,
of managing
managing
(routing
(routing and
like Bocewicz
(routing and gridlock
gridlock
et al.gridlock
and (2007) and
prevention)
prevention)
Perronnetaa
prevention)
a
from (Wu
nets drawbacks.and The banker’s
Zhou, 2007), algorithm
but they can elimi-
both suffer fleet of vehicles, like Bocewicz et al. (2007) and Perronnet
from drawbacks. The banker’s algorithm can elimi- fleet of vehicles,
et al.of(2014). Forlike Bocewicz
likeinstance, et al. (2007)
inetPerronnet and
et Perronnet
al.Perronnet
(2014) a
from drawbacks.
nate valid solutions, Thesignificantly
banker’s algorithm
algorithm
compromising can elimi- the fleet
elimi- et al. vehicles, Bocewiczin al. (2007)et and
from
nate
nate
drawbacks.
valid
valid solutions,
solutions,
The banker’s
significantly
significantly compromising
compromising
can the
the et al. (2014).
protocol
et al. (2014).
is
(2014).
For
For instance,
proposed
For instance,
to
instance, avoid in
Perronnet
in gridlock
Perronnet
Perronnet et al.
(deadlock
et al. (2014)
al. (2014)
caused
(2014)
aa
a
efficiency
nate valid ofsolutions,
the management, significantly while Petri nets can
compromising the protocol is proposed to avoid gridlock (deadlock caused
efficiency
efficiency of
of the
the management,
management, while
while Petri
Petri nets
nets can
can protocol is proposed
by the interaction
protocol is proposed to avoid
avoid intersections)
of multiple
to gridlock (deadlock
gridlock (deadlock caused
in a network
caused
cause livelocks. Moreover the number of AGV in the
can by the
efficiency
cause
cause
of theMoreover
livelocks.
livelocks.
management,
Moreover the
the number
while Petri
number of
of AGV
AGV
netsin
in the
the by by the interaction
interaction
of intersections
the interaction
of
of multiple
allowing
of multiple intersections)
intersections)
more intersections)
multiple than one vehicle
in aa network
in network
per
in a per zone
network
network
cause with these
livelocks. Moreover solutions
the is limited:
number of a resource
AGV in the of intersections allowing more than one vehicle zone
network
network with
with these
these solutions
solutions is
is limited:
limited: a
a resource
resource of
of intersections
between
intersections allowing
intersections.
allowing more
This
more than
algorithm
than one
one isvehicle
based
vehicle per
peron zone
the
zone
is an arcwith
network between thesetwo two nodes and
solutions can only
is limited: be held between intersections. This algorithm is based on the
a resource
is
is an
an arc
arc between
between two nodes
nodes and
and can
can only
only be
be held
held between
principle
between intersections.
of path
intersections. This algorithm
reservation:
This algorithm
a vehicle ishas
is based
based to on the
follow
on thea
by an
is onearcAGV betweenat a two time.nodes Otherand solutions
can only relying
be held on principle of path reservation: a vehicle has to follow a
by
by one
one AGV at
at a
AGV deadlock a time.
time. Other
Other solutions
solutions relying
relying on principle
given path,
on principle of itpath
of path reservation:
asks reservation:
the global server aa vehicle
vehicle
for the hasauthorization
has to follow
to follow aa
siphon-based
by one AGV at a time.prevention
Other are optimal
solutions but the
relying on given path, it asks the global server for the authorization
siphon-based deadlock prevention are optimal but the given path, it asks the global server for the authorization
siphon-based
minimal siphon
siphon-based deadlock
is a NP
deadlock prevention
problem,are
prevention are optimal
thus it cannot
optimal but the
but be to follow this path, then the global server reserves the
the to
givenfollow
path, this
it path,
asks thethen
global the global
server forserver
the reserves
authorization the
minimal
minimal siphon
siphon is
is aa NP
NP problem,
problem, thus
thus it
it cannot
cannot be to
be to
path follow
follow this
of thisthis path,
vehicle and
path,and then
informs
then the
the global server
intersection
global reserves
servers.
serverservers.
reserves the
Then,
the
considered
minimal for real
siphon is atime
NP applications.
problem, thus it cannot be path of this vehicle informs intersection Then,
considered
considered for real
forrouting
real time
time applications.
applications. path
eventual
path of this
of this vehicle and
intersections
vehicle andofinforms
informs
vehicles intersection servers.
are locallyservers.
intersection managed Then,
Then,by
• considered
Conflict-freefor real time based solutions
applications. are also pro- eventual intersections of vehicles are locally managed by
•• Conflict-free routing based solutions are also pro- eventual
one of the intersections
previously of vehicles
presented are locally
solution with managed
respect by
to
Conflict-free
posed in Nishirouting
• Conflict-free routing
and Tanaka based(2012)
based solutions are also
and Miyamoto
solutions are and eventual
also pro-
pro- one of the intersections
previously of vehicles solution
presented are locally with managed
respect by
to
posed
posed in
in Nishi
Nishi theyand
and TanakaTanaka
Tanaka (2012)
(2012) and Miyamoto
and Miyamoto
Miyamoto and one
the
one of
of the
constraints
the previously
stated
previously presented
by
and the constraints stated by the global protocol. the
presented global solution
protocol.
solution with
with respect
respect to
to
Inoue in
posed (2016),
Nishi and prevent(2012)the intersection
and problemand
Inoue
Inoue (2016),
(2016), they
they prevent
prevent the
the intersection
intersection problem
problem the
the constraints
constraints stated
stated by
by the
the global
global protocol.
protocol.
but they
Inoue require
(2016), they a centralized
prevent the management,
intersectionand some These solutions are originally designed for road vehicles
problem
but they require aa centralized management,
management, and and some
some These solutions are
but
but they
validthey requireare
solutions
require a centralized
eliminated.management,
centralized and some These solutions
but cansolutions
These be applied aretooriginally
are originally
AGV in order
originally
designed
designed for
for road
to increase
designed for road vehicles
vehicles
the vehicles
road number
valid
valid solutions
solutions are
are eliminated.
eliminated. but can be applied to AGV in order to increase the number
valid solutions are eliminated. but
of
butAGV can
can be
in
be aapplied
network,
applied to
to AGV
and
AGV in
improve
in order
order to
the
to increase
overall
increase the number
performance.
the number
Due to the recent development of cooperative intersec- of AGV in a network, and improve the overall performance.
Due to
to the
the recent development of
of cooperative intersec- of AGV in
Therefore, in aainnetwork,
network,
the context and improve
improve the overall
of this paper, overall performance.
we assume a net-
Due
tion management
Due to the recent for
recent development
road vehicles,
development of cooperative
new solutions
cooperative have of
intersec-
intersec- AGV
Therefore, in the and
context of this the
paper, we performance.
assume aa net-
tion
tion management
management for road
for traffic vehicles,
road vehicles,
vehicles, new
new solutions
solutions have Therefore,
work
have Therefore, with in
a the
traffic context
in the contextcontrol of this
protocol
ofprotocol paper,
this paper, in we
charge assume
we assume of the net-
grid-
agrid-
net-
emerged to improve efficiency with both colli- work with
tion
emerged
emerged
management
to
to improve
improve
for road
traffic
traffic efficiency
efficiency
new with
with
solutions
both
both
have
colli-
colli- work
lock
work with aaa traffic
prevention
with traffic control
controlmore
allowing
traffic control protocol
than
protocol
in
inone
in
charge
charge
vehicle
charge
of
of the
of the grid-
between
the grid-
sion and to
emerged deadlock
improveavoidance. One can with
traffic efficiency quoteboth the colli-
well- lock prevention allowing more than one vehicle between
sion and deadlock
deadlock avoidance. One can
can quote the well- lock prevention allowing
two intersections. We thenmore more thananone
consider one vehicle between
intersection between
in this
sion
known
sion and
andReservation-Based
deadlock avoidance.
avoidance. ProtocolOne (RBP)
One can quote
wherethe
quote the ve- lock
well-
thewell- two prevention
intersections. allowing
We then than
consider an vehicle
intersection in this
known
known Reservation-Based
Reservation-Based Protocol
Protocol (RBP)
(RBP) where
where the
the ve-
ve- two
network
two intersections.
with
intersections. We
multiple
We then
unlocked
then consider
consider an
vehicles,
an intersection
i.e. vehicles
intersection in that
in this
this
hicle
known sends a reservation
Reservation-Based request
Protocol of space
(RBP) and time
where to
the the
ve- network with multiple unlocked vehicles, i.e. vehicles that
hicle sends
hicle sends a reservation
sends aa reservation
reservation request request
request of of space
of space
space and and time
and time
time to to the network with multiple unlocked
the network with multiple unlocked vehicles, i.e. vehicles that
to the vehicles, i.e. vehicles that
hicle
Copyright 2016 IFAC
2405-8963 © 2016, 497 Hosting by Elsevier Ltd. All rights reserved.
IFAC (International Federation of Automatic Control)
Copyright
Copyright © 2016 IFAC 497
Peer review©
Copyright 2016
©under IFAC
2016 responsibility
IFAC 497
of International Federation of Automatic
497Control.
10.1016/j.ifacol.2016.07.669
IFAC MIM 2016
498
June 28-30, 2016. Troyes, France Alexandre Lombard et al. / IFAC-PapersOnLine 49-12 (2016) 497–502

can move without causing a gridlock. In order to avoid • Stop line: the nearest border of the box junction. For
collisions between AGV, we assume that the intersection safety reason, the AGV has to stop by default before
is regulated with an intersection server using a SBP and a the stop line (default-deny principle).
first-in, first-out scheduling policy.
According to the chosen protocol, the inconsistency of the
presence list brings different risks. Deadlock and collision
are then possibles. In the following we will consider a
centralized architecture of SBP (C-SBP) named Transpar-
ent Intersection Manager (TIM) (Perronnet et al., 2013).
In TIM vehicles synchronize their speeds according to
the presence list received from the server. There are two
advantages of C-SBP. The first one is the default-deny: if
a vehicle is not able to establish a communication with the
server, it has to brake before the box junction. The second
one is that the results of C-SBP can easily be extended to
RBP.
However, due to potential communication problems (mes-
sage losses), the sequence built according to the order of
arrival of messages from the AGV can be different from the
physical order of the vehicles, resulting in a deadlock situa-
tion. The scope of this paper is to propose a re-sequencing
algorithm able to avoid the deadlock without introducing
any risk of collision (3), even with an unreliable com-
munication. In order to assess the proposed algorithm,
simulations and intersections of robots are performed.
This paper is organized as follows; first it presents the
protocol TIM and the conditions of the problem. Then,
the paper introduces the deadlock problem as well as the Fig. 1. TIM
collision risk due to a bad re-sequencing. Therefore, it
presents the re-sequencing algorithm and shows that the
resulting sequence is collision-free. Before concluding, the The three obstacles are presented in 1. For each obstacle,
paper discuss the results of simulations as well as the the AGV computes an acceleration. Hence, there are three
results of the cooperative intersection of robots. computed accelerations:

2. PROBLEM STATEMENT • ar to keep a safe distance from the visible obstacles.

2.1 Brief presentation of the Transparent Intersection
Manager
Vehicles in TIM are able to observe the obstacles in
the surrounding environment and to adapt their speed
accordingly. AGV move autonomously if all obstacles are
visible by the sensors. In order to enhance the intersection
safety and throughput, a negotiation protocol based on
wireless communication to synchronize the AGVs speed
is proposed by TIM. Every AGV has to use wireless
communication to inform an intersection server about
its movement. More precisely, every AGV communicates
its current position and speed as well as the desired
destination and the remaining distance before the exit
of the box junction. Accordingly, the intersection server
broadcasts these data through an ordered presence list to
all the AGV closed to the intersection. In the received
presence list, the first vehicle has the highest priority and
so on. Each AGV considers three kinds of obstacles:
• Visible obstacles: mainly a precedent AGV in the
same buffer lane. Visible obstacles are detected by
the sensors without any message received from the
server.
• Virtual obstacles: conflicting AGV with higher prior-
ity.
• ai to keep a safe distance from priority conflicting
• as to allow a vehicle to stop in the case of a dangerous
situation.
The three accelerations contribute to determine the accel-
eration of the AGV as follows :
• ai is determined as the minimum acceleration from
all conflicting vehicles with a higher priority.
• If the AGV has not received a presence list it has
to stop before the box junction. In other words,
the intersection map is known before requesting the
presence list.
• Vehicles are not allowed to overtake in the buffer zone
of the intersection.
• The speed synchronization is done near the junction
box.
Formally, the acceleration of each AGV is computed as
a = min(ar , max(ai , as ))
We draw the reader attention to the fact that only ai
depends on the presence list sent by the server. The
vehicle must know the intersection map before getting into
the buffer zone. We highlight also the fact that, default
deny (as ) is used when the vehicle has not yet received a
presence list or when it is not in the presence list. It is also
used when the vehicle is not able to keep a safe distance
from the virtual obstacles.

498
 IFAC MIM 2016
June 28-30, 2016. Troyes, France Alexandre Lombard et al. / IFAC-PapersOnLine 49-12 (2016) 497–502 499

2.2 Considered assumptions

TIM allows several strategies for synchronizing the speed

of vehicles as well as many policies for ordering the
presence list. In the reminder we consider the following
assumptions:
• The error on the position is below one meter.
• With respect to the positioning accuracy, the posi-
tions received by the manager can refer to old posi-
tions of the vehicles, but never to a position that the
vehicle has not reached yet. In other words, more than
one meter error is due communication delays between
the AGV and the intersection server.
• ai is computed under the constraint that two conflict-
ing vehicles cannot simultaneously access to the box
junction.
• When t → ∞, all present vehicles will be discovered
at the good position.
• The presence list is built according to ”first-come,
first-served” (FCFS).

2.3 Loss of packets

Using wireless communication, latency and loss of packets
are common problems. And as the number of communi-
cating AGV increases in a network, they become a major
preoccupation. Therefore the order of arrival of vehicles at
the intersection can differ from the order of arrival of the
messages notifying their arrival.
The default-deny allows better safety conditions with the
intersection. Indeed, if an AGV is not able to communicate
with other AGV through the intersection server, it has
to stop before the box junction. As long as an AGV is
not in the presence list sent by the server, it has to stop.
However, in such a case the default deny can block the
traffic (Lombard et al., 2015). More precisely, if one vehicle
stops then all vehicles behind it in the same buffer zone
will stop too. Furthermore all conflicting AGV with a
lower priority will stop. The main raised issue is how to
safely recover the traffic movement when a missed vehicle
establishes a connection.
As said before the focus is put on the FCFS policy. With
this policy the first AGV arrived at the buffer zone will be
the first allowed to cross the conflict area according to the
presence list. When considering an instantaneous wireless
communication, the order of arrival of vehicles is defined
by the order of arrival of the messages emitted by the AGV
toward the intersection server.
Let us consider the example presented in 2. AGV are num-
bered according to the messages received by the server.
One can note that 10 and 11 were initially missed. Because
vehicle 10 is missed, only vehicles 1 and 4 are able to cross
the intersection. If the intersection server and the AGV
operate under normal conditions, the server will gradually
discover the missed vehicles. When these vehicles will be
discovered, the server will have to include them in the
presence list.
If vehicles are ordered according to the received message
there is a deadlock. Indeed, there is a circular wait: 10
 9  7  (6,5)  (3,4)  2  10 (i  j means that i
is waiting for j to cross the intersection). Only AGV #1
Fig. 2. Two missed vehicles: The number associated to
vehicles shows the order of the received messages by
intersection server.
and #4 are able to cross the intersection. If the AGV are
ordered according to their arrival time, a collision may
happen. Indeed, AGV #4 discovers a new virtual obstacle
(#10) that has not been initially considered for keeping a
safe distance. Moreover, if we consider that the AGV #11
has arrived before AGV #7, the latter has not adapted its
speed to avoid collision with #11. Hence, the main raised
issue is to reconsider the presence list with missed vehicles
without deadlock and collision risks.
3. APPROACH FOR SAFE COOPERATIVE
INTERSECTION MANAGEMENT
To achieve the objective, in the following we will provide
an algorithm that first prevents deadlock and then avoids
collision.
3.1 Collision free
First let us draw the reader attention to the fact that when
a new vehicle is discovered, if it is not ordered correctly a
collision may happen, whether or not a cycle is discovered.
Indeed, let us consider the example given in 2. If only AGV
#1, #2, #3 and #4 are present and if we discover vehicle
#10, there is no circular wait, but there is a risk of collision
between #10 and #4. This is also the case of vehicle 11 for
which there is no cycle. In Qian et al. (2014), the authors
introduced and proved the collision free condition. There
is a collision risk if a vehicle discovers a new obstacle.
From this condition we address a conflict vector of the
new discovered vehicle with all present vehicles and we
proceed as follows:
• The new discovered vehicle is ordered after the last
conflicting vehicle.
• Each AGV in the same lane physically behind the
new discovered AGV move back in the list until it is
behind the vehicle physically in front.

499
IFAC MIM 2016
500
June 28-30, 2016. Troyes, France Alexandre Lombard et al. / IFAC-PapersOnLine 49-12 (2016) 497–502

function sizeof(s)  size of the collection s

function l(v)  lane of the vehicle v

function d(v)  distance of the vehicle v from the intersection

function s(v)  position of the vehicle v in the sequence

function conflict(v1, v2)  true if v1 and v2 are in conflict,

false otherwise

procedure move(s, i, j)  move the element at position i in the

sequence s to position j

procedure insert(s, i, o)  insert the object o at position i in s

procedure InsertionSequence(v)
expectedP osition ← sizeof (sequence)
for all v  in sequence do
if l(v) = l(v  ) and d(v) < d(v  ) then
expectedP osition ← min(expectedP osition, s(v  )
end if
end for
insertionP osition ← expectedP osition
if expectedP osition = sizeof (sequence) then Fig. 4. Graph representation of the intersection: a- before
for i from expectedP osition to sizeof (sequence) − 1 do receiving message from vehicle 10, b- after receiving
if conf lict(v, sequence[i]) then message from vehicle 10
insertionP osition ← i
end if
break the cycle by removing edges. However, only virtual
end for
moved, i ← 0
obstacles can be reconsidered.
while i < insertionP osition − moved do Let G = (N, E) be a directed graph of dependency in
if l(v) = l(sequence[i] then
which the set of nodes N = 1, 2, ..., N ∪ S represents the
move(sequence, i, insertionPosition)
moved ← moved + 1
vehicles with the stop constraint and the set of directed
else edges E connects each node v to node v  if the vehicle v
i←i+1 has to adapt its speed according to the obstacle v  . Hence,
end if E = Er ∪ Ei ∪ Es according to the minimal acceleration
end while used to cross the intersection. 4-a and 4- b show the graph
insertionP osition ← insertionP osition − moved before and after discovering vehicle 10, respectively. For
end if the sake of readability, edges of real obstacles are removed
insert(sequence, insertionPosition, v) in 4 if the order is respected by edges of virtual obstacles.
end procedure
One can note that in both situations there is no collision
Fig. 3. Collision and deadlock free algorithm (CDF algo- risk because vehicles are stopped, waiting for vehicle 10.
rithm) In 4-b there are four cycles given in 5. The cycles can be
broken by removing two edges of virtual constraints. There
The optimized algorithm called CDF is given in 3. The are four possible solutions:
algorithm is collision free since there is no vehicle which • 10  9 and 10  8,
will discover new obstacle. Indeed, first, the conflicting • 9  7 and 8  7,
vehicles with the new discovered vehicle will all pass • 7  5 and 7  6,
before. Second, the vehicles behind can only loose priority. • 5  3 and 6  3.
It remains to show that it is deadlock free.
From the CDF algorithm presented in 3, edges 5  3 and
6  3 are removed. The new sequence when vehicle 10 is
3.2 Recovering liveness discovered is 1 ≺ 4 ≺ 5 ≺ 6 ≺ 7 ≺ 8 ≺ 9 ≺ 10 ≺ 2 ≺ 3.
When vehicle 11 is discovered, the new sequence will be 1
There are four conditions to have a deadlock situation ≺ 4 ≺ 5 ≺ 6 ≺ 7 ≺ 9 ≺ 10 ≺ 2 ≺ 3 ≺ 11 ≺ 8
Coffman et al. (1971). If they are satisfied simultaneously,
a deadlock takes place:
• Mutual exclusion
• Hold and wait
• No preemption
• Circular wait
The three first conditions are inherent to the traffic. Hence,
when there is a circular wait, a deadlock happens. The
circular wait can be represented as a directed cycle in
the graph of dependency. To recover liveness, we need to Fig. 5. The four circular wait

500
 IFAC MIM 2016
June 28-30, 2016. Troyes, France Alexandre Lombard et al. / IFAC-PapersOnLine 49-12 (2016) 497–502 501

It can be shown that there is a circular wait if the two observe a deadlock situation after 880 seconds for a loss
following conditions are satisfied: rate of 70%, 380 seconds for a loss rate of 95%, and 38
seconds for a loss rate of 90%. Even if is not represented
• The set S of vehicles physically behind the new
here, we have observed that a smaller amount of lost
discovered of vehicle in the same lane is nonempty
packets can also lead to a deadlock situation after enough
• There is at least one vehicle in S which precedes a
time. These statements confirm our assumption that, even
conflicting vehicle with the new discovered vehicle
if in real conditions the risk is low, deadlock cannot be
Then, since all vehicles physically behind the new discov- ignored.
ered vehicle are ordered after, the link that exists with
a conflicting vehicle is removed. Hence the cycle is inter-
rupted.

4. TESTS AND RESULTS

4.1 Simulation framework

Simulations have been realized to experiment this solution.

They were realized in Java with a simple yet realistic AGV
model, introducing inaccuracy and delays in the position-
ing system, in the front radar measure, and delays in the
application of the acceleration and direction command. In
order to test our deadlock solution, AGV are placed on an
8-shaped circuit where the crossing point is the intersec-
tion. The vehicles use the TIM protocol (Perronnet et al.,
2013) in order to get the right-of-way at the intersection.

Fig. 7. Evacuation of vehicles without deadlock solution

Fig. 6. Illustration of the simulation circuit

A communication rate has been fixed to 0.5s and different
loss rates have been tested, in order to simulate the issues
of a wireless communication. To represent the loss rate,
when a message is sent (either by the vehicle or the
intersection manager), it has a fixed probability to be
ignored. The simulations have been realized using the loss
rates 5%, 10%, 50%, 70%, 90% and 95%, with and without
the deadlock prevention.
The reader may note that these loss rates do not reproduce Fig. 8. Evacuation of vehicles with deadlock prevention
accurately the loss rates of a V2I communication, but are
used to stress-test our solution in critical conditions (if As it can be seen on 8, with deadlock prevention vehicles
you can do the big things, you can do the little things were sometime stopped at the intersection due to the loss
as well). Moreover the presentation of the number of of messages (if they do not receive the right-of-way from
vehicles evacuated is used to show the liveness or the the manager, their default behavior is to stop), but they
deadlock situation of the intersection, but is not used here always ended up moving out of the intersection. More-
to evaluate the throughput of the intersection using TIM over, no collisions were detected: a hazardous situation is
as it has already been studied in Perronnet et al. (2013). counted every time the distance between two vehicles is
below one meter, this did not happen in our simulation.
4.2 Results
4.3 Tests with EV3 robots
The following figures (7 and 8) represents the number of
vehicles crossing the intersection of the circuit in function Finally, to confirm the feasibility of the solution, and its
of the time for different loss rates. These measures have collision-free property, real tests have been made using
been made with six vehicles on the circuit. On 7, we can Lego EV3 robots. Our test bench is presented in 9.

501
IFAC MIM 2016
502
June 28-30, 2016. Troyes, France Alexandre Lombard et al. / IFAC-PapersOnLine 49-12 (2016) 497–502
Fig. 9. EV3 intersection test bench
Positioning of the robots was made using the odometry, the
color sensor and the color marks on the ground, the front
obstacle detection using the EV3 sonar and the network
using Edimax Wi-Fi dongles. The intersection manager
was an Android phone acting as a wireless access point for
the robots. As expected, during first tests, nor collision
neither deadlocks were observed, but the quality of the
wireless connection was good (only five devices were using
Wi-Fi and they were close to each other). A fake network
disturbance has been introduced similar to the one used in
simulations in order to stress-test our protocol. It consisted
of a fixed probability that the intersection manager ignores
a received message. It slowed down the performance of
the intersection but collisions and deadlocks were not
introduced.
5. CONCLUSION
During our work to extend the tests on cooperative inter-
section management to three vehicles and more, we have
faced deadlock situations. The purpose of this paper has
been twofold. In a first part, it has been shown that the
impact of vehicular communication issues (message delays
and losses) can lead to a specific case of deadlock resulting
of the interdependency between the right-of-way and the
position of a vehicle in its lane. In a second part, a solution
to prevent deadlocks has been presented, and it has been
shown that the proposed solution will effectively prevent
deadlock and will not provoke collisions.
Finally, simulations and tests have been realized. They
showed up that the presented kind of deadlock cannot be
ignored, especially when the message loss rate increases.
Moreover, it confirmed that the proposed solution prevents
the risk of deadlock and do not provoke any collision.
REFERENCES
Bobanac, V. and Bogdan, S. (2008). Routing and schedul-
ing in multi-agv systems based on dynamic banker algo-
rithm. In Control and Automation, 2008 16th Mediter-
ranean Conference on, 1168–1173. IEEE.
Bocewicz, G., Wójcik, R., and Banaszak, Z. (2007). Design
of admissible schedules for agv systems with constraints:
A logic-algebraic approach. In Agent and Multi-Agent
Systems: Technologies and Applications: First KES In-
ternational Symposium, KES-AMSTA 2007, Wroclaw,
Poland, May 31– June 1, 2007. Proceedings, 578–587.
Coffman, E.G., Elphick, M., and Shoshani, A. (1971).
System deadlocks. ACM Computing Surveys (CSUR),
3(2), 67–78.
de La Fortelle, A. (2010). Analysis of reservation algo-
rithms for cooperative planning at intersections. In
502
View publication stats
Intelligent Transportation Systems (ITSC), 2010 13th
International IEEE Conference on, 445–449.
Dresner, K. and Stone, P. (2004). Multiagent traffic
management: A reservation-based intersection control
mechanism. In Proceedings of the Third International
Joint Conference on Autonomous Agents and Multia-
gent Systems-Volume 2, 530–537. IEEE Computer Soci-
ety.
Ezpeleta, J., Tricas, F., Garcı́a-Vallés, F., and Colom,
J.M. (2002). A banker’s solution for deadlock avoidance
in fms with flexible routing and multiresource states.
Robotics and Automation, IEEE Transactions on, 18(4),
621–625.
Lawley, M., Reveliotis, S., and Ferreira, P. (1998). The
application and evaluation of banker’s algorithm for
deadlock-free buffer space allocation in flexible man-
ufacturing systems. International Journal of Flexible
Manufacturing Systems, 10(1), 73–100.
Lombard, A., Perronnet, F., Abbas-Turki, A., El Moudni,
A., and Bouyekhf, R. (2015). V2x for vehicle speed
synchronization at intersections. In 22nd ITS World
Congress, EU - ITS-2788, 5-9 October 2015, Bordeaux,
France. ITS-WC.
Miyamoto, T. and Inoue, K. (2016). Local and random
searches for dispatch and conflict-free routing problem
of capacitated {AGV} systems. Computers & Industrial
Engineering, 91, 1 – 9.
Nishi, T. and Tanaka, Y. (2012). Petri net decompo-
sition approach for dispatching and conflict-free rout-
ing of bidirectional automated guided vehicle systems.
Systems, Man and Cybernetics, Part A: Systems and
Humans, IEEE Transactions on, 42(5), 1230–1243.
Perronnet, F., Abbas-Turki, A., and El Moudni, A. (2013).
A sequenced-based protocol to manage autonomous ve-
hicles at isolated intersections. In Intelligent Transporta-
tion Systems-(ITSC), 2013 16th International IEEE
Conference on, 1811–1816. IEEE.
Perronnet, F., Abbas-Turki, A., and El Moudni, A. (2014).
Vehicle routing through deadlock-free policy for cooper-
ative traffic control in a network of intersections: Reser-
vation and congestion. In Intelligent Transportation
Systems-(ITSC), 2014 17th International IEEE Confer-
ence on, 2233–2238. IEEE.
Qian, X., Gregoire, J., Moutarde, F., and De La Fortelle,
A. (2014). Priority-based coordination of autonomous
and legacy vehicles at intersection. In Intelligent Trans-
portation Systems (ITSC), 2014 IEEE 17th Interna-
tional Conference on, 1166–1171. IEEE.
Wu, N. and Zhou, M. (2007). Shortest routing of bidirec-
tional automated guided vehicles avoiding deadlock and
blocking. Mechatronics, IEEE/ASME Transactions on,
12(1), 63–72.
Zohdy, I.H., Kamalanathsharma, R.K., and Rakha, H.
(2012). Intersection management for autonomous ve-
hicles using icacc. In Intelligent Transportation Systems
(ITSC), 2012 15th International IEEE Conference on,
1109–1114. IEEE.
Zohdy, I.H. and Rakha, H. (2012). Game theory algorithm
for intersection-based cooperative adaptive cruise con-
trol (cacc) systems. In Intelligent Transportation Sys-
tems (ITSC), 2012 15th International IEEE Conference
on, 1097–1102. IEEE.