Distributed constraint reasoning applied to multi-robot exploration

Arnaud Doniec, Noury Bouraqadi, Michael Defoort

Univ Lille Nord de France, F-59000 Lille, France
EMDouai, IA, F-59508 Douai, France
{doniec,bouraqadi,defoort}@ensm-douai.fr
Van Tuan Le, Serge Stinckwich
GREYC, UMR 6072 (CNRS, Université de Caen)
UMI UMMISCO 209 (IRD/UPMC/IFI-MSI)
[email protected]
[email protected]

Abstract One of the major applications of multi-robot systems is

the exploration of an unknown environment. It consists,
Exploration of an unknown environment is one of the ma- for a fleet of robots, in navigating while incrementally con-
jor applications of Multi-Robot Systems. Many works have structing a map of the environment. In this application, the
proposed multi-robot coordination algorithms to accom- robots have to collaborate to:
plish exploration missions based on multi-agent techniques.
• spread out on the ground (and consequently: speed up
Some of these works focus on multi-robot exploration under
the exploration and use less energy),
communication constraints. In this paper, we propose an
original way to formalize and solve this issue. Our pro- • keep in touch with each other in order to exchange par-
posal relies on distributed constraint satisfaction problems tial maps and share areas that have not been explored
(disCSP) which are an extension of classical constraint sat- yet.
isfaction problems (CSP). Compared to other works, our
proposal is fully distributed and guaranties the exploration Therefore, a good collaboration scheme for multi-robots ex-
of an unknown environment with maintenance of connectiv- ploration has to conciliate these two antagonist constraints.
ity between all the members of a robots’ team.
2 Multi-robot exploration related works

The multi-robot exploration issue has been addressed in

1 Introduction the literature using different approaches and has been orig-
inally initiated by: [9] and [15].
Multi-robot systems (MRS) consist in a set of au- In [15], robots try to build and use a global map for the
tonomous mobile robots which collaborate to perform a exploration. The postulate of this work is the following: to
mission. This collaboration is allowed by communica- speed up the exploration, robots have to gain new informa-
tion abilities which usually rely on radio communication tion about the environment. Therefore, they have to move
technologies. For example, Mobile Ad Hoc Networks towards the boundary between open space and unexplored
(MANET) are frequently employed to support explicit com- area. In practice, some robots try to get closer to the bound-
munications in MRS: each robot becomes a node of the net- ary but without real cooperation with the rest of the fleet.
work and is able to send, receive and relay data to other Nevertheless, this article states the concept of frontier-based
robots. In order to have an accurate and efficient collabora- exploration which has inspired many other works.
tion between robots, each node (robot) of the network has to Main improvings of the frontier-based exploration have
be reachable at each instant. This implies that, in addition to consisted in introducing coordination between robots. This
their collaborative task, the robots have to perform an extra has firstly been performed in a centralized way. In [12],
task which consists in maintaining the network connectiv- the authors propose to assign a target destination to each
ity. robot so that the expected information gain over time will be
maximized. Since the computation of the optimal solution same subnetwork. At the end of a constant time, the robot
is intractable in practice, they proposed an heuristic based who provides the best bid is declared as winner and is al-
on bids construction. Each robot estimates the utility and lowed to move towards the target. This process restarts for
the cost to travel towards various locations. A central server all the remaining robots.
receives all the bids and assigns a location target to each The use of distributed bidding algorithms does not guar-
robot taking possible overlaps in the coverage of the ground anty the maintenance of connectivity. Indeed, the introduc-
into account. tion of the nearness measure in the bid calculation is just
More recently, Rooker and Birk [8] also proposed a cen- an heuristic which tends to make robots stay close to each
tralised coordination ensuring that, during the exploration, other. Robots exhibit certain clustering behaviors during the
no robot will loose the connection with the rest of the fleet. exploration but deconnections can occur and lead to the par-
To achieve this goal, a central entity collects the current po- tition of the network. The authors explicitely mention this
sitions of all the robots and generates a set of future config- problem and talk about inconsistency between maps of the
urations for the fleet (i.e. the future possible positions of the disjoint subnetworks.
robots). Due to the high number of available combinations, The aim of this article is to provide a fully distributed
all configurations can not be considered but only a limited algorithm for exploration with maintenance of connectivity.
number of them. Among this number of generated config- In the following, we present an original way to formalize
urations, the central entity chooses the best one according and solve this issue considering distributed constraint satis-
to an utility function. This function gives penality when faction problems.
the evaluated future position is occupied by an obstacle or
when it puts a robot out of the communication range with 3 DisCSP for multi-robot coordination
the other robots.
Clearly, centralized coordination algorithms dedicated to 3.1 Distributed Constraint Satisfaction Problems
frontier-based exploration seem difficult to employ in real (disCSP)
applications. They are not fault tolerant since the entire sys-
tem will fail whether the central entity fails. Moreover, for Constraint satisfaction is a classical and powerfull tool
large scale MRS, the use of an entity which centralizes all in artificial intelligence [2] whose traditional applications
data accounts for a bottleneck in terms of decision comput- concern planning, scheduling, placement, logistics and so
ing and communication. on. A constraint satisfaction problem (CSP) can be viewed
To avoid these drawbacks, some works have investigated as a triplet (X, D, C) in which: X is a finite set of vari-
multi-robot coordination in a distributed way. In [14], the ables, each variable xi ∈ X is associated to a finite domain
authors propose an exploration algorithm based on the se- dom(xi ) ∈ D and relates to a finite set of constraints C.
lection of different behaviors: avoiding obstacles, main- Solving a CSP requires to find for each variable xi a value
taining network connectivity, exploring around the frontier. in dom(xi ) which is consistent with C (i.e. which does not
This selection takes into account the current network condi- violate any constraint of C).
tion which is known by each robot thanks to periodically Common algorithms for solving a CSP [5] consist in a
exchanged messages. To achieve the connectivity of the systematic search based on the backtracking principle. Such
network, each robot analyses the topology of the network algorithms incrementally try to construct a partial solution
and makes distinction between a simple articulation and a affecting to each variable xi a value which satisfies all the
bridge. A simple articulation is a link whose deconnec- constraints in C. When this process fails before reaching
tion does not imply the loss of connectivity in the network. the last variable of X, a backtracking is performed to the
A bridge is a link whose deconnection creates two uncon- most recently affected variable that still has alternative val-
nected subnetworks. ues available in its domain. In other words, the backtrack-
Many other works use multi-agent bidding algorithms to ing principle allows to prune a subspace from the cartesian
achieve multi-robot coordination. An example of such an product of all variable domains when a partial solution vio-
approach can be found in [10]. The authors introduce a lates one of the constraints of C. Let us note that the order
bid calculation allowing robots to find their best target lo- of values in each dom(xi ) (1 ≤ i ≤ card(X)) has an influ-
cations. The bid calculation is based on a ponderated sum ence in the resolution of the problem since a backtracking
of three elements: the potential information gain of the tar- algorithm successively tests, for each variable, each value
get location, the distance from the current position of the present in its domain. Clearly, this notice is used by many
robot to the target location and a nearness measure intended heuristics designed to speed up the resolution of a CSP or
to characterize the ability to maintain communication links to favour some solutions among all possible solutions.
with other robots. To perform coordination, robots periodi- The concept of distributed CSP has been introduced to
cally broadcast their best bids to all other robots within the formalize and solve naturally distributed decision problems
[17] which generally deal with a set of data, shared out • periodically, the robots exchange their map so that they
among many sites and whose centralization is often impos- can own a global map and know the position of the
sible. frontier,
A disCSP (X, D, C, A) is an extension of the
triplet (X, D, C) where A is a finite set of agents • each robot is able to know its own position and the
{A1 , A2 , . . . , Ap } in which other robots’ ones.
T each Ak (1 ≤ k ≤ p) owns a
subset of X: XAk with Ak ∈A XAk = ∅ and a subset of
To express the multi-robot exploration as a distributed
C: CAk . From the point of view of agent Ak , variables
constraint satisfaction problem, we have to discretize the
of set XAk are called ”owned variables” whereas the set
space of actions of robots. The movement of a mobile robot
X \ XAk refers to the ”foreign variables”.
is usually the result of a combination of lateral and longitu-
Most of algorithms to solve disCSP are distributed and dinal accelerations. To simplify our speech, we consider 8
asynchronous. To execute such algorithms, an agent has to possible directions corresponding to the cardinal points of
be able to send messages to any other agents of its accoin- a compass: go to the north, go to the north-east, go to the
tance set1 . As the algorithms are asynchronous, delays can east, go to the south-east, etc.
occur when messages are exchanged through the commu- Considering this discretization, each robot has to find a
nication network. Nevertheless, we assume that when two decision among these 8 possibilities such that three require-
agents exchange successive messages, the sending order is ments are fulfilled:
preserved when receiving.
The first algorithm introduced in literature, ABT (Asyn- 1. its future location does not break the connectivity of
chronous BackTracking) [16], is an adaptation of the back- the network,
tracking principle to the multi-agent context. To achieve
completeness and avoid infinite loops during resolution, this 2. its future location does not induce overlaps between
algorithm uses a total statically order on A; which means the sensor ranges,
that Ai has a smaller priority than Aj if i > j. In ABT,
3. its future location allows to discover new unexplored
an agent also keeps in memory two sets: the agent view,i.e
areas.
the set of values that it believes to be assigned to the agents
of its accointance set and the nogood store, the set of val- To evaluate if a movement will break or not the con-
ues which implies an impossible assignment (this set can nectivity of the network, a robot has to be aware about it.
be viewed as a justification of inconsistent values). Each In practice, a robot has to identify among all these direct
agent concurrently instanciates its variables and sends its connections the one that allows to access to the rest of the
solution to other agents of its accointance set. When receiv- network (the others being able to be broken if necessary).
ing a new assignment, an agent updates its agent view and, Our algorithm presented in [6] allows the robots to obtain
if necessary, tries to change its own assignment to be consis- such information: based on exchanged messages in the net-
tent with its constraints. If an agent cannot find any consis- work, each robot constructs a table containing the id of ac-
tent value with the different assignments already received, cess robots with which it has to maintain a direct connection
it generates new nogoods and sends a backtracking message in order to stay in touch with a reference node2 .
to the closest (considering the total order defined in A) in-
volved agent. ABT terminates when a solution is found,
or when an empty nogood is generated, meaning that the
disCSP has no solution. Recent works improve ABT (see
[11] or [1] for details).

3.2 Modelling the multi-robot exploration prob-

lem as a disCSP

To expose our proposal, we consider the traditional hy-

pothesis made in works related to multi-robot exploration:

• each robot, during its movement, updates its map with

the new areas discovered, Figure 1. A networked robotics system
1 In MAS, the term accointance refers to the set of agents with which 2 When all robots maintain a communication path towards a reference

an agent can exchange messages knowing their address and id node, then the connectivity of the network is assured.
 For instance, considering the networked robotics system • X = {x1 , x2 , . . . , xp } is composed of variables stor-
in Figure 1, robots A7 and A3 are able to construct the ing the next heading of each robot of A.
following tables with A1 as the reference node:
• D = {dom(x1 ), . . . , dom(xp )} with dom(xi )(1 ≤
i ≤ p) is the set of all 8 cardinal directions that
Access Robots Access Paths a robot Ai can choose to plan its next movement.
• table of A3 :
A2 (A2 , A1 ) The domain is ordered by the following rela-
tion: v1 4 v2 ≡ dist(f p(Ai , v1 ), f rontier) <
dist(f p(Ai , v2 ), f rontier) with (v1 , v2 ) ∈
Access Robots Access Paths dom(xi )2 .
• table of A7 : A3 (A3 , A2 , A1 )
A4 (A4 , A2 , A1 ) • C = C1 ∪ C2 where:

C1 = {∀Ai ∈ A, ∃Aj ∈ ARAi ,

In the following, we denote ARAi the set of access
dist (f p(Ai , xi ), f p(Aj , xj )) < cr}
robots of a robot Ai . In the previous example, we have:
ARA3 = {A2 } and ARA7 = {A3 , A4 }. C2 = {∀Ai ∈ A, ∀Aj ∈ A \ Ai ,
Thus, the requirement 1. can be expressed as a constraint 2sr < dist (f p(Ai , xi ), f p(Aj , xj ))}
defined by an inferiority test between two values: the dis-
tance between the future positions of the two robots and the with:
communication range (Figure 2).
To maximize the discovery of new unexplored areas, a • sr the sensor range of a robot3 ,
robot has to be closed to the frontier. The requirement 3.
• dist(p1 , p2 ) the euclidian distance between the posi-
can be expressed as a specific ordering between the 8 car-
tion p1 and the position p2 ,
dinal positions such that the first cardinal position allows to
have the lowest distance between the robot and the frontier, • f p(Ai , xi ) the future position of Ai considering its fu-
the second cardinal position allows to have the second low- ture direction xi and its current vector speed,
est distance, etc. To enforce the efficiency of the exploration
by reducing overlaps (requirement 2), we can also impose • cr the communication range of a robot.
that the distance between the future positions of any two
robots of A have to be superior to the sum of their sensor To explore an unknown environment, the robots of set A
ranges. have to periodically solve this disCSP in order to be able to
choose a heading compatible with the requirements previ-
ously introduced. We present the resulting algorithm in the
following.

3.3 Multi-robot exploration algorithm based on

distributed constraint reasoning

The proposed multi-robot exploration algorithm consists

in repeating the following sequence until the end of the mis-
sion:

1. update maps and connectivity tables for each robot,

2. construct the disCSP based on connectivity tables and

Figure 2. Connectivity constraint between current positions of robots,
two robots
3. order the value of each domain taking the distance to
the frontier into account,
This previous statement can be expressed as the follow-
4. solve the disCSP to obtain future directions of each
ing disCSP:
robot,
• A = {A1 , A2 , . . . , Ap } denotes a fleet of p robots ex- 3 To simplify our speech, is made the assumption that all robots have
ploring an unknown environment and sharing a com- identical sensor capacities. This assumption is also made for the commu-
mon map of already explored areas. nication range.
 5. operate the movement of each agent during a fixed Considering the situation of Figure 4, the strict applica-
time period. tion of relation 4 between W and SE gives W 4 SE since,
in term of the euclidian distance, f p(Ai , W ) is closer to the
The mission is finished when there is no more area to ex-
frontier than f p(Ai , SE). Obviously, the frontier is not di-
plore. To be able to detect this state, each robot has to own
rectly reachable from f p(Ai , W ) because of the obstacle.
a global view of the environment. Clearly, this implies that
To take into account such a situation, the set of all possible
robots build this global view by exchanging periodically
headings is splited into two subsets: the set H̄ whose head-
their local map. In pratice, our algorithm can be run con-
ings lead into obstacles and the set H which contains the
currently with a process realizing simultaneous localisation
remained free headings.
and mapping (SLAM) [4].

algorithm explore

1: begin
2: while f rontier exists do
3: update map and connectivity table
4: XAi ← {xi }
5: CAi ← {dist(f p(Ai , xi ), f p(Aj , xj )) < cr}Aj ∈ARAi
6: CAi ← CAi ∪ {2 sr < dist(f p(Ai , xi ), f p(Aj , xj ))}Aj ∈A\Ai
7: heading ← {N, N E, E, SE, S, SW, W, N W }
8: H̄ ← ∅
9: for each h ∈ heading do
10: if there is an obstacle towards h then
11: H̄ ← H̄ ∪ {h}
Figure 4. Future possible positions according
12: end
13: end
to current speed and obstacles position
14: H ← heading\H̄
15: dom(xi ) ← {va ∈ H/∀(va , vb ) ∈ H 2 , a < b ⇒ va 4 vb }
16: dom(xi ) ← dom(xi ) ∪ H̄ At line 15, the domain of xi is initialized with the val-
17: solveDisCSP ues of H ordered with the relation 4. In some cases, the
18: move towards direction xi during dt seconds only way to maintain connectivity is to choose a direc-
19: end tion towards an obstacle. Consequently, the values of the
20: end set H̄ are added at the end of the domain dom(xi )(line
16). In the example, such an ordering policy gives
Figure 3. Distributed exploration algorithm {N, N E, S, E, SW, SE, N W, W } as the domain for the
variable xi .
Figure 3 details the exploration algorithm run by robot The disCSP is solved at line 17. Here any disCSP solv-
Ai . The algorithm stops when there is no more unexplored ing algorithm can be used such as ABT [16], AWS [17],
area as expressed at line 2. Line 3 concerns the updates of ABTnot [1], etc. Finally, the movement of the robot is oper-
the map and the connectivity table of Ai using the algorithm ated at line 18. To take the obstacle into account in a finer
presented in [6]. manner, a navigation algorithm can be used [3].
From lines 4 to 16, Ai constructs its part of the disCSP.
At line 5, Ai is aware about its own connectivity with the
4 Implementation and Validation
rest of the fleet and is able to initialize its set CAi with the
constraints of connectivity maintenance. At line 6, Ai com-
pletes its set CAi with the constraints of non overlap be- 4.1 Implementation details
tween sensor ranges.
Lines 7 to 16 concern the ordering of the do- To evaluate the feasibility and the relevance of our ap-
main of xi . Among the set of all possible headings proach, we have implemented the exploration algorithm in
{N, N E, E, SE, S, SW, W, N W }, the robot Ai selects the NetLogo. NetLogo is a multi-agent programmable mod-
one which brings him closer to the frontier. The order re- eling environment [13] which allows to prototype quickly
lation 4 compares the distances to the frontier from two systems of situated agents evolving in a two dimensions
future positions. The future position of a robot can be cal- world. Our considered environments are closed and mod-
culated using the current speed and heading of the robot and eled as a grid with 100 × 100 cells. Each cell can be empty,
the position of known obstacles. occupied by a robot or an obstacle, explored or unknown.
 Figure 5. Examples of environments with different complexity and density of obstacles

Many parameters of the robots like the sensor range, 4000

wifi range = 25
wifi range = 50
the communication range and dt can be tuned. Many data
3500
are recorded at each simulation: exploration duration, trav-
elled distance by each robot, number of exchanged mes- 3000
sages by robots. In addition, specific data about disCSP
solving are available: number of ”nogood” messages, num- 2500
time

ber of ”ok” messages, number of cycles, number of checked 2000

constraints. The disCSP solving algorithm used for our ex-
periments is ABT whose implementation has been provided 1500

by [7] in community models of NetLogo.

1000

4.2 Simulations 500

1 2 3 4 5 6 7 8
number of robots

In our simulations, we have considered environments

Figure 6. Experiments with a "simple" envi-
with different levels of complexity depending on: the shape
ronment
of the obstacles (concave vs convexe), the size of the obsta-
cles (compared to the size of one robot), the density of the
3500
obstacles, the minimum distance between two close obsta- wifi range = 25
wifi range = 50
cles. Figure 5 gives two examples of possible environments. 3000
At the beginning of each simulation, the robots are ini-
tially aligned in the bottom right corner of the environment 2500
with a distance between them inferior to the communication
range. For our experiments, we have considered two differ-
time

2000
ent communication ranges: a short one and a large one. Fig-
ures 6 and 7 are examples of obtained results respectively 1500
for a simple and a complex environment. They present the
average of exploration durations according to the number of 1000

robots deployed in the environment.

Our different experimentations show that the typology 500
1 2 3 4 5 6
number of robots
of the environment has a great impact on the duration of
the exploration: a ”simple” environment is faster explored Figure 7. Experiments with a "complex" envi-
than a ”complex” one. Moreover, for closed environments, ronment
adding robots improves the speed of the exploration to a
certain limit: when robots are too many, they interfer with
each other and spend more time avoiding each other than
exploring. can be interpreted as a deadlock situation. To overcome this
Clearly, a large communication range also helps to im- problem, the authors propose to take into account the lim-
prove the exploration: robots are less constrained in their ited movements of the robots which is more difficult to de-
movements and can cover a larger area of exploration. On tect. In our approach, the deadlock detection will be easier:
the other hand , we have noticed that initial positions of all robots are blocked when the solving disCSP algorithm
robots in the environment have a minor influence on the ex- does not find any solution.
ploration. As future works, we also plan to test our approach on
Figure 8 presents the average of exchanged messages for several Wifibot4 robots. For practical experiments, we will
different sizes of robots’ team to accomplish a full explo- have to choose a value for dt. Let us consider a robot at
ration. When the number of robots increases, the size of a time instant t. In the interval [t, t + dt], the robot has
the disCSP grows and consequently more messages are re- to operate the movement planned during the interval [t −
quired to find a solution. Meanwhile, the increase of the dt, t] and solve a disCSP to determine its next movement.
number of exchanged messages is balanced by the fact that Clearly, dt has to be sufficiently long to allow the resolution
the duration of the exploration is reduced with more robots. of a disCSP and operate the current movement taking the
Let us remind that we have used in our prototype the sim- physic limitations of the robot into account. But dt has not
plest algorithm (ABT) for solving the disCSP. Using a more to be too long in order to avoid too large errors in the future
sophisticated algorithm will help to reduce the number of position estimation.
exchanged messages at each decision step.
6 Conclusion
25000

This paper proposes an original way to deal with multi-

Number of exchanged messages

20000 robot exploration under limited communication range.

From a robot point of view, such a task consists in choos-
15000 ing its best decision such as it can explore unknown areas
and maintain connectivity with the rest of its fleet. To state
10000
this coordination problem, we use a well known formalism
in artificial intelligence: the distributed constraint satisfac-
tion problems (disCSP). Based on this foundation, a new
5000
distributed algorithm for the coordination of robots is intro-
duced. Using its own connectivity awareness, each robot
0
1 2 3 4 5 6 7 8 constructs a part of the disCSP by adding constraints with
number of robots
the rest of the fleet. The distributed resolution of the disCSP
Figure 8. Number of exchanged messages gives the future direction of each robot.
during an exploration Simulations give interesting results: reasonable number
of exchanged messages, decrease of the exploration dura-
tion when the number of robots grows, robust connectivity
maintenance, easier deadlock detection. Next step of this
5 Future works work will consist in implementing our approach on several
Wifibot robots.
Works in progress investigate particular environ- We also believe that the multi-robot exploration issue,
ments composed with walls and rooms (http://vst. presented in this article, provides a relevant and real case of
ensm-douai.fr/robotics/exploration). Such application of distributed constraint reasoning which could
environments introduce another level of complexity since be declined as an interesting benchmark to evaluate and
deadlocks can occur when robots are stucked in separate compare disCSP solving algorithms.
rooms.
In their article, Rooker and Birk address this issue con- References
sidering that a deadlock can be detected when the frontier
remains unchanged for an ”enough” long period of time. [1] C. Bessiere, I. Brito, A. Maestre, and P. Meseguer. Asyn-
They state that this information can be deduced by each chronous backtracking without adding links: A new member
robot with an analysis of the map. Unfortunately, at the in the ABT family. Artificial Intelligence, 161:7–24, 2005.
end of the exploration, when almost the entire environment [2] R. Dechter. Constraint Processing. Morgan Kaufmann,
has been explored, robots have to travel a larger distance to 2003.
reach the frontier. Consequently, the end of the exploration 4 http://www.wifibot.com/
 [3] M. Defoort, J. Palos, A. Kokosy, T. Floquet, and W. Perru-
quetti. Performance-based reactive navigation for nonholo-
nomic mobile robots. Robotica, 27(2):281–290, 2009.
[4] H. Durrant-Whyte and T. Bailey. Simultaneous localization
and mapping: Part I and II. IEEE Robotics & Automation
Magazine, pages 99–117, 2006.
[5] V. Kumar. Algorithms for constraint-satisfaction problems:
A survey. AI Magazine, 13(1):32–44, 1992.
[6] V. T. Le, N. Bouraqadi, V. Moraru, S. Stinckwich, and
A. Doniec. Making networked robot connectivity-aware.
In Proceedings of the IEEE International Conference on
Robotics and Automation (ICRA), pages 3502–3507, 2009.
[7] I. Muscalagiu, J. M. Vidal, and H. Jiang. Random binary
CSPs using asynchronous backtracking. NetLogo Models,
http://discsp-netlogo.fih.upt.ro/, 2007.
[8] M. N. Rooker and A. Birk. Multi-robot exploration under
the constraints of wireless networking. Control Engineering
Practice, 15(4):435–445, 2007.
[9] H. Shatkay and L. P. Kaelbling. Learning topo-logical maps
with weak local odometric information. In Proceeding of
the International Joint Conference on Artificial Intelligence
(IJCAI), pages 920–929, 1997.
[10] W. Sheng, Q. Yang, J. Tan, and N. Xi. Distributed multi-
robot coordination in area exploration. Robotics and Au-
tonomous Systems, 54:945–955, 2006.
[11] M. Silaghi and B. Faltings. Asynchronous aggregation and
consistency in distributed constraint satisfaction. Artificial
Intelligence Journal, 161(1-2):25–53, 2005.
[12] R. Simmons, D. Apfelbaum, W. Burgard, D. Fox, M. Moors,
S. Thrun, and H. Younes. Coordination for multi-robot ex-
ploration and mapping. In Proceedings of the National Con-
ference on Artificial Intelligence (AAAI), 2000.
[13] S. Tisue and U. Wilensky. Netlogo: Design and implementa-
tion of a multi-agent modeling environment. In Proceedings
of the Agent 2004 Conference, 2004.
[14] J. Vazquez and C. Malcolm. Distributed multirobot explo-
ration maintaining a mobile network. In Proceedings of the
2nd International IEEE Conference on Intelligent Systems,
volume 3, pages 113–118, 2004.
[15] B. Yamauchi. Frontier-based exploration using multiple
robots. In Proceedings of the Second International Confer-
ence on Autonomous Agents (Agent’98), 1998.
[16] M. Yokoo. Distributed constraint satisfaction for DAI prob-
lems. In Proceedings of the 10th International Workshop on
Distributed Artificial Intelligence., 1990.
[17] M. Yokoo. Distributed Constraint Satisfaction: Foundations
of Cooperation in Multi-agent Systems. Springer, 2001.