Reactivity and Social Cooperation in a Multi-Robot System
Atef Gharbi
1
, Nadhir Ben Halima
2
and Hamza Gharsellaoui
1
1
Department of Computer sciences, INSAT, Tunis, Tunisia
2
College of Computer Science and Engineering, Taibah University, Yanbu Branch, Yanbu Al-Bahr, K.S.A.
Keywords:
Multi-Robot Systems, Distributed Planning, Five Capabilities Model, Benchmark Production System.
Abstract:
Multi-Robot System (MRS) is an important research area within Robotics and Artificial Intelligence. The
balancing between reactivity and social cooperation in autonomous robots is really considered as a challenge
to get an effective solution. To do so, we propose to use the concept of five capabilities model which is based
on Environment, Self, Planner, Competence and Communication. We illustrate our line of thought with a
Benchmark Production System used as a running example to explain our contribution.
1 INTRODUCTION
In this paper, the term Multi-Robot System (MRS) in-
dicates a team of two or more autonomous robots per-
forming in a dynamic environment where uncertainty
and unforeseen changes can happen due to robots.
Besides, the control is not centralized, but rather dis-
tributed over the members of the team. Robots, to
achieve their goals successfully in dynamic and un-
predictable environments, must be able to generate
plans in a timely way, monitor changes in their envi-
ronment, change and adapt their plans accordingly. In
this paper, we regard Multi-Robot System as a partic-
ular form of Multi Agent System (MAS), by specif-
ically addressing planning and social abilities. This
mapping seems at the first look suprising as a robot
is considered a physical entity ensuring a variety of
tasks. However, there are many similarities between
Agent and Robot which pushes in this way (for exam-
ple, the robot like the agent has several activities and
responds to its environment). Besides, we believe that
the multi-agent technology applied to robotic system
is able to increase the flexibility of the system as a
whole.
This article is concerned with two important mat-
ters: how to define the Multi-Robot System in a man-
ner such that it has more utility to deploy it, and how
to use such a MRS for the advanced software. The
Multi-Robot System must discover the action to be
taken by supervising the application and its environ-
ment and analyzing the data obtained.
With Multi-Robot System, we face two important
matters: (i) the detection of a need for action: the need
for action must be discovered by supervising the ap-
plication and its environment and analyzing data ob-
tained. (ii) the planning of the action: it consists to en-
visage the action (by proposing which modifications
need to be made) and by programming it.
Using a multi-agent approach, the robot’s archi-
tecture can be decomposed into flexible autonomous
subsystems (agents). The architecture can then be de-
scribed at a higher level, defining the agents that have
to be in the system, the role of each of them, the in-
teractions among them, the actions each of them per-
forms, and the resources they need. Since the multi-
agent system is inherently multi-threaded, each agent
has its own thread of control; each agent decides
whether or not to perform an action at the request
of another agent (autonomy); agents establish agree-
ments among themselves, while keeping their auton-
omy sharing their knowledge and acting together to
accomplish specific common goals. Agents need to
interact to coordinate their activities so that control
of the robot is achieved. All of those processes, the
agent’s own decision making, interaction and coordi-
nation need to be highlighted. To do so, we propose
the design of a Robotic Agent according to the 5 Ca-
pabilities model (5C) proposed by (van Aart, 2004),
(C. J. van Aart and Schreiber, 2004). The 5 Capabil-
ities model is separated into five dimensions: Envi-
ronment, Self, Planning, Competence and Communi-
cation. These dimensions are said models where each
model represents one specific capability of the robotic
agent. First of all, a robotic agent needs to interact
with the environment in which it operates thanks to
sensors (providing data) and actuators (executing ac-
tions) therefore we define the Environment Model. To
know what tasks to be executed, we define the Self
Model. The self model is used to know the robotic
agent’s perception of its own being and state. In other
253
Gharbi A., Ben Halima N. and Gharsellaoui H..
Reactivity and Social Cooperation in a Multi-Robot System.
DOI: 10.5220/0005482202530260
In Proceedings of the 10th International Conference on Software Engineering and Applications (ICSOFT-EA-2015), pages 253-260
ISBN: 978-989-758-114-4
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
terms, it consists of ongoing tasks. The planning of
ongoing tasks is the concern of the Planner Model.
A planner model is ensuring some kind of reasoning
about task selection, execution control, time monitor-
ing and emergency handling.
The Competence Model ensures the methods the
abilities and the knowledge that enables the robotic
agent to execute the task that is designed for it. The
multi-robot system has the appropriate Communica-
tion Model in order to avoid non-feasible, unsecured
and fortuitous actions that can provoke undesirable re-
sults by a single robotfor the whole system, the differ-
ent robotic agents have to interact together following
a specific communication protocol.
This paper introduces a simple Benchmark Produc-
tion System that will be used throughout this article
to illustrate our contribution which is developped as
Robot-based application. We implement the Bench-
mark Production System in a free platform which
is JADE (JavaTM Agent DEvelopment) Framework.
JADE is a platform to develop multi-agent systems
in compliance with the FIPA specifications (Salva-
tore Vitabile, 2009), (Chuan-Jun Su, 2011), (Bordini
and all., 2006).
In the next section, we present the Benchmark
Production System. The third section introduces the
EnvironmentModel by specifying the sensors, the ac-
tuators and the safety requirements. The fourth sec-
tion presents the Self Model which describes the dif-
ferent tasks to be executed by the robotic agent with
a formal specification. We introduce in the fifth sec-
tion the Planner Model. The sixth section presents
the Competence Model based on Fuzzy Logic Sys-
tem. Finally, we study the Communication Model in
particular the message exchanged through a commu-
nication protocol. We conclude in the last section.
2 BENCHMARK PRODUCTION
SYSTEM
As much as possible, we will illustrate our contri-
bution with a simple current example called RARM
(Branislav Hrz, 2007). We begin with the description
of it informally,but it will serve as an example for var-
ious formalism presented in this article. The bench-
mark production system RARM represented in the fig-
ure 1 is composed of two input and one output con-
veyors, a servicing robot and a processing-assembling
center. Workpieces to be treated come irregularly one
by one. The workpieces of type A are delivered via
conveyor C1 and workpieces of the type B via the
conveyorC2. Only one workpiece can be on the input
conveyor. A robot R transfers workpieces one after
another to the processing center. The next workpiece
can be put on the input conveyor when it has been
emptied by the robot. The technology of production
requires that first one A-workpiece is inserted into the
center M and treated, then a B-workpiece is added
in the center, and last the two workpieces are assem-
bled. Afterwards, the assembled product is taken by
the robot and put above the C3 conveyer of output.
the assembled product can be transferred on C3 only
when the output conveyor is empty and ready to re-
ceive the next one produced.
A
Conveyor C1
A
B
C
o
n
v
e
y
o
r
C
3
B
Conveyor C2
Position p1
Position p2
Position p3 Position p4
P
osit
io
n
p
5
P
o
sitio
n
p6
Robot r
Processing unit
M
Figure 1: The benchmark production system RARM.
3 ENVIRONMENT MODEL
Perception is responsible for collecting runtime infor-
mation from the virtual environment. The perception
component supports selective perception, enabling a
robotic agent to direct its perception to its current
tasks. The perception component interprets the rep-
resentation resulting in a percept. A percept consists
of data elements that can be used to update the robotic
agent’s current knowledge.
Actuators
The system can be controlled using the following ac-
tuators: (i) move the conveyorC1 (act1); (ii) move the
conveyorC2 (act2); (iii) move the conveyorC3 (act3);
(iv) rotate robotic agent (act4); (v) move elevating the
robotic agent arm vertically (act5); (vi) pick up and
drop a piece with the robotic agent arm (act6); (vii)
treat the workpiece (act7); (viii) assembly two pieces
(act8).
Sensors
The control program receives information from the
sensors as follows: (i) Is there a workpiece of type A
ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications
254
at the extreme end of the position p1? (sens1) (ii) Is
the conveyor C1 in its extreme left position? (sens2)
(iii) Is the conveyor C1 in its extreme right position?
(sens3) (iv) Is there a workpiece of type A at the pro-
cessing unit M? (sens4) (vi) Is the conveyor C2 in its
extreme left position? (sens5) (vii) Is the conveyorC2
in its extreme right position? (sens6) (viii) Is there a
workpiece of type B at the extreme end of the posi-
tion p3? (sens7) (ix) Is there a workpiece of type B at
the processing unit M? (sens8) (x) Is the conveyorC3
in its extreme left position? (sens9) (xi) Is the con-
veyor C3 in its extreme right position? (sens10) (xii)
Is there a workpiece of type AB at the processing unit
M? (sens11) (xiii) Is the robotic agent arm in its lower
position? (sens12) (xiv) Is the robotic agent arm in its
upper position? (sens13)
4 SELF MODEL
We represent the whole system (i.e. RARM) with the
petri net. A token in the place p
A0
(resp. p
B0
) repre-
sents a situation when the conveyor C1 (resp. C2) is
empty. The event ”a workpiece occurred at the sys-
tem input” changes the input state. Afterwards the
workpiece is prepared for manufacturing. The event
is represented by firing t
A
. Transition t
A
(resp. t
B
)
is fireable and on its firing the token from p
A0
(resp.
p
B0
) is removed and is placed into the place p
A1
(resp.
p
B1
). In the Petri net it is not specified when the event
happens. If the robotic agent is available (a token is in
p
R
), a workpiece A is available (a token in p
A1
), and
the previous machine process is completed (a token in
p
MF
) then the robotic transfer of the workpiece A can
start (firing transition t
1
). Place p
TA
is occupied by a
token during the transport. The end of the transport
and loading the workpiece into machine M is speci-
fied by transition t
2
. After firing t
2
a token is placed
in p
MA
. End of processing of the workpiece A is
specified by transition t
3
and the start of the transport
of B into the machine by t
4
. Then a token comes
to the place p
TB
, which indicates that the transfer of
the workpiece B is in progress; t
5
denotes end of the
transfer and start of the assembly operation; t
6
spec-
ifies end of the assembly operation; t
7
specifies start
of the product transfer from M onto the conveyor C3
(t
7
is fired and a token appears in p
TO
). After the
transfer of the product AB on C3 and its leaving out
the cell, the token moves to the p
OF
place. The place
p
MF
ensures that the next part A is loaded into ma-
chine M only after the assembly process has been fin-
ished. The next workpiece A can be loaded at the
earliest on the conveyor C1 (firing t
A
) when the pre-
ceding workpiece A is in the transfer to the machine
M (p
TA
marked). Considering it we have: t
1
fires and
then t
A
fires. First, A has to arrive to be processed in
M (a token in p
MA
) and after that it waits in the ma-
chine (token in p
AW
) and only then B can be loaded
into the machine. The place p
ABW
corresponds to the
machine output. A token is in p
ABW
if the assembled
product AB is at the machine output. After the prod-
uct leaves machine M, M is free again.
PA0
PMA
tA t1
t2 t3
PR
PAW
PB0
PAl PTA
PMAB
PO
t6
t7
t8
t9
PABW
PTO
P2
tB
PMF
PB1
t4
PTB
t5
Figure 2: Petri Net of RARM.
5 PLANNER MODEL
Distributed planning is considered as a very complex
task (David Jung, 1999), (Oscar Sapena, 2008). In
fact, distributed planning ensures how the multi-robot
system should plan to work together, to decompose
the problems into subproblems, to assign these sub-
problems, to exchange the solutions of subproblem,
and to synthesize the whole solution which itself is
a problem that the robotic agents must solve (Sergio
Pajares Ferrando, 2013), (Pascal Forget, 2008), (Ma-
lik Ghallab, 2014). The actions of the other robotic
agents can induce a combinatorial explosion in the
number of possibilities which the planner will have
to consider, returning the space of research and the
size of solution exponentially larger.
The Decision making component encapsulates a
behavior-basedaction selection mechanism. Decision
making is responsible for realizing the robotic agent’s
tasks by invoking actions in the virtual environment.
To enable a situated multi-robot system to set up col-
laborations, behavior-based action selection mecha-
nisms are extended with the notions of role and situ-
ated commitment. The conceptual model of a robotic
agent is constituted by three components including (i)
the planner, (ii) the plan-execution agent, and (iii) the
ReactivityandSocialCooperationinaMulti-RobotSystem
255
world in which the plans are to be executed (the for-
mal representation is based on the work (Malik Ghal-
lab, 2004)).
The planner’s input includes descriptions of the
state transition system denoted by Σ, the initial
state(s) that Σ might be in before the plan-execution
robotic agent performs any actions, and the desired
objectives (e.g., to reach a set of states that satisfies a
given goal condition, or to perform a specified task,
or a set of states that the world should be kept in or
kept out of, or a partially ordered set of states that we
might want the world to go through). If the planning
is being done online (i.e., if planning and plan execu-
tion are going on at the same time), the planner’sinput
will also include feedback about the current execution
status of the plan or policy. The planner’s output con-
sists of either a plan (a linear sequence of actions for
the robotic agent to perform)or a policy (a set of state-
action pairs with at most one action for each state).
P
o
s
i
tion
p
6
P
o
s
i
t
i
on
p
5
Robot r
Processing
unit M
S
12
take3
put3
take3
put3
Robot r
Processing
unit M
S
15
Conveyor
C1
C
o
n
v
e
y
o
r
C
3
Conveyor
C2
Conveyor
C2
Conveyor
C1
C
o
n
v
e
y
o
r
C
3
A
B
AB
P
o
s
i
tio
n
p
6
Positio
n
p5
Robot r
Processing
unit M
S
13
Conveyor
C2
Conveyor
C1
C
o
n
v
e
y
o
r
C
3
AB
P
o
sit
i
o
n
p5
P
o
s
it
io
n
p
6
Robot r
Processing
unit M
Conveyor
C1
C
o
n
v
e
y
o
r
C
3
Conveyor
C2
AB
P
ositi
o
n
p
5
P
osition
p
6
S
16
R3_left R3_right
S
12
Robot r
Processing
unit M
S
14
Conveyor
C1
C
o
n
v
e
y
o
r
C
3
Conveyor
C2
AB
Po
sit
i
o
n
p
5
P
os
i
t
i
o
n
p
6
C3_left C3_right
Figure 3: The third state-transition for RARM.
Running Example
According to figure 3 :
• A set of positions {p
1
, p
2
,...} : A position is used
to localise the workpiece A, B or AB;
• A set of robotic agents {r
1
, r
2
, ...} : Each robotic
agent transfers a workpiece one after one to be
processed;
• A set of workpieces of type A {a
1
,a
2
, .. .};
• A set of workpieces of type B {b
1
,b
2
, .. .};
• A set of workpieces of type AB {ab
1
,ab
2
, .. .};
• A set of conveyors {C
1i
, C
2i
, C
3i
} : A conveyorC
1i
(resp. C
2i
, C
3i
) is responsible for transfering set of
workpieces of type A (resp B, AB);
• A set of processing Centers M {M
1
, M
2
,...} : first
one A-workpiece is inserted into M and processed,
then one B-workpiece is added into the center M,
and last both workpieces are assembled.
The set of states is {s
0
, s
1
, s
2
, s
3
, s
4
, s
5
, s
6
, s
7
, s
8
,
s
9
, s
10
, s
11
, s
12
, s
13
, s
14
, s
15
, s
16
}
There are nine possible actions in the domain.
• a workpiece of type A is trasnported to the left
from position p1 to position p2;
• the robotic agent transports a workpiece of type
A;
• the piece is put in the processing unit M;
• a workpiece of type B is trasnported to the left
from position p3 to position p4;
• the robotic agent transports a workpiece of type
B;
• the piece is put in the processing unit M;
• the robotic agent picks up the assembled piece;
• the assembled piece is put on the conveyor C3;
• a workpiece of type AB is trasnported to the right
from position p5 to position p6.
The set of actions is {C1
le ft, C1 right, R1 left,
R1 right, C2 left, C2 right, R2 left, R2 right,
C3
le ft, C3 right, R3 left, R3 right, take
1
, take
2
,
take
3
, load
1
, load
2
, load
3
, put
1
, put
2
, put
3
, process
1
,
process
2
}
The current configuration of the domain is de-
noted using instances of the following predicates,
which represent relationships that change over time.
• occupied(c1) (resp. occupied(c2), occupied(c3)):
conveyor c1 (resp. c2, c3) is already occupied by
a workpiece of type A (resp. B, AB);
• empty(c1) (resp. empty(c2), empty(c3)): conveyor
c1 (resp. c2, c3) is already ready to transport a
workpiece of type A (resp. B, AB);
• at(r,p2) (resp. at(r,p4), at(r,p5)): robotic agent r
is currently at position p2 (resp. p4, p5);
• loaded(r, a) (resp. loaded(r, b), loaded(r, ab)) :
robotic agent r is currently loading the workpiece
a (resp. b, ab) of type A (resp. B, AB);
ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications
256
• put(r, a) (resp. put(r, b),put(r, ab)): robotic agent
r is currently putting the workpiece a (resp. b, ab)
of type A (resp. B, AB);
• empty(r): the robotic agent r is empty;
• empty(a) (resp. empty(b), empty(ab)): there is no
workpiece of type A (resp. B, AB).
6 COMPETENCE MODEL
The Competence Model of a robotic agent is based
on a Fuzzy Logic Control. The Fuzzy Logic Con-
trol is a methodology considered as a bridge on the
artificial intelligence and the traditional control the-
ory (Mria Kukov, 2013). This methodology is usu-
ally applied in the only cases when exactitude is not
of the need or high importance (Jianhua Dai, 2013).
As it is stated in (Marijana Gorjanac Ranitovi, 2014)
, Fuzzy Logic is a methodology for expressing oper-
ational laws of a system in linguistic terms instead of
mathematical equations.
The basic form of a fuzzy logic agent consists of
(Zadeh, 2008): Input fuzzification, Fuzzy rule base,
Inference engine and Output defuzzification.
Running Example
(i) Fuzzification
The number of defected pieces is measured through
a sensor related to the system. The range of number
of defected pieces varies between 0 to 40, where zero
indicates the rate of defected pieces of A that is null
(each piece is well) and 40 indicates the rate of de-
fected pieces of A is very high.
Now assume that the following domain meta-data val-
ues for these variable, VF = very few, F = few, Md
= medium, Mc = much, VMc = very much. Assume
that the linguistic terms describing the meta-data for
the attributes of entities are: VF = [0,..,10], F =
[5,..,15], Md = [10,..,20], Mc = [15,..,25] and VMc
= [20,..,40].
Based on the metadata value for each attribute the
membership of that attribute to each data classifica-
tion can be calculated. In the Figure 4, triangular and
trapezoidal fuzzy set was used to represent the state
of defected pieces from A classifications (i.e. state of
defected pieces from A classification levels: VF , F,
Md, Mc, VMc whereas state of defected pieces from B
classification levels: F, Md, Mc).
(ii) Rule Engine
We take as example, the first column from the Table 1:
IF number of defected pieces from A is Very Few and
number of defected pieces from B is Few Then Pro-
duction is High.
Number of
defected pieces A
VF Md VMc
1
0
Degree of
Membership
F Mc
5 10 252015
Figure 4: Fuzzy State of defected pieces from A.
IF number of defected pieces from A is Few and num-
ber of defected pieces from B is Few Then Production
is High.
...
Table 1: Fuzzy Control rules for the robotic agent.
A B F Md Mc
VF H H M
F H H M
Md H M L
Mc M L N
VMc M L N
(iii) Defuzzification
Defuzzification is the conversion of a fuzzy quantity to
a precise quantity. There are many methods to cal-
culate it such as Max membership, Centroid method,
Weighted average method, Mean max membership,
Center of sums, Center of largest area and First (or
last) of maxima. Obviously, the best defuzzification
method is context-dependant (Zadeh, 2008).
7 COMMUNICATION MODEL
Communication is responsible for communicative in-
teractions within a multi-robot system. Message ex-
change enables robotic agents to share information di-
rectly and set up collaborations. The communication
module processes incoming messages and produces
outgoing messages according to well-defined com-
munication protocols. To do that, we implement the
different robotic agents with the platform JADE (for
more details, we refer to (Fabio Bellifemine, 2010b),
(Caire, 2009), (Fabio Bellifemine, 2010a)).
Running Example
The robotic agent RARM
1
is going to reduce the
production. The another robotic agent RARM
2
may
either decrease the production (in which case the
robotic agents can cooperate together) or increasethe
production (in which case neither robotic agent can
ReactivityandSocialCooperationinaMulti-RobotSystem
257
cooperate). Thera are two possible actions can be
modeled as nondeterministic outcomes.
7.1 Message Exchanged Between
Robotic Agents
A multi-robot system never interact through method
calls but by exchanging asynchronous messages. Ob-
viously, inter-agent interaction will be very difficult
until all robotic agents adopt the same communi-
cation language, and fortunately ACL standards en-
sure this requirement. This format comprises a num-
ber of fields and in particular: (1) the sender of
the message, (2) the list of receivers, (3) the com-
municative intention (also called performative) indi-
cating what the sender intends to achieve by send-
ing the message (for example the performative can
be REQUEST, INFORM, QUERY
IF, CFP (call for
proposal), PROPOSE, ACCEPT
PROPOSAL, RE-
JECT PROPOSAL, and so on). (4) The content i.e.
the actual information included in the message which
may be string in simple cases; otherwise we need a
content language, a corresponding ontology, and a
protocol. (5) The ontology i.e. the vocabulary of the
symbols used in the content and their meaning (both
the sender and the receiver must be able to encode ex-
pressions using the same symbols to be sure that the
communication is effective).
7.1.1 Sending a Message
Sending a message to another robotic agent is as sim-
ple as filling the fields of an ACLMessage object and
then call the send() method of the Agent class.
Running Example
//The code below informs a robotic agent whose
nickname is Robot1 that the production must be
decreased.
ACLMessage msg = new ACLMes-
sage(ACLMessage.INFORM);
msg.addReceiver(new AID(”Robot1”,
AID.ISLOCALNAME));
msg.setOntology(”Production”);
msg.setContent(”We must decrease in the produc-
tion”);
send(msg);
7.1.2 Receiving a Message
A robotic agent can pick up messages from its mes-
sage queue by means of the receive() method. This
method returns the first message in the message queue
(removing it) or null if the message queue is empty
and immediately returns.
Running Example
ACLMessage msg = receive();
if (msg ! = null) {
// Process the message
}
7.1.3 Blocking Behavior Waiting a Message
Some behaviors must be continuously running and at
each execution of their action() method, must check
if a message is recceived and perform some action.
Running Example
public void action() {
ACLMessage msg = myAgent.receive();
if (msg ! = null) {
// Message received. Process it
...
}
else {
block();
}
}
7.1.4 Selecting a Message
When a template is specified, the receive() method
returns the first message (if any) matching it,
while ignores all non-matching messages. Such
templates are implemented as instances of the
jade.lang.acl.MessageTemplate class that provides
a number of factory methods to create templates in a
very simple and flexible way.
Running Example
The action() method is modified so that the call to
myAgent.receive() ignores all messages except those
whose performative is REQUEST:
public void action() {
MessageTemplate mt = MessageTem-
plate.MatchPerformative(ACLMessage.REQUEST);
ACLMessage msg = myAgent.receive(mt);
if (msg ! = null) {
// REQUEST Message received. Process it
...
}
else {
block();
}
}
8 RELATED WORK
In literature, we can find three kinds of architecture.
The first one is Shakey’s architecture which is decom-
ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications
258
posed into three functional elements: sensing, plan-
ning, and executing (Nilsson, 1980). The sensing sys-
tem translates the camera image into an internal world
model. The planner takes the internal world model
and a goal and generates a plan (i.e., a series of ac-
tions) that would achieve the goal. The executor takes
the plan and sends the actions to the robot. This ap-
proach is called the sense-plan-act (SPA) paradigm.
But the SPA paradigm has problems. First, plan-
ning in any realworld domain takes a long time, and
the robot would be blocked waiting for planning to
complete. Second, and more importantly, execution
of a plan without involving sensing is dangerous in a
dynamic world.
The second kind of architecture is Reactive plan-
ning in which plans are generated quickly and relied
more directly on sensed information (Fir, ). The most
influential work, however, is the Subsumption archi-
tecture of Rodney Brooks of MIT (Brooks, 1986a). A
Subsumption architecture is built from layers of in-
teracting finite state machines each connecting sen-
sors to actuators directly. These finite state machines
were called behaviors (leading some to call Sub-
sumption ”behavior-based” or ”behavioral” robotics
(Arkin, 1998)).
However, it is very difficult to compose behaviors
to achieve long-range goals and it is almost impossi-
ble to optimize robot behavior.
The last kind of architecture is Layered Robot
Control Architectures. Among the many instances
of layered architectures, we can distinguish between
two fundamental classes: horizontally layered archi-
tectures (such as the ones developed by (Brooks,
1986b), (Kaelbling, 1990), and (Ferguson, 1992))
and vertically layered architectures (such as MECCA
(D. D. Steiner and Lerin, 1993) and (Muller and Pis-
chel, 1994)). Whereas all the layers of an agent have
access both to the perception and action components
in horizontal architectures, only one (and normally:
the lowest) layer has a direct interface to these facili-
ties in the vertical approach.
Currently two kinds of contributions in the multi-
robot systems are considered related to this paper,
namely reactivity and social cooperation.
The difference of the architecture proposed in this pa-
per lies in emphasis on reactivity and social cooper-
ation between multiple robots, while most of other
architectures analysed by Goodwin (Goodwin, 2008)
are concentrating on controlling multiple parts of a
single robot.
The main difference of the proposed architecture
from STEAM (Tambe, 1997) is in the task allocation
mechanism. While STEAM uses the main task coor-
dinator (team leader) that issues orders to team mem-
bers, the proposed architecture focuses on individual
agents being able to apply for tasks, therefore increas-
ing their autonomy.
In ALLIANCE (Simmons, 1994), it is harder to
replace actual robots with a software simulator, be-
cause single point of interface between deliberative
and reactive layers makes it easier to test application.
The main difference from CENTIBOTS is in com-
plexity of each robot CENTIBOT system uses robots
with complex sensors and processing units while
the proposed architecture focuses on use of low-cost
hardware and in order to achieve sufficient environ-
mental data quality, supplements deliberative level
with appropriate algorithms and intelligence.
9 CONCLUSION
The main aim of this paper is how to ensure a dis-
tributed planning in Multi-Robot System composed
of several intelligent autonomous robotic agents able
to take the initiative instead of simply reacting in
response to its environment. Our solution to this
problem is the use of the 5 Capabilities Model (as
it was presented, 5 levels: Environment, Self, Plan-
ner, Competence and Communication). The 5 Ca-
pabilities Model can be easily implemented where
each model is represented with a process collaborat-
ing with the other processes. The 5C Model, based
on the principle of separation of concerns, has the
following interests: (i) The design is general enough
to cope with various kinds of embedded-software ap-
plication (therefore, the 5C Model is uncoupled from
the application); (ii) The robotic agent is represented
through five dimensions where each model is inde-
pendent from the other which permits to change one
without having to change the other.
Our future work is the design of an autonomous
robot integrating cognitive abilities with other capa-
bilities such as locomotion, prehension and manipu-
lation.
REFERENCES
Arkin, R. C. (1998). Behavior-based robotics. MIT Press,
Cambridge MA.
Bordini, R. and all. (2006). A survey of programming lan-
guages and platforms for multi-agent systems. Infor-
matica, 30(1):33–44.
Branislav Hrz, M. Z. (2007). Modeling and control of
discrete-event dynamic systems with petri nets and
other tools. page 67.
ReactivityandSocialCooperationinaMulti-RobotSystem
259
Brooks, R. A. (1986a). A robust layered control system for
a mobile robot. IEEE Journal of Robotics and Au-
tomation, 2:14–23.
Brooks, R. A. (1986b). A robust layered control system
for a mobile robot. In IEEE Journal of Robotics and
Automation, 2.
C. J. van Aart, B. J. W. and Schreiber, A. T. (2004). Or-
ganizational building blocks for design of distributed
intelligent system. International Journal of Human-
Computer Studies, 61(5):567599.
Caire, G. (2009). Jade Tutorial: Jade Programming For
Beginners.
Chuan-Jun Su, C.-Y. W. (2011). Jade implemented mo-
bile multi-agent based, distributed information plat-
form for pervasive health care monitoring. Applied
Soft Computing, 11(1):315–325.
D. D. Steiner, A. Burt, M. K. and Lerin, C. (1993). The
conceptual framework of mai2l. In Pre-Proceedings
of MAAMAW’93, Neuchatel, Switzerland.
David Jung, A. Z. (1999). An architecture for dis-
tributed cooperative planning in a behaviour-based
multi-robot system. Robotics and Autonomous Sys-
tems, 26(23):149–174.
Fabio Bellifemine, Giovanni Caire, T. T. G. R. (2010a).
Jade Programmers Guide.
Fabio Bellifemine, Giovanni Caire, T. T. G. R. R. M.
(2010b). Jade Administrators Guide. The publishing
company.
Ferguson, I. A. (1992). Touring machines: An architec-
ture for dynamic, rational, mobile agents. PhD thesis,
Computer Laboratory, University of Cambridge, UK.
Goodwin, J. (2008). A unified design framework for mobile
robot systems. PhD Thesis. University of the West of
England.
Jianhua Dai, H. T. (2013). Fuzzy rough set model for set-
valued data. Fuzzy Sets and Systems, 229(3):54–68.
Kaelbling, L. P. (1990). An architecture for intelligent re-
active systems. In J. Allen, J. Hendler, and A. Tate,
editors, Readings in Planning, pages 713–728.
Malik Ghallab, Dana Nau, P. T. (2004). Automated Plan-
ning.
Malik Ghallab, Dana Nau, P. T. (2014). The actor’s view
of automated planning and acting: A position paper.
Artificial Intelligence, 208(3):1–17.
Marijana Gorjanac Ranitovi, A. P. (2014). Lattice represen-
tations of interval-valued fuzzy sets. Fuzzy Sets and
Systems, 236:50–57.
Mria Kukov, M. N. (2013). Principles of inclusion and
exclusion for fuzzy sets. Fuzzy Sets and Systems,
232(3):98–109.
Muller, J. P. and Pischel, M. (1994). Integrating agent inter-
action into a planner-reactor architecture. In M. Klein,
editor, Proc. of the 13th International Workshop on
Distributed Artificial Intelligence, Seattle, WA, USA.
Nilsson, N. J. (1980). Principles of artificial intelligence.
Tioga Press, Palo Alto CA.
Oscar Sapena, Eva Onaindia, A. G. M. A. (2008). Engi-
neering applications of artificial intelligence. Some
Fine Journal, 21(5):698–709.
Pascal Forget, Sophie DAmours, J.-M. F. (2008). Multi-
behavior agent model for planning in supply chains:
An application to the lumber industry. Robotics and
Computer-Integrated Manufacturing, 24(5):664–679.
Salvatore Vitabile, Vincenzo Conti, C. M. F. S. (2009). An
extended jade-s based framework for developing se-
cure multi-agent systems. Computer Standards & In-
terfaces, 31(5):913–930.
Sergio Pajares Ferrando, E. O. (2013). Context-aware
multi-agent planning in intelligent environments. In-
formation Sciences, 227:22–42.
Simmons, R. (1994). Structured control for autonomous
robots. IEEE, 10(1):34–43.
Tambe, M. (1997). Agent architectures for flexible, prac-
tical teamwork. Proceedings of the National Confer-
ence on AI, Providence, Rhode Island, USA, pages 22–
28.
van Aart, C. J. (2004). Organization principles for multi-
agent architectures. PhD thesis, University of Amster-
dam, Faculty of Social and Behavioural Sciences.
Zadeh, L. A. (2008). Is there a need for fuzzy logic? Infor-
mation Sciences, 178(13):2751–2779.
ICSOFT-EA2015-10thInternationalConferenceonSoftwareEngineeringandApplications
260
