1104 Communication Methods: Instructors’ Positions At Istanbul Aydin University Distance Education
Institute
credibility of the process and created confidence for
the researcher.
As the reflection is central essence of the work
based learning, having high level of responsibility on
pre-planned actions provided confidence for the
researcher and reflection, negotiation on these
activities increased the learning.
5. Conclusions
It is work based project which requires
collaboration of the colleagues to propose change for
better working practice. Reflections and learning
experiences of researcher from research process are
aimed to share with academic community.
High risks and questionable rewards are the reality
for most complex organizations experiencing rapid
change. Work, even in higher education, is shifting
toward greater interdependence among individuals to
create collective and synergistic products and services
using advanced technology. As the boundaries
between traditional positions blur, role clarification
becomes increasingly important. In this learning
environment, the role of the ODL instructor requires
the merging of multiple roles. The convergence of
advances in computer technology, rapidly growing
enrollment needs, and cost cutting measures for higher
education suggest that innovative solutions are
required. The findings of this study illustrate the
complexity of the role of the online instructor through
a unique perspective in which two types of roles were
examined in great detail.
References
[1] Middlesex University Module Guide Handbook, (2008).
[2] T.V. Eilertsen, N. Gustafson, P. Salo, Action research and
micropolitics in schools, Educational Action Research 16
(2008) 295-309.
[3] H. Thomas, An analysis of the environment and
competitive dynamics of management education, Journal
of Management Development 26 (2007) 9-21.
[4] J. Kamau, Challenges of course development and
implementation in a dual mode institution in Botswana, in:
Pan Commonwealth Forum on Open Learning:
Empowerment through Knowledge and Technology,
Darussalam, Brunei, Mar. 1-5, 1999.
[5] I. Mugridge, Response to Greville Rumble’s article “The
competitive vulnerability of distance teaching
universities”, Open Learning 7 (1992) 59-62.
[6] G. Srikanthan, J. Dalrymple, A Synthesis of a quality
management model for education in universities,
International Journal of Educational Management 18
(2004) 266-279.
[7] J. Frahm, Developing communicative competencies for a
learning organization, Journal of Management
Development 25 (2006) 201-212.
[8] S. Parka, S.T. Oliverb, T.S. Johnsonc, P. Grahamd, N.K.
Oppongb, Colleagues’ roles in the professional
development of teachers: Results from a research study of
National Board certification, Teaching and Teacher
Education 23 (2007) 368-389.
[9] R.M. Pallof, K. Pratt, The Virtual Student, John Wiley &
Sons, San Francisco, 2003.
[10] R.S. Hubbard, B.M. Power, The Art of Classroom
Inquiry, Heinemann, USA, 1993.
[11] E. Saito, P. Hawe, S. Hadiprawiroc, S. Empedhe,
Initiating education reform through lesson study at a
university in Indonesia, Educational Action Research 16
(2008) 391-407.
[12] S.D. Gilbert, How to be a Successful Online Student,
McGraw-Hill, San Francisco, 2001.
[13] Z. Berge, The role of the online instructor/facilitator，
Educational Technology 35 (1995) 22-30.
[14] N.W. Coppola, S.R. Hiltz, N.G. Rotter, Becoming a
virtual professor: Pedagogical roles and asynchronous
learning networks, Journal of Management Information
Systems 18 (2002) 169-189.
[15] P.C. Lim, P.T. Cheah, The role of tutor in asynchronous
discussion boards: A case study of a pre-service teacher
course, Education Media International 40 (2003) 33-47.
[16] D. Maor, The teachers’ role in developing interaction and
reflection in an online learning community, Education
Media International 40 (2003) 127-137.
[17] D. Silverman, Doing qualitative research, SAGE, London,
2000.
[18] H. Altrichter, P. Posch, B. Somekh, Teachers Investigate
Their Work, Routledge, London, 1993.
Journal of Communication and Computer 10 (2013) 1105-1113

Coordination in Competitive Environments

Salvador Ibarra-Martinez, Jose A. Castan-Rocha and Julio Laria-Menchaca

The Engineering School, Autonomous University of Tamaulipas, Victoria 87000, Mexico

Received: August 02, 2013 / Accepted: August 19, 2013 / Published: August 31, 2013.

Abstract: Despite several researches in autonomous agents important theoretical aspects of multi-agent coordination have been
largely untreatable. Multiple cooperating situated agents support the promise of improved performance and increase the task
allocation problems in cooperative environments. We present a general structure for coordinating heterogeneous situated agents that
allows both autonomy of each agent as well as explicit coordination of them. Such situated agents are embodied for taking into
account their situation to solve any action. Indeed, organizational features have been used as metaphor to achieve highest levels of
interactions in an agent system. Then, a decision algorithm has been developed to perform a match between the situated agent
knowledge and the requirements of an action. Finally, this paper presents preliminary results in a simulated robot soccer scenario
showing an improvement of around 92% between the worst and the best cases.

Key words: Multi-agent coordination, e-institutions, interactive norms, soccer robotics.

1. Introduction way and under a wide-range of conditions or

constraints. In fact, an agent system will have to be
Coordination depends on how autonomous agents
handled with a great level of awareness because the
make collective decisions to work jointly in real
failure of a single agent may cause a total degradation
cooperative environments [1]. Nowadays, several
of the system performance. For thus, this paper aims
researchers have proposed that autonomous agent
to introduce a decision algorithm based on the e-I
systems are computational systems in which two or
(electronic Institution) features [3], which it represents
more agents work together to perform some set of
the rules needed to support an agent society.
tasks or satisfy some set of goals. Research in
Specifically, such algorithm uses knowledge of the
multi-agent systems is then based on the assumption
agent situation regards to three perspectives:
that multiples agents are able to solve problems more
interaction with social information and other relevant
efficiently than a single agent does [2]. Special
details to entrust in other agents or humans; awareness
attention has been given to MAS developed to operate
representing the knowledge of the physical body
in dynamic environments, where uncertainty and
reflecting in the body’s skills; and world including
unforeseen changes can happen due to presence of
information perceived directly from the environment.
other physical representation (i.e., agents) and other
But each type of agent reacts to its perception of the
environmental representations that directly affect the
environment in different ways, modifying the overall
agents’ decisions. Such coordination allows agents to
system performance. In particular, a match function
reach high levels of interaction and increase their
has been formulated to reach a suitability rate based
successful decisions, improving the performance of
on the situated agents’ capabilities and the actions’
complex tasks. Agents must therefore work in some
requirements. In fact, agents can select those actions
for which they are the best qualified. The
Corresponding author: Salvador Ibarra-Martinez, doctor, effectiveness of this work is illustrated by developing
research fields: intelligent systems, autonomous robots and
coordination algorithms. E-mail: sibarram@uat.edu.mx. several examples that analyze cooperative agents’
1106 Coordination in Competitive Environments
behavior considering different situations in a
cooperative environment. Section 2 introduces
formal coordinated structure introduced in
approach. In Section 3 an example of
implementation is presented. Finally, the results
conclusions are showed in Section 4.
2. Our Approach
A group of situated agents are here presented as
cooperative systems constituted by a group of
autonomous agents who must cooperate among
themselves in order to reach specific goals within real
cooperative environments. When agent interaction
exists, each element of the agent group must be able
to be differentiated from the others. These agents
require a sense of themselves as distinct and
autonomous individuals obliged to interact with others
within cooperative environments (i.e., they require an
agent identification) [4]. This identification refers to
the property of each agent to know who it is and what
it does within the group. In this sense, this work
proposes two agent classifications: CA (coach agents)
and SA (situated agents).
2.1 Adopting e-Features
In order to imitate the ideology of the e-I (i.e., e-I
uses a set of rules to manage the action performance
in groups of agents), the paper describes how agents
that work in temporal groups are able to achieve
collective behaviour. Such behaviour is possible by
using communication among agents. Let us suppose a
scene sα as a spatial region where a set of actions must
be performed by a group of situated agents sα.
 s i , s j  S | si  s j where S  {s1 , s2 , s3 ,..., sn }
S is the set of all possible Scenes.
Let us define a coach agent caα in charge of
supervising the execution of the actions in a particular
sα.
 cai , ca j  GCA | cai  ca j  GCA  CA
where CA  {ca1 , ca2 , ca3 ,..., can }
real where CA is the set of all possible supervisor agents.
the When saα has identified its s, it must claim
this information in order to know which actions must be
the achieved in such s. It is possible, then, to define a saα
and as sensitive to the events that happen in real
cooperative environments based on the agent
paradigm [5].
Let us define a situated agent sai as an entity that
has a physical representation on the environment and
through which the systems can produce changes in the
world.
 sai , sa j  GSA | sai  sa j  GSA  SA
where SA  {sa1 , sa2 , sa3 ,..., san }
SA is the set of all possible situated agents.
In this sense, sai could be represented in many ways,
(i.e., one autonomous robot with arms, cameras,
handles, etc.) but for the scope of our proposal; sai is
embodied as an entity which is characterized by the
consideration of three parameters: interaction,
awareness and world.
In fact, the paper argues coordination at two
meta-levels (cognitive level—supervision of the
intentions; physical level—execution of the action in
the world, Fig. 1), where the coach agent coordinates
among them to allocate of a set of actions for a group
of situated agents.
Let us define a norm ni that is denoted as a rule that
must be respected or must fix the behavior that a sai
must keep at trying to perform an action in a sα. We
indicated the conception of a norm within a scene
Fig. 1 Levels of interaction.
 Coordination in Competitive Environments
following a set of rules such that:
if (ni) do/dont {action}
 ni , n j  N ( s ) | ni  n j  N ( s )  N
where N ( s )  {n1 , n2 , n3 ,..., na }
Let us define an obligation obl as the imposition
given to some sai to perform some action, which it is
established following a set of rules. In order to denote
the notion of obligation obl the predicate [3] is present
obl ( pai , , s )
as follows:
where a sai is obligated to do  in sα.
2.2 Cooperative Actions
Studies about which actions are involved in
determine scene are needed to perceive knowledge
that make possible the organization of any determined
scene. Once a coach knows in which scene it will
develop its function, it must identify the goals to be
accomplished in such spatial region, indicate the tasks
that must be performed to achieve these goals, and
what roles are necessary for the task achievement.
Then, a coach is defined in its knowledge base
KB(caα) by the consideration of a set of goals G, a set
of tasks T and a set of roles R.
KB(ca )  G ( s )  T ( s )  R( s )
where KB ( sa ) is the information of all the issues
regarding to a specific scene sα, such that: G ( s ) is
the set of goals, T ( s ) is a set of all tasks, and R( s )
is the set of all roles involved in determined scene sα.
Indeed, it is necessary to propose a priority index pi
that represents the importance of every action. A saα
will know both the order in which the goals and the
tasks must be performed and the order of the role
allocation process regarding its supervised sα. Such
priority index will be established according to system
requirements (i.e., timeline) in order to achieve the saα
aims.
Goals then embody the overall system purpose;
however, a caα could achieve a particular goal without
the necessity of performing another goal at the same
sα.
1107
 g i , g j  G ( s ) | g i  g j  G ( s )  G
where G ( s )  {g1 , g 2 , g 3 ,..., g o }
 g i  G ( s )  p( g i )  PG ( s ) | 0  p( g i )  1
where G is the set of all possible Goals and G ( s  ) is
g involved in sα.
Let us to define a set of tasks T which represent the
issues that must be performed to achieve a specific g .
Goal then could be achieved without the implicit
necessity of performing all its involved tasks.
Therefore, the tasks selected are independent, but their
development could affect in a positive or negative
way the development of other tasks.
 ti , t j  T ( g  ) | ti  t j  T ( g  )  T ( s )  T
where T ( g  )  {t1 , t 2 , t3 ,..., t p }
 ti  T ( g  )  p (ti )  PT ( g  ) | 0  p (ti )  1
where T is the set of all possible Tasks.
Let us define a set of roles R which represent the
actions that a pai must fulfil to perform a t within a sα.
 ri ,r j  R(t ) | ri  r j  R(t )  R( s )  R
where R(t )  {r1 , r2 , r3 ,..., rq }
 ri  R(t )  p(ri )  PR (t ) | 0  p(ri )  1
where R is the set of all possible Roles.
In order to illustrate how this process is performed,
let us suppose a scene s1 which is supervised by the
coach ca1 performing a decision process to define
which goal must be attended firstly (Fig. 2).
2.3 Embodying Situated Agents
Supposing that a situated agent lives in a real
environment, therefore, it has the ability to consider
Fig. 2 The coach ca1 defines which goal must be
performed first.
1108 Coordination in Competitive Environments
its physical representation in such world. Although
these characteristics could supposedly take a lot of
“things” regarding the environment our proposal takes
three kinds of knowledge that seek to reference all the
information that characterize the perception of
particular sai.
2.3.1 Interaction
Interaction I refers to the certainty that an agent
wants to interact with other agents to assume a
specific behavior with successful and high reliability
to achieve any action proposed within any determined
scene. Such information is useful in the interaction
process of the agents because they can trust in other
agents based on the result of their previous
interactions. Obviously, if a sai has a positive
performance of its actions, its interaction level
increases; but if the outcome of the action is negative,
its interaction level decreases. Such knowledge is
obtained when a sai has a direct relationship with a
caα.
 sai  GSA  I r ,s ( sai )  I ( sai )
where I r ,s ( sai ) is the interaction level of a sai to
perform rγ in the sα.
2.3.2 Awareness
Awareness A refers to the set of physical
self-knowledge that a physical agent has represented
about its skills and physical characteristics to execute
any proposed action. Such physical representation is
considered as the embodiment of the physical features
that constitute all the information that physical agents
can include in their decision-making.
Physical agents could be any physical objects
“handled” by an intelligent agent (i.e., an autonomous
robot, a machine or an electric device). Such pai has
features that consider their physical body properties
(i.e., their dynamic, their physical structure) usually
when they commit to perform some task or to assume
a specific behaviour within a cooperative environment.
This fact represents the skill of the physical agents to
know that actions will be performed based on the
knowledge of the physical agents’ bodies, which is
achieved through representation of them on a
capabilities basis.
 pai  PA  At ,s ( pai )  A( pai )
where At ,s ( pai ) is the Awareness of pai to perform
t in the sα.
2.3.3 World
World W refers to the set of environmental
knowledge that physical agents have to perform the
proposed set of actions. Such domain representation is
considered as the embodiment of the environment
knowledge that represents all the physical information
that has influence in the physical agents’ reasoning
process.
Let us define a set of world conditions that
represent information about empirical knowledge of
the environmental state, such that:
 pai  PA  Wt ,s ( pai )  W ( pai )
where Wt  ,s (pa i ) is the environmental condition of
pai to perform t in a sα; saα uses the above
information to know the physical situation of each pai.
All knowledge of a particular pai KB(pa i ) is then
constituted by the information provided for the three
modules, such that:
 pai  KB( pai )  [ I ( pai ) A( pai ) W ( pai )]
In particular, all knowledge related to a specific t
in sα is given such that:
KB( pai ) t ,s  [ I t ,s ( pai ) At ,s ( pai ) Wt ,s ( pai )]
2.4 Communication Process
The humans have a communication process that
allows transmit information or ideas in a common
language to make sure and reliable commitments
between us. Likewise, artificial intelligence has
several approaches showing the same process [5-6] to
exploit the advantages of expressing communication.
To accomplish an action, a group of agents must
establish communication (to coordinate them). On
such coordination agents must “converse” among
them to agree who is who within the group (Fig. 3).
Then, a communication with three simple dialogues
 Coordination in Competitive Environments 1109

based on the KQML specification is presented as values that establish the relevance of each parameter

Request( sa , sa  , , sn )
follows: related to a sα. These values are in the range [0, 1]. In
this sense, the sa responsible in s uses the
where saα asks to sa its θ in the scene sα KB ( pai ) t ,s and the ID(s ) to perform a match
Reply( sa  , sa , )
where sa responds to saα its decision  based on the
 ID(s )
function by means of Eq. (1).
 KB( pai ) t ,s
3

Inform( sa , sa ,  , s )

information dispatched.
match(ID(s ), KB( pai ) t ,s ) 
j 1
( j)

(1  ID(s )
( j)
(1)

where saα informs sa its state  in the scene sα. 3

3

( j) )
j 1
This process helps to the saα to communicate
A sa uses the match to determine which pai must
among them and with a pai.
perform rq in a s, assigning the higher pai for the
Otherwise, some concepts have been explained
most prior rq in s. In addition, Fig. 4 shows an
throughout this research work, but none of them has
example of the match process.
clarified how a saα could decide who is the pai (or
group) that will take part in the action of its
responsible sα. saα then considers an ID (influence
degree) to all these actions involved in a sα by the
tupla ID(sα) based on the consideration of the
aforementioned parameters to generate an utility
function that helps them in their decision making
structure.

ID(s )  [id EC (s ) id PK (s ) idTV (s )]

where id EC (s ) , id PK (s ) and id TV (s ) are Fig. 3 Conversation between the sa1 and sa2.

Fig. 4 Empirical example of a match process.

1110 Coordination in Competitive Environments

2.5 Decision Algorithm amount of PA, such as R(sα). Suppose that the system
has enough amount of PA to take all the defined roles.
An important criterion for the development of
To know, every saα is able to exclude a pai that
collective actions within real cooperative
presents a lowest action capability.
environments is the traffic of the information available
Stage 4. Show-time. A pai knows the rq that must
from the perception of the intentions to the execution
perform. This involves physical changes in the
of them. We have therefore determined a particular
environment. Now, the environment has been
decision algorithm of four simple stages.
modified. So, a new consensus among the SA could be
Stage 1. Refers to the property of a saα to perceive
performed to adjust it to the current changes in the
which sα must manage, therefore, a saα then knows its
environment.
goals, tasks, roles (the priority of every item is also
perceived) and ID involved in its sα. Hence, the 3. Implementation
knowledge base of each saα could be achieved.
In our implementation, each physical agent has a
Stage 2. All the sa (of the entire SA) must organize
different movement controller which differentiates
them to define which will be the order in that they
from others. Then, we have segmented the scenario
could begin the recruitment of pa to perform the
into three spatial regions (Fig. 5) to represent each sα.
actions within its sα. For thus, the sa must converse
For sake of simplicity, we only have defined one
among them using the developed dialog.
goal per scene G(s1) = g1; G(s2) = g2 and G(s3) = g3.
Stage 3. Based on the order obtained above, a saα is
The consensus to define the execution order of the
approved to start the communication with the entire
scenes is derived as as shown in Fig. 6.
PA to determine that pai will be the selected to
The cbp is the current ball position on the
perform every action. For thus, a saα must obtain the
environment. So, the spatial regions are limited
physical knowledge of each pai by means of directly
according to the simulator dimensions (axis x: [0 220];
communication with them; the environment
conditions and trust value of each pai are obtained
when the saα uses the modules aforementioned
(respectively for each parameter).
Once a saα completes the KB(pa i ) of the entire PA,
it takes such information to perform the match using
the Eq. (1), considering the priority index of all the
roles. Then, saα has a list detailed (form higher to
lowest coefficient) of the entire PA. After, saα knows
that pai must perform that role; therefore, it is able to
obligate a determined pai to perform a role which
Fig. 5 Geographic segmentation of the experimental
represents that action must be performed.
environment.
Hence, the best pai (of the entire PA) will choose
the prior role to perform and then others successively,
until all the roles finish in such sα. Such process
guaranty allocates us a suitable role because the rq
always allocated to the best pai. Indeed, a saα knows
how many PA needs because it needs the same Fig. 6 Supervisor agent consensus.
 Coordination in Competitive Environments 1111

axis y: [0 180]). Moreover, specific tasks are defined in s respectively and  is from 1 to Q( s ) and ω is
in order to accomplish each gi such that: from 1 to Q' ( s ) ; that are the number of awards and

T(g 1 )  {t1 , t 2 }  T ( g 2 )  {t 3 , t 4 }  T ( g 3 )  {t 5 ,t 6 }
where t1 is make-pass, t2 is shooting, t3 is player-on, t4
is kick-ball, t5 is protect-ball and t6 is covering a
position.
Following the rule presented for the goals, the tasks
also use the cbp as a reference to determine its
execution order.
Then, using the ranges above, a saα may decide the
task to perform at any time. But, to attempt to achieve
such tasks a saα must define which roles it must
perform and the priority order of such roles. Therefore,
by means of human analysis we have proposed four
punishments in s.
Awareness here called Physical Knowledge, PK
represents the knowledge of the agents about their
physical capabilities to perform any proposed task. In
particular, the introspection process is performed by
using neural networks taking into account the
knowledge that a pai has related to perform t in sα.
PK t ,s ( pai )  [0,1]
Consider that a high by
representing a suitable pai.
World here called Environmental Conditions, EC is
a value related to the distance between the current
location of a pai and the location of the ball. Eq. (4)

ecty , sa ( pai )  (1  d ( pai, r (ty, sa ))

roles that could be used to perform any task such that: shows the calculation:

R(t  )  r1 , r2 , r3 ,r 4 }
/ d max(( sa )) ecty , sa ( pai )  [0, 1]
(4)

where, r1 goes to the ball, r2 kicks the ball, r3 covers a where ect ,s ( pai ) is the value of a pai to perform a
zone and r4 takes a position to be used in each t.

t in s; d ( pai , r (t , s )) is the distance between the

In addition, we have performed a combination with
the information involved in the environment-based
knowledge. Such combination is used by saα to
pai with r (t , s ) and d max(s ) is the maximal
perform the match process considering the
aforementioned parameters. Then, a binary distance of all pa in s. Then, Eq. (5) shows the
combination lets us generate eight influence degrees d max(s  ) calculation where m is the total number of

d max( sa )  max(d (1, sa ),..., d (m, sa )) d max  [0, 1]

(Table 1). We present a review to show how we have pa in IAS.
implemented these parameters in the robot soccer
testbed. (5)
Interaction here called Trust TV represents the In order to show how our approach performs the
social relationship among agents taking into account role allocation process we present a possible situation
the result of past interactions of a sa with a pai. Eq. (Fig. 7) where the ball is within the s2 and we use all
(2) shows the trust calculation if the aim is reached. Table 1 Influence degree consideration (0: is not
Otherwise, using Eq. (3) shows the trust calculation if considered; 1: is considered).
Influence degree TV PK EC

tvt ,s ( pai )  tvt ,s ( pai )  A( s ,  ) (2)

the aim is not reached.
ID0 0 0 0

tvt  , s ( pai )  tvt  , s ( pai )  P ( s ,  ) (3)

ID1 0 0 1
ID2 0 1 0

where the tvt ,s ( pai )  [0,1] and higher tvt ,s ( pai )
ID3 0 1 1
ID4 1 0 0

represents the best pai to perform t in s, A( s ,  )

ID5 1 0 1
ID6 1 1 0
and P(s  , ) are the awards and punishments given ID7 1 1 1
1112 Coordination in Competitive Environments

Exp. 2, a league of 28 games was performed to

confront the IAS among them. So, the IAS
performance increases when using jointly all the
parameters. In fact, the IAS(ID7) shows a better
average (improvement rate: +92%) than IAS(ID0).
As conclusions we argue the need of agent
meta-coordination to exploit the advantages of the
abstract environment knowledge (by the supervisor
agents) and use it to influence the reasoning process
of the physical agents.
Fig. 7 Possible situation for the PA in the environment. In addition, a combination (named Influence
Degree) describes the consideration among these
the influence degrees generated to perform the pa parameters giving to the sa the ability to determine a
selection. Then, we only showed the allocation for one decision process to perform a match between the
action (kick the ball). In Table 2 we present the values scene requirements and the physical agent capabilities.
of a pai regarding to the proposed action. In Table 3 In fact, the best performance is obtained when our
we show the match values obtained by means of Eq. team agent took into account all the parameters in its
(1). Then, is possible to see it will be the pai selected decision process. But it is really interesting to analyze
by the sa2 to perform the proposed action.
Table 2 Physical agents’ knowledge bases.
Additionally, the remained physical agents follow a
pa Trust Intro. Prox.
fix strategy which was defined to consider actions to
KB(pa1)tkickball,s2 0.43 0.47 0.31
the entire PA.
KB(pa 2)tkickball,s2 0.65 0.52 0.46
4. Results and Conclusions KB(pa 3)tkickball,s2 0.71 0.69 0.79
KB(pa 4)tkickball,s2 0.83 0.77 0.63
We ran two experimental evaluations to validate the
proposed approach. In particular, in the experiments Table 3 Some examples of physical agent selection.
our IAS uses all the binary combination of the ID to ID(s2) pa1 pa2 pa3 pa4 pa Selected
perform the match process. In Exp. 1, our IAS ID1(s2) 0.31 0.46 0.79 0.63 pa3
ID2(s2) 0.47 0.52 0.69 0.77 pa4
competed against a blind opponent in 30 games. Here,
ID3(s2) 0.39 0.49 0.74 0.70 pa3
the IAS performance is improved when all the ID4(s2) 0.43 0.65 0.71 0.83 pa4
parameters are considered. So, IAS(ID7) shows a ID5(s2) 0.37 0.55 0.75 0.73 pa3
better average (improvement rate: +81% better) than ID6(s2) 0.45 0.58 0.70 0.80 pa4
IAS(ID0) (any parameter considered). Then, in the ID7(s2) 0.40 0.54 0.73 0.74 pa4

Table 4 Our approach vs. other approaches.

ID T I P VS
0 0 0 0 References take at least one of these parameters.
1 0 0 1 Not references yet.
2 0 1 0 [4-7]
3 0 1 1 [8-10]
4 1 0 0 No references yet.
5 1 0 1 [11-13]
6 1 1 0 [14]
7 1 1 1 [15]
 Coordination in Competitive Environments
how the cooperative IAS performance increases when
the system takes the parameters into consideration. In
conclusion, the situation matching approach is a
promising method to be used as utility function
between task requirements and physical agent
capabilities in MAS.
In Table 4 we show some approaches regarding
architecture for multi-agent cooperation. In particular,
these architectures express behavior by implementing
different kinds of knowledge, which can be related to
our approach.
References
[1] S. Ibarra, C. Quintero, J. A.Ramon, J. Ll de la Rosa, J.
Castan, PAULA: Multi-agent Architecture for
coordination to intelligent agent systems, in: Proc. of
European Control Conference (ECC’07), Kos, Greece,
July 2-5, 2007.
[2] D. Jung, A. Zelinsky, An architecture for distributed
cooperative-planning in a behaviour-based multi-robot
system, Journal of Robotics & Autonomous Systems
(RA&S) 26 (1999) 149-174.
[3] M. Esteva, J.A. Rodriguez, C. Sierra, J.L. Arcos, On the
formal specification of electronic institutions, Agent
Mediated Electronic Commerce Lecture Notes in
Computer Science 1991 (2001) 126-147.
[4] B. Duffy, Robots social embodiment in autonomous
mobile robotics, Int. J. of Advanced Robotic Systems 1
(2004) 155-170.
[5] S. Russell, P. Norving, Artificial Intelligence: A Modern
Approach, 3rd ed., Ed. Prentice Hall, London, Dec. 2009,
p. 1152.
[6] M. Luck, P. McBurney, O. Shehory, S. Willmott, Agent
Technology: Computing as Interaction (A Roadmap for
Agent Based Computing), AgentLink, 2005.
[7] A. Oller, DPA2: Architecture for co-operative dynamical
1113
physical agents, Doctoral Thesis, Universitat Rovira I
Virgil.
[8] C.G. Quintero, J. Ll. de la Rosa, J. Vehi, Self-knowledge
based on the atomic capabilities concept—A
perspective to achieve sure commitments among physical
agents, in: 2nd International Conference on Informatics in
Control Automation and Robotics, Barcelona, Spain, Sep.
14-17, 2005.
[9] L. Pat, An adaptative architecture for physical agents, in:
IEEE /WIC/ACM International Conference on Intelligent
Agent Technology, Sep. 19-22, 2005 pp. 18-25.
[10] D. Busquets, R. Lopez de Mantaras, C. Sierra, T.G.
Dietterich, A multi-agent architecture integrating learning
and fuzzy techniques for landmark-based robot
navigation, Lecture Notes in Comp. Science 2504 (2002)
269-281.
[11] C.G. Quinero, J. Zubelzu, J.A. Ramon, J. Ll. de la Rosa,
Improving the decision making structure about
commitments among physical intelligent agents in a
collaborative world, in: In. Proc. of V Workshop on
Physical Agents, Girona, Spain, Mar. 25-27, 2004, pp.
219-223.
[12] R.S. Aylett, D.P. Barnes, A Multi-robot architecture for
planetary rovers, in: 5th ESA Workshop on Advanced
Space Technologies for Robotics and Automation,
ESTEC, Noordwijk, The Netherlands, Dec. 1-3, 1998.
[13] R. Simmons, T. Smith, M. Bernardine, D. Goldberg, D.
Hershberger, A. Stentz, et al., A layered architecture for
coordination of mobile robots, in: Multi-robot Systems:
From Swarms to Intelligent Automata, May, 2002.
[14] C.G. Quintero, J.L. de la Rosa, J. Vehi, Physical
Intelligent Agents’ Capabilities Management for Sure
Commitments in a Collaborative World, Frontier in
Artificial Intelligence and Applications, IOS Press, 2004,
pp. 251-258.
[15] S. Ibarra, C. Quintero, J. Ramon, J.L. De la Rosa, J.
Castan, Studies about multi-agent teamwork coordination
in the robot soccer environment, in: Proc. of 11th Fira
Robot World Congress 2006, 2006, pp. 63-67.