INTERVAL BASED INTEGRATED REAL-TIME COORDINATION

FOR MULTI-AGENT SYSTEMS

Ghulam Mahdi and Abdelkader Gouaich

LIRMM, Univ. Montpellier 2, UMR 5506 - CC 477, 161 Rue Ada, 34095 Montpellier Cedex 5, France

Keywords:

Coordination, Real-time, Interval algebra, Games.

Abstract:

Real-time computations in agent based simulations and (serious) games possess an inherent element of tempo-

ral relationships as well as time constraints for their performance and utility measures. Such time relationships

and temporal constraints can be observed in individual agent behaviors as well as coordination process involv-

ing multiple agent. The temporal relationships and time constraints in multi-agent coordination come in terms

of message passing, resource management and negotiations. The idea behind such temporal relationships and

time constraints is to efﬁciently handle complex interactions as different patterns of coordination emerge as

per the updated situations under certain time durations. Here we propose our position about integrating both

dimensions of individual and collective coordination in a uniﬁed manner where the coordination patterns are

expressed through Allen’s interval paradigm. We also introduce concept of “timers” to ensure real-time with

explicit expressiveness of the interval paradigm.

1 INTRODUCTION

One of important issues in multi-agent systems

(MASs) based games and simulation is to deal with

the interaction of the user(s)’s actions at one end

while keeping track of internal coordination of the

game/simulation objects. In most of the situations,

users need to interact with an application with little

care of number, time or resource availability of partic-

ipating game/simulation characters. However, if these

aspects of interaction are taken for granted, the results

may appear in either irrelevance to the real situation

or very poor user interaction.

The problem becomes more clear when we visu-

alize evolution of user actions and games objects’ in-

ternal interactions. Different scenarios emerge as the

system progresses as a result bringing complex pat-

terns of coordination. The games need to introduce

increasing complexity, maintain user interaction with

the game and progressively difﬁcult scenarios to keep

up the user’s attention. Although all these aspects

need to be carried on throughout the game but deﬁn-

ing every user, game interaction brings a quite cum-

bersome and bulky task for the game designers. Here

we are particularly interested in addressing coordina-

tion in agent-based simulations and (serious) games in

a way that complex patterns of temporal coordination

are expressed through very minimal primitive time

relationships between two events. We suggest Inter-

val script paradigm (Pinhanez et al., 1997) based on

Allen’s Algebra (Allen, 1984) to express temporal re-

lationships in real-time coordination for agent-based

(serious) games and simulations. Here we argue that

it is not only possible to express all temporal aspects

of coordination through a limited set of time relation-

ships but it may bring the “real-time” phenomenon

for these application domains by introducing “timer”

mechanisms.

The rest of article is structured as follows: the

ﬁrst section discusses some motivations from earlier

works and real world scenarios for investigating co-

ordination issues; then the following section is about

characterizing some key aspects of coordination and

inﬂuence of “real-time” on them. We then describe

integrated approach of real-time coordination; then a

basic discussion on interval paradigm’s possible us-

age in the integrated real-time coordination is pre-

sented and ﬁnally, we conclude the article by a con-

clusion and future works of this study.

2 MOTIVATIONS

Real-time systems and multi-agents systems have in-

dividually contributed to many complex, heteroge-

664

Mahdi G. and Gouaich A..

INTERVAL BASED INTEGRATED REAL-TIME COORDINATION FOR MULTI-AGENT SYSTEMS.

DOI: 10.5220/0003188406640669

In Proceedings of the 3rd International Conference on Agents and Artiﬁcial Intelligence (ICAART-2011), pages 664-669

ISBN: 978-989-8425-40-9

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

neous and diverse real-world applications even before

joining hands to be applied in domains particularly

known for distributed, time-critical and autonomous

features. A transfusion of both disciplines has shown

quite interesting results in diverse domains ranging

from sensor networks (Sierra and Sonenberg, 2005)

and (Soh and Tsatsoulis, 2005) to virtual class rooms

(Liu et al., 2003) and from e-commerce applications

(DiPippo et al., 2001) to soccer robots (Kim et al.,

1997). Apart from these applicative studies of multi-

agent and real-time systems, many works have tried

to deﬁne, develop and implement efﬁcient agent mod-

els presenting features of both disciplines (like (Julian

and Botti, 2004) and (Julian et al., 2004)) Some oth-

ers have focused on frameworks simulating such real-

time agents (Micacchi and Cohen, 2008); even others

align to developing joint architectures for such type

of agents (Soler et al., 2002) and (Carrascosa et al.,

2008)

However a fundamental issue of coordinating

MASs under real-time constraints has remained

largely unaddressed, even if some of works which

have tried to address the agent coordination are lim-

ited to focus on only one aspect of coordination in-

stead of addressing the issue in an integrated manner.

Our motivations to address the problem has theoreti-

cal as well applicative inspirations. From the perspec-

tive of applications, most of the real-time MASs are

functioning in resource-constrained environments.

In agent-based (serious) games and simulations

often situations presume a real-time behavior in game

and simulation objects’ behavior as well their inter-

action, communication and resource sharing. Taking

a simple example of a war jet in a game, it assumes

that when it is launched it has real-time perception of

enemy attacks, devising updated strategy of further

plan (reasoning), receiving and delivering message

instantaneously(message passing), sharing air space

with other fellow pilots (resource sharing) and abil-

ity for collected decision making (reaching an agree-

ment as a result of negotiations) all at once. Hereby,

addressing only one aspect of temporal constraints

may not address the overall issue but probably bring-

ing serious challenges to the performance and util-

ity for the users. Although the problem seems quite

challenging one and addressing all aspects of indi-

vidual agent performance and interaction with other

agents at each instant of time even unrealizable but

here we suggest a mechanism which introduces con-

straints on particular time relations. The time interval

relationships phenomenon introduced by James Allen

in (Allen, 1984) suggests introducing temporal rela-

tionships based on 13 primitive relations. These time

relations like a description based on what to do af-

ter/before/meet another interval. We ﬁnd this line of

research quite interesting for agent based games and

simulations due to simplicity, explicit expressiveness

and generalization of the interval algebra. As the for-

malism is based on Allen Algebra’s 13 primitive time

relationships so almost all temporal aspects of coordi-

nation are expressed through primitive relationships;

similarly time relationships’ explicit characterization

brings a clear picture of the coordination at any par-

ticular instant; ﬁnally, as we per our fundamental the-

sis that all aspects of temporal constraints need to

be addressed under one roof, here this model fulﬁlls

the condition in a very general manner. This “posi-

tion” nature of our work examines agent coordination

studies at different levels and suggests an integrated

as well comprehensive framework for efﬁcient “real-

time coordination”.

Having a brief idea about the implications of “time

factor” in agent coordination, now we individually de-

scribe the factors involved in amelioration of “time

factor” in agent computations as argued by Mahdi et

al. in (Mahdi et al., 2010). By description of these

“performance contributors”, we argue that these pro-

cesses may be viewed as constituting aspects in global

agent coordination process for an integrated view of

“real-time coordination” in MASs. Here we describe

different studies on these “performance contributors”

and their role in integrated and comprehensive under-

standing of “real-time coordination” of MASs.

3 TIME NOTION UNDER

DIFFERENT ASPECTS OF

AGENT COORDINATION

3.1 Time in Message Passing

Assuming agents process incoming messages atom-

ically as soon as they receive them (or buffered in

the message inbox), we need to take care of how

much time it takes to deliver a message. Having tim-

ing constraints on the delivery of messages may play

a substantial role in managing temporal behavior of

the overall system. Embracing monitors that ensure

timely dispatch of the messages do not have to come

in conﬂict with the timing constraints in message pro-

cessing in a way that message delivery is not to breach

the agent encapsulation of how and when the message

is processed, rather it’s sole concern would be about

timely delivery of the incoming messages.Jamali et al.

(Jamali and Ren, 2005) suggest similar approach for

MASs in resource allocation. Although the approach

works quite nicely, but its performance improvement

INTERVAL BASED INTEGRATED REAL-TIME COORDINATION FOR MULTI-AGENT SYSTEMS

665

is limited to resource management, in other words,

it covers only one aspect of real-time performance in

MASs.

3.2 Time in Agent Reasoning

Once a message is passed to the concerned agent, it

may take time some time to read the message, eval-

uate the contract content and subsequently reply in

denial or follow the message contents. If time factor

is not involved in such message processing or agent

reasoning it would unnecessarily affect the agent per-

formance resulting in delay of overall coordination

process. Many works of Julien et al. (like (Ju-

lian and Botti, 2004), (Julian et al., 2004) and (Soler

et al., 2002)) are addressed on development, design

and implementation of real-time agents without con-

sidering coordination as the main subject of studies.

Here we need to make such models enough ﬂexible

with other real-time computations like the above men-

tioned ones. In absence of timing constraints the sys-

tem processes the messages may take too much time

and leading to affect the overall progress of the sys-

tem.

3.3 Time in Resource Management

Agents being part of open systems compete for re-

sources due to sharing of independent computations.

Such competition to acquire resources leads to func-

tional and non-functional dependencies. Functional

dependencies are about whether sufﬁcient resources

are available or not, how to acquire and release cer-

tain resources and how to deal with multiple requests

of the same resource at the same time. By non-

functional dependencies, we mean that availability (or

at least information of unavailability) of the required

resources in certain time bounds. Such availability

or unavailability information would be seen as an im-

portant factor in overall agent coordination. Here we

need to manage autonomy of agents in a way that

agents are not to be let to accumulate all the resources

so here some type of resource management behavior

is also recommended. Jamali et al. (Jamali and Ren,

2005)’s work on real-time resource allocations is sem-

inal on the subject that it not only ensures real-time

in resource allocations but also handles excessive re-

source acquisitions problem common to agents based

on actor model.

3.4 Time in Negotiations

Although agent coordination doesn’t imply cooper-

ation but many times coordination is seen as a co-

operative process to maintain heterogeneous body of

agents in an environment. Agent negotiations are

used as a means to reach an accord through commu-

nications. Agent negotiations are usually seen as a

compromising tool to mutual beneﬁts of efﬁcient re-

source usage and task distribution. Despite beneﬁts

of reaching an agreement, agent negotiations process

is presumed as a costly and time consuming prac-

tice. When agent negotiations are left to work on

their momentum it would not only delay the coordina-

tion process but also consume unnecessary resources.

A model for real-time agent negotiations for sensor

networks is presented in (Soh and Tsatsoulis, 2005),

other important works on the subject include Kraus et.

al(Kraus et al., 1995) and Fatima et al. (Fatima et al.,

2002) but both of these works address negotiations to

be constrained by time rather than directly treating it

as real-time issue.

After a brief introduction of different real-time

mechanisms in MASs, we return to our earlier propo-

sition that coordination should be viewed as a meta-

collection of different aspects based on an integrated

approach covering all aspects of coordination. Here

we brieﬂy discuss our proposition of an integrated

real-time coordination mechanism suggested for fully

realizing real-time through all aspects of temporal be-

havior.

4 INTEGRATED REAL-TIME

COORDINATION FOR

MULTI-AGENT SYSTEMS

Coordination in MASs has been studied from dif-

ferent perspectives of “real-time”, namely reasoning,

message passing , resource management and negotia-

tions. All these perspectives of real-time coordination

can be seen as “aspects” or “dimensions” in overall

agent computation and coordination process. Real-

time distributed computing processes can be viewed

as a composition of aspects (instead of a single co-

ordination process) where each process coordinates

with its comprising components along with other sub-

systems at its stage while being part of the global co-

ordination process. There are two approaches to see

the real-time coordination in MASs:

• Incorporating time constraints on individual pro-

cesses and coordination of any of the individual

process would of course bring amelioration in the

performance of that system but not at the optimum

level.

• Setting up a meta approach of real-time as well as

coordination in a way that the coordination is in-

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

666

volved at all aspects’ level which improves over-

all real-time performance of the system at global

level.

Our vision to see real-time agent coordination dis-

tinct from other coordination mechanisms may prove

useful in understanding both coordination as well as

real-time performance of MASs. Due to the differ-

ences in the architectures and performance measures

of different systems we suggest an integrated treat-

ment of “real-time” problem at the level of each as-

pect. Human societies also adopt coordination mech-

anisms which may involve myriad aspects, at some

extent seem even irrelevant but after all serve a global

purpose. Like an ofﬁce working procedure may adopt

different procedures and aspects for their coordina-

tion and time constraints but after all it serves timely

performance of the main objective. Real-time MASs

have special architectural foundations and design as-

pects current approaches in real-time agent systems

were not set forth with those considerations therefore

there is a performance as well as efﬁciency gap in

effective agent coordination. Earlier studies on the

subject have either dealt scalable agent coordination

(Durfee, 2001) and (Durfee, 2004) or particularizing

coordination for different application domains (like

(Liu et al., 2003) and (DiPippo et al., 2001)). As

per our knowledge we haven’t seen studies on real-

time agent coordination which involves real-time in

all aspects, although similar studies are carried out

in robotics and communication domain from the per-

spective of coordination (Bouroche et al., 2006) and

communication(Mock and Nett, 1999).

5 ALLEN’S ALGEBRA FOR

INTERVAL-BASED

INTEGRATED COORDINATION

Given any situation of integrated coordination differ-

ent agents’ possible temporal interactions can always

be expressed through disjunctions of Allen’s primi-

tive time relationships between two intervals. A major

bottleneck in integrated coordination has been the ex-

pressiveness of the interactions at each instant of time

and the severity of the situation increases with any at-

tempt of describing all possible time points with all

four dimensions. The result of such situation emerges

in such a complexity that it becomes almost impossi-

ble to track interactions even if temporal constraints

are not there. Some of the primary motivations in-

clude unavailability of some strict timers to handle the

constraints (i.e. It is more inclined to timely comple-

tion of agent intervals than interrupting them unneces-

sarily); then there is relatively smooth and convenient

job at the programmers’ end that they have to handle

only the relevant relationship for any time instead of

managing all time instants; also the notion of disjunc-

tion of intervals can be incorporated to deﬁne mul-

tiple paths and interactions in a coordination mech-

anism, which makes it the best candidate mechanism

for any integrated approach; and ﬁnally a programmer

can integrate quantitative aspects of temporal behav-

ior through introducing timers to the qualitative no-

tion of time interval as per our requirements.

Interval script paradigm (Pinhanez et al., 1997)

based on Allen’s Algebra (Allen, 1984) to ex-

press temporal relationships between two inter-

val. The paradigm uses basic thirteen relation-

ships(summarized in Fig. 1) to express all type of

possible temporal relationships among agents. The

temporal relationships provide an explicit declaration

of time intervals between agents at one hand through

limited number of time relationships; on the other

hand it is enough ﬂexible to determine implicit rela-

tionships between different agent (like if agent A is

BEFORE interval agent B and B is BEFORE agent

C, it can infer that interval agent A is BEFORE C).

Based on our proposed integrated approach of coor-

Figure 1: The possible 13 primitive relationships between

two agents (Allen, 1984).

dination we suggest that almost all of coordination

patterns can always be expressed through disjunction

of these these primitive time relationships. Taking a

game object’s coordinating with other game objects

can occur when the object either STARTS or FIN-

ISHES or happens DURING or is EQUAL to the in-

terval when the game object is in movement. In other

words, the time relationships between a game object’s

movement and interaction can be described by the dis-

junction of START,FINISH,DURING,EQUAL. Sim-

ilarly an agent’s behavior, negotiations and resource

INTERVAL BASED INTEGRATED REAL-TIME COORDINATION FOR MULTI-AGENT SYSTEMS

667

management can be expressed through these primi-

tive time relationships as in real occurrence any one

of the relationships can happen at a time. The main

task for the designer to do is determine the relation-

ships between the agent intervals along with deﬁning

corresponding sensing and actuating routines for the

game objects.

5.1 Introducing Timers

Although one of the deﬁning characteristic of the in-

terval paradigm is that it does not require any explicit

speciﬁcation of the interval duration but as in our case

we address real-time issue we need some speciﬁc time

references. Hence, we suggest some timer mecha-

nism for an interval agent in a way that we have a

desired and actual constraints on the intervals as sug-

gested by (Pinhanez et al., 1997). The desired in-

terval can be used for turning the timer on and off

under speciﬁc time constraints; while actual interval

is used for triggering other actions as the timer ex-

pires. The usage of desired constraints ensures time

constraints on the interval agents while actual timers

are about the solution enough generic to handle nor-

mal routines where there is not any involvement of

time constraints.

5.2 Integrated Real-time Coordination

for Agent-based Games and

Simulations

Again taking the earlier mentioned example of a war

jet in a game, we can describe possible time relation-

ships and temporal constraints in a scenario. Con-

sidering space limitations and position nature of the

paper here we provide only few situation where the

interval phenomenon shows its applicability and sim-

plicity to apply:

• When a jet ﬂies, wait for a message from the base

station BEFORE launching any attack.

• When in the air, in case of attack iBEFORE (after)

change the position to x units.

• When an enemy jet MEET your targeted range do

not wait for the message.

similarly all other temporal relationships of agent

reasoning, communication, negotiations and resource

management can be explicitly expressed through a

limited set of 13 primitive time relations; in addition

to that the “timers” can be implemented that in a way

that how many time units to wait BEFORE launching

an attack or any other situation in the game.

One of most difﬁcult aspect of agent behavior and

coordination comes in keeping track of different ac-

tivities of game/simulation objects in the case of in-

tegrated coordination. It is often the case that in an

attempt to address all aspects of coordination the user

is unable to track even a single aspect and in our case

of real-time coordination it becomes even more dif-

ﬁcult. The problem can be handled through the in-

terval approach in a way that there are conditions on

the time relationships rather than time instants how-

ever to maintain the real-time spirit timers ensure all

temporal constraints.

6 RELATED WORKS

Allen’s interval algebra has been used in different

application domains as a means to provide a deli-

cate balance between explicit expressiveness and efﬁ-

ciency of its deductive engine. PNF calculus based on

Allen’s interval algebra developed by Pinhanez and

Bobick(Pinhanez and Bobick, 1996) for deﬁning a

method to propagate occurrence information through

a network of intervals. The PNF calculus based

scripts were used by Pinhanez et al. in (Pinhanez

et al., 1997) for testing a story-based, interactive sys-

tem named SingSong. The system provides reac-

tive and interval script following computer-generated

partners to the human performers. Another important

work using interval script paradigm is handling vis-

itors’ musem intelligent visits intelligently(Pinhanez

and Bobick, 2003). The work handles time intervals

for visitors’ information, but once it discovers that the

visitor has changed its visiting course, modiﬁes its in-

tervals as per the visitors’ latest location.

7 CONCLUSIONS

AND PERSPECTIVES

Here in this position paper, we have tried to un-

derstand the peculiarities of integrated real-time co-

ordination for MASs. We suggest that current ap-

proaches on the subject present some serious con-

cerns for understanding and applying coordination

models in MASs. Here in this paper, we have ar-

gued that real-time coordination has some different

implications when viewed from a global perspective

of “real-time” at different levels of multi-agent com-

putation and coordination. We are of the view that,

in order to support efﬁcient coordination mechanisms

for real-time MASs, we need to understand the key

differentiating factors that make real-time coordina-

ICAART 2011 - 3rd International Conference on Agents and Artificial Intelligence

668

tion different in MASs. We have discussed different

aspects of “real-time coordination” in MAS and sug-

gested an interval based paradigm to address the par-

ticular related issues. Finally, we suggest “timers” to

introduce real-time in interval-based integrated real-

time coordination.

The approach discussed here would let both MAS

and real-time communities to see each other’s re-

quirements and prospectus in their domains. More

precisely, the agent community to see coordination

in MASs deal differently than it has been and the

real-time community to take a more realistic picture

about the agents’ functionality and effectiveness in

MASs. Clearly there is much left to be done. As a

future work, we plan to work on a formalism based

on Allen’s interval algebra for our suggested approach

for integrated real-time agent coordination.

REFERENCES

Allen, J. (1984). Towards a general theory of action and

time. Artiﬁcial intelligence, 23(2):123–154.

Bouroche, M., Hughes, B., and Cahill, V. (2006). Real-

time coordination of autonomous vehicles. In IEEE

Intelligent Transportation Systems Conference, 2006.

ITSC’06, pages 1232–1239.

Carrascosa, C., Bajo, J., Julian, V., Corchado, J., and Botti,

V. (2008). Hybrid multi-agent architecture as a real-

time problem-solving model. Expert Systems with Ap-

plications, 34(1):2–17.

DiPippo, L., Fay-Wolfe, V., Nair, L., Hodys, E., and

Uvarov, O. (2001). A real-time multi-agent system ar-

chitecture for e-commerce applications. In isads, page

357. Published by the IEEE Computer Society.

Durfee, E. (2001). Scaling up agent coordination strategies.

Computer, 34(7):39–46.

Durfee, E. (2004). Challenges to Scaling-Up Agent Coor-

dination Strategies. An Application science for multi-

agent systems, pages 113–132.

Fatima, S., Wooldridge, M., and Jennings, N. (2002). Multi-

issue negotiation under time constraints. In Proceed-

ings of the ﬁrst international joint conference on Au-

tonomous agents and multiagent systems: part 1, page

150. ACM.

Jamali, N. and Ren, S. (2005). A layered architecture for

real-time distributed multi-agent systems. ACM SIG-

SOFT Software Engineering Notes, 30(4):8.

Julian, V. and Botti, V. (2004). Developing real-time multi-

agent systems. Integrated Computer-Aided Engineer-

ing, 11(2):135–149.

Julian, V., Soler, J., Moncho, M., and Botti, V. (2004). Real-

Time Multi-Agent System Development and Imple-

mentation. Recent advances in artiﬁcial intelligence

research and development, page 333.

Kim, J., Shim, H., Kim, H., Jung, M., Choi, I., and Kim,

J. (1997). A cooperative multi-agent system and its

real time application to robot soccer. In IEEE Interna-

tional Conference on Robotics and Automation, pages

638–643. Institute of Electrical Engineers Inc (IEEE).

Kraus, S., Wilkenfeld, J., and Zlotkin, G. (1995). Multia-

gent negotiation under time constraints. Artiﬁcial In-

telligence, 75(2):297–345.

Liu, X., Zhang, X., Soh, L., Al-Jaroodi, J., and Jiang, H.

(2003). A distributed, multiagent infrastructure for

real-time, virtual classrooms. In Proceedings of the

International Conference on Computers in Education,

pages 2–5. Citeseer.

Mahdi, G., Goua

ıch, A., and Michel, F. (2010). Towards an

Integrated Approach of Real-Time Coordination for

Multi-agent Systems. Agent and Multi-Agent Systems:

Technologies and Applications, pages 253–262.

Micacchi, C. and Cohen, R. (2008). A framework for simu-

lating real-time multi-agent systems. Knowledge and

Information Systems, 17(2):135–166.

Mock, M. and Nett, E. (1999). Real-time communication

in autonomous robot systems. In isads, page 34. Pub-

lished by the IEEE Computer Society.

Pinhanez, C. and Bobick, A. (1996). PNF Calculus: Rep-

resenting and propagating time constrains in Allen’s

interval algebra. Technical report, Technical Report

389, MIT Media Laboratory Perceptual Computing

Section.

Pinhanez, C. and Bobick, A. (2003). Interval scripts: a pro-

gramming paradigm for interactive environments and

agents. Personal and Ubiquitous Computing, 7(1):1–

21.

Pinhanez, C., Mase, K., and Bobick, A. (1997). Interval

scripts: A design paradigm for story-based interactive

systems. In Proceedings of the SIGCHI conference on

Human factors in computing systems, page 294. ACM.

Sierra, C. and Sonenberg, L. (2005). A Real-Time Negoti-

ation Model and A Multi-Agent Sensor Network Im-

plementation. Autonomous Agents and Multi-Agent

Systems, 11(1):5–6.

Soh, L. and Tsatsoulis, C. (2005). A real-time negotiation

model and a multi-agent sensor network implementa-

tion. Autonomous Agents and Multi-Agent Systems,

11(3):215–271.

Soler, J., Julian, V., Rebollo, M., Carrascosa, C., and Botti,

V. (2002). Towards a real-time multi-agent system ar-

chitecture. COAS, AAMAS, 2002.

INTERVAL BASED INTEGRATED REAL-TIME COORDINATION FOR MULTI-AGENT SYSTEMS

669