Objective Coordination in Multi-Agent System Engineering: Design and Implementation

137 Pages·2001·1.06 MB·English

by  Michael Schumacher (auth.)

Lecture Notes in Artificial Intelligence 2039
SubseriesofLectureNotesinComputerScience
EditedbyJ.G.CarbonellandJ.Siekmann
Lecture Notes in Computer Science
EditedbyG.Goos,J.HartmanisandJ.vanLeeuwen
3
Berlin
Heidelberg
NewYork
Barcelona
HongKong
London
Milan
Paris
Singapore
Tokyo
Michael Schumacher
Objective Coordination
in Multi-Agent
System Engineering
Design and Implementation
1 3
SeriesEditors
JaimeG.Carbonell,CarnegieMellonUniversity,Pittsburgh,PA,USA
Jo¤rgSiekmann,UniversityofSaarland,Saabru¤cken,Germany
Author
MichaelSchumacher
Berninastrasse85,8057Z(cid:252)rich,Switzerland
E-mail:[email protected]
Cataloging-in-PublicationDataappliedfor
DieDeutscheBibliothek-CIP-Einheitsaufnahme
Schumacher,Michael:
Objectivecoordinationinmulti-agentsystemengineering:designand
implementation/MichaelSchumacher.-Berlin;Heidelberg;NewYork;
Barcelona;HongKong;London;Milan;Paris;Singapore;Tokyo:
Springer,2001
(Lecturenotesincomputerscience;Vol.2039:Lecturenotesin
arti(cid:222)cialintelligence)
ISBN3-540-41982-9
CRSubjectClassi(cid:222)cation(1998):I.2.11,C.2.4,D.1.3,D.2.12,I.2.9
ISBN3-540-41982-9Springer-VerlagBerlinHeidelbergNewYork
Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis
concerned,speci(cid:222)callytherightsoftranslation,reprinting,re-useofillustrations,recitation,broadcasting,
reproductiononmicro(cid:222)lmsorinanyotherway,andstorageindatabanks.Duplicationofthispublication
orpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawofSeptember9,1965,
initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer-Verlag.Violationsare
liableforprosecutionundertheGermanCopyrightLaw.
Springer-VerlagBerlinHeidelbergNewYork
amemberofBertelsmannSpringerScience+BusinessMediaGmbH
http://www.springer.de
'Springer-VerlagBerlinHeidelberg2001
PrintedinGermany
Typesetting:Camera-readybyauthor,dataconversionbyChristianGrosche,Hamburg
Printedonacid-freepaper SPIN:10782507 06/3142 543210
Foreword
This book addresses the engineering of Multi-Agent Systems (MAS) based
on suitably-de(cid:12)ned coordination models and languages. Its contribution is
twofold: on the one hand, the theoretical part de(cid:12)nes a new coordination
model (ECM) for designing MAS and proposes two corresponding coordina-
tion languages (STL++ and Agent&Co) for implementing designed sys-
tems in distributed environments; on the other hand, the experimental part
presents a software prototype by starting o(cid:11) the existing Pt-pvm (a dis-
tributed threads environment developed by OliverKrone, a formerPhD stu-
dent of our PAI research group) and validates the theoretical part with two
case studies. De(cid:12)nitelyoneofthe mainresult is thatthe ECM modeland its
explicit distinction of objective and subjective coordination de(cid:12)ne an origi-
nalandcomprehensiveframeworkformodeling,designing,andimplementing
interactions in complex MAS.
This work has been carried out within the framework of the Parallelism
andArti(cid:12)cialIntelligenceresearch group(PAI)attheUniversityofFribourg,
Switzerland (http://www-iiuf.unifr.ch/pai). The PAI group originated
from a synergy between the (cid:12)elds of Distributed Computing and Arti(cid:12)cial
Intelligence. Since 1986, it has been involved in massively distributed pro-
gramming methodologies, and since 1995 in decentralized control strategies
as far as they are suited for providing adaptive capabilities in network com-
puting. Research topics of the group follow an engineering trend, which can
be characterized by:
{ The abandonment of centralized control and stringent hierarchical data
structures infavorofdecentralized controlstrategies based oninteractions
through influences. These strategies, which require autonomous compo-
nents, lead to solutions, which are more flexible, more tolerant to pertur-
bations, and which support the emergence of new properties;
{ A bio-inspired approach, which draws its models from neuro-biology and
thestudyofanimalsocietiesandparticipatesintheconceptofembodiment
of intelligence now surfacing in AI;
Between 1995 and 1999, PAI’s research (cid:12)tted inside the AXE project
(AXE is a compressed acronym for \Autonomy and Coordination Supports
Evolution") whose aim was to devise a self-organization and a coordination
viii Foreword
theory for massively distributed systems. AXE’s main acquisitions concern
(i) the development of coordination platforms for distributed applications;
(ii) the de(cid:12)nition of models for autonomous, adaptive, and evolutive agents;
(iii)the designofcoordinationandevolutionmodels,namelycollectiveintel-
ligence strategies. Since 1998, human computer interaction aspects are also
part of PAI’s research, in the measure that universal networking and ubiq-
uitous computing formulate special requirements in this (cid:12)eld. With AXE’s
terminationatthebeginningof2000,thisdomainsets thenewfocusadopted
byPAIinsidethenewWELCOMEframework,buildingontheknow-howac-
quired on autonomous systems and coordination.
It was a pleasure for me to work with Michael Schumacher. His multi-
disciplinaryresearch results have been well received in the MAS and coordi-
nationcommunities.Themodeling,design,andimplementationofMAS,the
ECM coordination model, and the STL++ and Agent&Co coordination
languages, which are the central parts of his PhD thesis described in this
book, are now used in our projects presently under development in WEL-
COME.
Finally,I warmly acknowledge the (cid:12)nancial support of the University of
Fribourg, the Swiss National Science Foundation, and the European Pro-
gramme of R&D for having granted PAI’s research projects for the past
fourteen years.
December 2000 B(cid:19)eat Hirsbrunner
Leader of the PAI research group
Preface
Thisbook,situatedattheintersectionofbehavior-basedarti(cid:12)cialintelligence
and concurrent and distributed computing, de(cid:12)nes programmingparadigms
to support the design and the concurrent and distributed implementationof
Multi-Agent Systems (MAS) that simulate collective robotics applications.
We analyze the research that has tried to (cid:12)ll the gap between agent theory
andapplications,observingthattheproposed methodologies,languages,and
tools are mostly concentrated on intra-agent aspects. In contrast to those
approaches, we propose that the modeling of MAS should be a bottom-up
and interaction-oriented process, grouping existing autonomous agents and
describinghowtheyinteract,thusmanagingthecoordinationofthese agents.
Tothataim,wedistinguishobjectivecoordination,whichhandlesinter-agent
dependencies (the organization of the environment and the agent interac-
tions), and subjective coordination, which handles intra-agent dependencies
often involvingmentalistic categories. We then promote a methodologythat
focuses the modelingof MAS on objective coordination,and we propose the
use of coordination models and corresponding languages (from the (cid:12)elds of
concurrentanddistributedcomputing,programminglanguages,andsoftware
engineering) in order torespectively support the design phase of aMAS and
allowits implementationon a concurrent and distributed architecture.
After reviewing coordination models and languages,we examine the pre-
requisites that acoordinationmodel andlanguageshouldinclude inorder to
support our target MAS.On this basis, ageneral coordinationmodel named
ECM and a corresponding object-oriented coordination language named
STL++ are presented. ECM/STL++ uses an encapsulation mechanism
as its primary abstraction, o(cid:11)ering structured separate name spaces which
can be hierarchically organized. Agents communicate anonymously within
and/or across name spaces through connections, which are established by
the matching of the communication interfaces of the participating agents.
Three basic communicationparadigmsare supported, namelypoint-to-point
stream,group,andblackboardcommunication.Furthermore,aneventmech-
anism is introduced for supporting dynamicity by reacting to state changes
of the communication interfaces.
The use of ECM/STL++ is illustrated by the simulationof a particular
collective robotics applicationand of the automationof a trading system.
x Preface
Acknowledgements
ThisbookcompilesfouryearsofresearchworkthatIcarriedoutasamember
of the Parallelism and Arti(cid:12)cial Intelligence group of the Computer Science
Department at the University of Fribourg inSwitzerland, working on a PhD
thesis funded by the Swiss National Science Foundation. I bene(cid:12)ted enor-
mously from the knowledge, the expertise, and the support of many people.
My special thanks go to: B(cid:19)eat Hirsbrunner who gave me the opportu-
nity to work in a highly quali(cid:12)ed research group and who has guided my
research activity during the past years; Rolf Ingold, Andrea Omicini, and
George Papadopoulos, for accepting the invitation to act as jury members;
FabriceChantemargueandOliverKrone forthe numerousdiscussions,ideas,
criticisms, and also for their constant help and enthusiasm; Antony Robert,
ValerioScarani,MicheleSchumacher,andRobertVanKommerforreviewing
previousversionsofthisdocument;ChristianWettsteinforimplementingthe
(cid:12)rst version of STL++; Ivan Doitchinov for implementing Agent&Co; all
colleagues in the PAI group for creating a stimulatingresearch environment;
and last but not least, my familyand my friends for their constant support
and encouragement.
January 2001 Michael Schumacher
Contents
1. Introduction:::::::::::::::::::::::::::::::::::::::::::::: 1
Part I. Positioning
2. Multi-Agent Systems ::::::::::::::::::::::::::::::::::::: 9
2.1 Introduction ........................................... 9
2.2 What Is an Autonomous Agent?.......................... 9
2.2.1 De(cid:12)nitions ...................................... 10
2.2.2 Autonomy and Embodiment ....................... 11
2.2.3 Generic Agent Architectures ....................... 12
2.3 Characteristics of MASs................................. 14
2.4 Modeling MASs ........................................ 15
2.4.1 Objective Coordination ........................... 17
2.4.2 Subjective Coordination........................... 19
2.4.3 Emergence ...................................... 21
2.5 Our Target Class of MASs............................... 21
2.5.1 A Generic Model for an Autonomous Agents’ System . 22
2.5.2 A Typical Application: Gathering Agents ............ 23
2.6 Implementing MAS Applications ......................... 24
2.6.1 Languages for MAS Applications ................... 24
2.6.2 Methodologies for MAS Applications................ 28
2.6.3 Using Coordination Models and Languages for
Designing and Implementing MASs ................. 29
3. Coordination Models and Languages:::::::::::::::::::::: 33
3.1 What Is Coordination?.................................. 33
3.2 What Are Coordination Models and Languages? ........... 33
3.2.1 Motivation ...................................... 33
3.2.2 Key Elements.................................... 34
3.3 Data-Driven Coordination Models ........................ 35
3.3.1 Linda .......................................... 36
3.3.2 Linda-Based Models ............................. 38
3.3.3 Models Based on Multiset Rewriting................ 39