From: AAAI Technical Report FS-01-03. Compilation copyright © 2001, AAAI (www.aaai.org). All rights reserved. A Negotiation Formal Model for Cooperating Agents Ahmed Hadj Kacem and Mohamed Jmaiel Universityof S fax LARISLaboratory B.E1088, 3018Sfax, Tunisia E-mail: Ahmed@fsegs. rnu. tn, Mohamed.Jmaiel@enis. rnu. tn Abstract Theworkpresentedin this paperconstitutesa part of a globalapproach aimingat definingthe theoreticalfoundations of a platformfor designingandimplementing Multi-Agents Systems.This paperis primarily centeredaroundthe studyof the negotiationas a technique neededto supportthe cooperativeactivity withinthis kindof systems.Ourcontributionis mainlyaboutthe definitionof formalmodels for negotiationandnegotiating agent.Thesemodelsenableto specifythe relations between the conceptsof plans, planproposalsand resourceallocations,on the onehand,andconceptsof roles, knowledge, beliefsandcapabilities,onthe other hand. Introduction Multi-Agent Systems(MAS)are designed and implemented with a set of agents whichinteract accordingto diversified cooperationmodes,in order to enrich the collective behaviors (Wooldridge& Jennings 1995, Smith & Davis 1980). Thenegotiation plays a fundamentalrole in the cooperative activity by enablingthe agentsto solve conflicts whichcould obstruct such behaviors. Durfee et al. (Durfee & Lesser 1989) define the negotiation as a process whichenhancing the agreementson the agents’ points of view, by reducing inconsistencies and uncertainties. This is reached by exchangingrelevant information. Also, Park et al. (Park Birmingham1995) describe the negotiation as a sequence of interactive communications whichare necessaryto solve conflicts betweenagents. In general, the negotiationis considered as a mechanism neededto coordinate the actions of an agent groupduring the problem-solvingprocess. Thenegotiation has been dealt with in several research works.Mtiller (Mailer1996)classified these researchesinto three categories. Thefirst oneconcernsthe studies on negotiation languages. Theyare interested in the communication primitives, their semanticsas well as the protocols monitoring their occurrenceduring negotiation process. Thesecond one includes researches on the negotiation decisions. These worksare interested in algorithms needed to comparenegotiation subjects, utility functions, andagentspreferences. Researches,in the third category, deal with the negotiation 119 process at high level of abstraction. Theyare mainlyinterested in semanticand behavioralaspects of the negotiation activity. Ourcontribution presented in this paper belongs to the third category. It mainly concerns the study of negotiation as a technique, supportingcooperativeactivity within MAS.Ourinvestigation aims at suggesting an adequateformal definition in terms of modelsfor negotiation and negotiating agents. Thestudy presentedin this paperis an extension of a previous workdefining formal modelsfor cooperation and cooperating agents (Hadj-Kacem&Jmaiel 1999; Jmaiel &Hadj-Kacem 2000). In our opinion, such a formal definition is essential due to the lack of standardsand unified definitions of conceptslike cooperationandnegotiation. Moreover, it is not clear at all howthis conceptsare related to each other. Ourcontribution attempts to identify these conceptsand to highlight their relations. Oncesucha definition has beenestablished, wecan consideredit as a basis for defining behavioral aspects of cooperatingand negotiating agents. In this way,wewill be able to analysetheir behavior. At a higherlevel this allows to formallyspecify behaviorof a MAS and consequently to reason about it. These models constitute a basis of a specification languagefor cooperating and negotiating agents. Our worksbelong to a global approachaimingat defining a platform for designing, in a formal manner, MAS. In order to define a modelfor negotiation, we proposeto highlight the links betweenthe concepts of plan, plan proposal, resource allocation, and commitment. Theseconcepts wouldbe treated from the points of view of agent group and individual agent as well. This enabled us to reformulate the conceptsof role and modeof cooperationspecified in (Jmaiel & Hadj-Kacem 2000). Concerningthe modelfor negotiating agents, wepropose to determine the relations betweenthe concepts of knowledge, belief and capability. Theseconceptsallow us to define the mentalstate of a negotiating agent. Moreover,we insist on the various properties of this kind of agents, such as autonomy,communication,proposal and evaluation. This paper is organized in three sections. Thefirst one presents the various approachesof negotiation within MAS. The secondone summarizesthe mainconcepts used to highlight the modelsof negotiation and negotiating agent. The last one presents a formalizationof the suggestedmodels. State of the Art Different negotiation techniques have been developed. Thesetechniquesare generally basedon the rich diversity of the humannegotiations in various contexts (Davis &Smith 1983). Oneof the moststudied techniquesis basedon an organisational metaphor(Ferber 1995; Davis &Smith 1983). MainlyIt deals with contract-net whichwasone of the most used techniques for MAS.This technique allows to coordinate the collective activities of the agentsby establishing contracts. The same approach has been adopted by Cammarata et al. (S. Cammarata&Steeb 1983). They developednegotiationprotocols to solve conflicts on action plans betweenagents whichoperate according to various cooperation strategies. In the context of our study, we attempt to approachthe negotiation from the point of view of behavioral modelsfor negotiating agents. In the literature, wedistinguish twomajor tendencies. Accordingto the first one, the negotiation arises as beingan inherited part of the solution generation process (Lander &Lesser 1993). It consists of specifying the proposals of solution componentsmadeby the agents until a completesolution has been reached. In order to implement the corresponding algorithm, the authors propose seven operators: Initiate Solution operatorgenerates a proposal whichwill be evaluated by others using the Critique Solution operator. Thegeneration of a feedback, in a conflict situation, is ensuredby the operatorExtendSolution. If the conflict persists, the RelaxedRequirement Solution operator allows, amongothers, the reassessmentof the existing solution. Theoperators Blind-Receive-Informationand Retrieve Informationensure an access to the knowledgebase. Theoperator TerminateSearchis used to indicate the endof the resolution. Thesecondapproachconsiders the negotiation as a mechanismwhichis independentof the solution generation process. It first consists of decomposing the initial probleminto a set of sub-problemswhichwill be distributed amongthe cooperating agents. Eachagent proceedsto plan its tasks and to be committedwith others, in order to achieve them. In this context, the authors in (Park &Birmingham 1995) proposea generic structure aimingat supportingthe variety of the negotiationsituations. This structure consists of three models: negotiating agent model,negotiation languageand protocol. Thefirst modelpoints out the need for maintaining, for each agent, a mentalstate andits associated operations. It tries to specify the behaviorof agents based on their beliefs, knowledge, plans and preferences. Thenegotiation languagedefines a set of performativessuch as: Propose, Counter-Propose, Ack, Accept, Reject, Retract, FeedBack. Theprotocol of negotiation controls the interactions betweenagents during the negotiation process. Negotiation Concepts In our approach, we makeuse of the notion of the model in order to determinethe conceptsrelated to the negotiation within MAS.Thus, we contribute to the design of the negotiation process by specifying two models,namely: ¯ the modelof negotiation, whichis madeup of three levels: 120 global plan, local plan andresourceallocation. ¯ the modelof negotiating agent specifies the internal concepts of an agent as well as their mainproperties. Negotiation Model Theprocess of negotiation enables to avoid any potential conflict situation. Sucha mechanism is based on the generation of proposals, their evaluation and commitment in the case of agreement.Wedistinguish three types of proposals, namely: ¯ proposalfor realization plan of a global goal, ¯ proposalto be responsiblefor a local goal, ¯ proposalfor time interval to access to a sharedresource. Eachproposal will be evaluated by every negotiating agent whoanswers whether with: ¯ an acceptanceif the proposalis accepted, ¯ a rejection if the proposalis rejected and no others are made, ¯ or a counter-proposalif the proposal is rejected and an another proposalis generated. After reachingan agreementbetweenthe agents about a proposal, a commitment will be taken to ensure its realization. There are three types of commitments correspondingto the proposals presented above. In the following, wepresent, in an abstract way,the concepts neededfor the qualification of a negotiation activity. Thisqualification is structuredin three levels: global plan, local plan andresourceallocation. ¯ A global plan is a decompositionof a global goal in an orderedset of local goals. - GlobalPlan Proposalis a suggestion of a global plan madeby an agent participating in the realization of the global goal. - Global Plan Commitment is an engagementto realize a global plan acceptedby all agents. ¯ Local Plangathers the various local goals whichconstitute the agentrole after negotiation. - Local GoalProposalis a suggestionof an agent to take responsibilityto achievea set of local goals. - Local Goal Commitmentis described by the assignmentof the local goals to the appropriateagents. ¯ ResourceAllocation relates to allocation of resources shared betweenagents. - Interval Proposalis a suggestionof a time interval to accessto a shared resource. - Interval Commitmentis an engagementto reserve a time interval to access to a shared resource during the local goal realization. Negotiating agent model The negotiating agent model expresses the individual aspects of an agent, maintainingcoherencewith the proposed negotiation model. Suchan agent is characterized by its mentalstate whichis the result of the interactions with its environment.This mentalstate plays the role of a regulator, whichdirects its behavior.So, certain properties characterize a negotiating agent such as communication,autonomy, reasoning, evaluation, etc. A description of these aspects, namelymental state and properties of negotiating agent, is presentedin a detailed way. ¯ MentalState is a set of beliefs, knowledge, plans and capabilities. - Beliefs are informationreportedby an external event or by another agent. Theydescribe the state of the world from the view point of an agent, and thus, the wayin whichit perceivesits environment. - Knowledgegathers information relating to the tasks achievedby the agent as well as those concerningits environmentand the other agents. Theydiffer frombeliefs by their degreeof certainty. - Planis the set of plans on whichthe agent is engaged to achieve. Theseplans can be of three types, namely: global plan, modeand local plan. - Capabilitydescribesthe ability of an agent to carry out domaintasks as well as control tasks. ¯ Propertiesdescribe the essential properties of an agent, in order to be able to negotiate. Suchan agent should be able to propose, to evaluate and to take decisions. In addition, a negotiating agent mustbe able to adapt its activities withinits environment, so that it couldagree with the other agents andcould solve the consideredconflicts. - Autonomyis an agent directed by tendencies deduced from its mental state and not by commandscoming fromthe user or anotheragent. - Proposingis an agent able to proposeplans for the realization of global goals, local goals andresourceallocation. - Evaluatingis an agent able to evaluate proposals made by other agents and to answer by an acceptance or a rejection accordingto its reasoning. - Communicating is characterized by its capabilities to send, receive and analyse messages. Thefollowingsection presents these two modelsin a formal way. Formal definition of negotiation This section proposesa formal definition of each concept, mentionedin the previoussection, and the relations between them, as well. This formalization, whichis based on a set language,enables to highlight the type of operationsof the different conceptscharacterizingthe negotiation. In our approach,the negotiation processis carried out in three stages. A first stage consists of proposingglobal plans decomposing a common objective into a set of global goals. This stage ends with a total acceptanceof a proposalby all 121 the agents. Oncea commitment on a global plan was made, the agents start the secondstage of negotiation whichends after distributing local goalson the different agents. Thelast stage aims at avoiding conflicts betweenagents.during the problemresolution by coordinatingthe access on the shared resources. Formal model for negotiation In this modelwesupposethe existence of the three following sets: ¯ A~is a set of all agents; ¯ BGis a set of all global goals; ¯ B£is a set of all local goals. A modelof negotiation is defined by a couple (A9, Oc) such as: Ag d~j {A1,A2..... An} where AiCJtG and n:>l Theset Agdenotes the agents society. Oc dof {Bgl, By2,..., (l<i_<n) Bgm} where BgIEB~ and m_>l (l_<i_<m) The set Oc denotes the commongoal which is madeup of set of global goals. Apart from the modelof negotiation, whichformsthe basis for the formaldefinition of the negotiation,wealso need the followingconcepts andnotations: ¯ It is supposedthat each agent has a mentalstate neededto create, modify,evaluate, accept or refuse proposals. The commitment on a retained proposal can update its mental state. Wedenote with EmA~the mental state of the agent Ai. ¯ Wedefine for each agent Ai a confidencecoefficient denoted with CoefA,. Thelatter associates to each agent a value describing the degreeof confidencegranted to its proposals. This value is obtained by a very rich mental state anda veryvast expertise. ¯ A negotiating agent mustbe able, using its mentalstate, to evaluatethe receivedproposals,and to submitits opinion (acceptance or refusal). For simplicity reasons, only consider in this study the case of acceptance, since the case of refusal generally reinitializes the processof negotiation. - Weexpress an acceptanceof a realization plan with a predicate whichtakes as parameters a proposal p and the mental state of an agent A~. Wedenote this predicate withAccept(EmA, , p ). - Similarly, an agent must knowif another agent is able to achieve a particular local goal. Weexpressthe fact that an agent Ai accepts that a local goal b is carried out by another agent Ak, with a predicate denoted Accept(EmA,, b, Ak). In this case, we say that the agent A~has confidence,accordingto its mentalstate, in the agent Akto reach the goal b. It is worthnoting that an agent alwaysaccepts its properproposal. That is to say an agent mustbe alwaysconvincedof its ownproposal. Realization Plan The process of negotiation starts with the definition of a global plan related to a global goal. This definition is the result of the negotiation of proposalsfor global realization plan whichis suggestedby the different agents. (a) Proposalfor a Global Plan A proposal for global plan is a decomposition of a global goal in a nonemptyset of local goals. At this stage, it shouldbe notedthat a global plan does not makeany association betweenagents and suggested local goals. A proposalof a global goal is a subset of B£: {bh b2..... bt} whereb~ E B£and l > 1 (1 _< i <_ l) Duringthe process of negotiation, an agent Ascan propose a global achievementplan for a global goal Bg~.Theset of the proposals relating to Bgj whichcan be submittedby the agent Ai is denotedwith: PG(AI, Bgj) where(l_<i_<n) and(l_<j<m) Theset of the proposalsfor global achievementplans for the global goal Bgj is defined as follows: Let Bgj be a global goal, Pg an accepted global plan belonging to PGEng(Bgj),and Ai an agent. A proposal for responsibility of local goal madeby the agent Ai relative to P9is, in the fact, a subset of P9. Theset of these proposals is denoted with PL(AI, Pg). The set of the proposals for responsibility of the local goals of all the agentsis the union of submitted proposals. This set is denoted by PL(Pg). Hence: PL(Pg) = 0 PL(AI, i=1 It shouldbe notedthat each local goal, belongingto the plan of resolution, mustbe part of at least oneproposalfor a local plan. (b) Local Goal Commitment A local goal will be subject of the negotiation whenit is proposedto be of the responsibility of at least twoagents. Formally: Let b a local goal in PL(Pg).Thelocal goal b is considered as subject of negotiationif andonly if: 3A C_ Ag. IAI _> 2 and b ¯ N PL(Ai, Pg) A~cA PG(Bgj) doj ~.j PG(A~,Bgj) i=1 (b) Global Plan Commitment A proposal for a global plan will be subject of commitment if it is accepted by all agents of the society. This proposal belongsto the set defined by: PGEn9d’d {Pg E PG(Bgj)lAcceptTot(Ag, P9)} (c) Global Plan AcceptanceThis can be reached according to two cases: - all the agents accept the sameproposal; - the agentsare in conflict (no proposalcan satisfy the first case). In this case, the agent whichhas the highest confidencecoefficient will imposeits proposal. Let Ai be an agent of the society, Bgj a global goal and Pgij a proposal belongingto the set PG(AI,Bgj). AcceptTot(Ag, P g~j ) iff (VAk E Ag\{AI} Accept(EmAk,Pgi~)) or (VAk E Ag\{Ai} Coef a~ > Coef k) Local GoalDistributionThefirst step of the negotiation considersonly the decomposition of a global goal in a set of local goals. Thenext step distributes the retained achievementplan on the agents of the society. Indeed,it affects to each agentthe set of local goals whichit has to be achieved. (a) Local Goal Proposal For the achievementof a global goal (after definition of the global plan) the agents diffuse their proposals for responsibility of local goals (whodoes what), Eachproposalis a subset of the selected global plan. 122 Theset A is regardedas the set of agents that proposedto deal with the local goal b as part of the global plan Pg. We denote with Aneg(b, Pg) the set of all agents whoproposed the goal b as part of Pg: Aneg(b, Pg) %f {As ¯ Aglb ¯ PL(AI, P9)} (c) Local Goal Acceptance The conflict between agents mustbe solved by the agents of the society whichselect the appropriateagent for carrying out a local goal, accordingto their mental states (knowledgeon competencesof the various agents) or accordingto the coefficients of confidence. Total acceptanceof a local goal can be reachedeither: - dependingon the mental state, ke. the agent which is readyto achievethe goal. - or accordingto the confidencecoefficient assignedto each agent. AcceptTot(Ag, A~, b, Pg) iff As ¯ Aneg(b, Pg) and ((YAk ¯ Ag\(Aneg(b, Pg)) Accept(EmA~,or (VAt ¯ (Aneg(b, Pg)\{Ai}) CoefA, > CoefA,)) After negotiation,weassign to eachagent a set of local goals called Role, defined as follows: Role(Ai, Pg) %f {b Pg l AcceptTot(Ag, Ai, b, Pg) It is obviousthat the roles of the agents mustbe pairwise disjunct, in order to ensure an optimalcooperation. ResourcesDistribution In the problemsolving process, agents generally access not only to local resources, but also to shared resources. Theaccess to shared resources can involve simultaneous accesses, to the sameresource, which can generate conflicts. In order to cure these anomalies,it is necessarythat each agent reserves the neededresources duringa set of timeintervals. (a) NotationsFor the formal description of the process of negotiation for allocating shared resources wesupposethe existenceof the followingsets: - 2 denotesthe set of all time intervals; - 7~ denotesthe set of sharedresources; Wedenote with Uses(Ai, b,r, I) the predicate whichindicates that the agent Ai, uses the resource r during the time interval [ whileresolvingthe local goal b. Let Ai be an agent, Pg a global plan. Wedefine the set of the intervals, allocated by an agent Ai within the framework of the global plan Pg, as: IntR(Ai,P9)d~__r{(I,r) E 27 x I 3b C Role(Ai, Pg).Uses(A~,b, I)} (b) Interval ProposalWedenote with PI(Ai, P9) the set of proposals for interval allocations of resourcesneededby the agent Ai to realize local goals as part of the resolution plan Pg. Anagent submits a proposal in term of resources allocation intervals with respect to global plan. Anagent should not proposeintervals whichoverlap intervals allocated for global plan not yet completed: PI(AI, Pg) def {(I,r) IntR(Ai, Pg )] YAkE Ag, I’ E 2-. Reserv(Ak, I’, r) ~ I n I’ = ~} (c) Interval Commitment A proposal for a resource allocation is subject of negotiationif the sameresourceis proposedby two different agents for time intervals whichoverlap. Wedenote the set of resource allocations whichare subject of negotiation with Neglnt(Ag,Pg). It is definedas: (I, r) E: Neglnt(Ag,Pg) iff 3Ai, Aj E Ag, I’ E 2-.i ¢ j and (I, r) E PI(AI, Pg) and (I’, r) E PI(Aj, Pg) and I n I’ = 0 A Commitment on an interval of resource allocation is made witha real reservationof the resourceduringthe interval indicatedafter a total agreementof the proposalby all agents. Reserv(Ai, I, r) iff AcceptTot(Ag, Ai, I, r) (d) Allocation AcceptanceA total agreement of a proposal for a resource allocation can be reachedeither by a correspondencebetweenthe mental state of each agent and the proposal, or by using the confidencecoefficient if none of the proposalscan satisfy the first case. AcceptTot(Ag, Ai, r, I) iff (YAke Ag\(Ai} Accept(Ai, r, I, EmA~)) (YAlE EngIR(r, I) Coef , >Coef A , ) or where EnglR(r,I) de=~ {A E AgISPg E PO.(I,r) E PI(A, Pg)} 123 (e) Local Plan For every agent, the process of negotiation enables to specify a local plan definedin term of roles and resourceallocations: Plan(Ai, Pg) d~d (Role(Ai, Pg), Res(Ai, where Res(gi, Pg) %f {(i, r)i3 b C Role(Ai, Pg). Uses(A~,b, I, r) and Reserv(Ai,I, r)} Wedenotewith Plansthe set of all the possible plans. Formal Model for Negotiating Agent MentalState The set of knowledgeor facts of an agent Ai is denotedby FactsA,. Toeach fact ¢ the agent A~assigns, by applying a function CCA,,a coefficient of certainty denoted CCA,(¢). This is becausean agent is not alwayssure of the truth of any given fact. Thevalue CCA,(¢) is always between0 and 1. Accordingto this value wedistinguish two typesof facts: ¯ Knowledge are facts whosecoefficient of certainty equals 1. ](~A, d¢_f {~bl~b E FaCtSA,and CCA,(~3 ) : 1} A knowledgeis a fact of whichthe agent is sure of its truth. ¯ Beliefs are facts whosecoefficient of certainty is strictly lower than 1. BeIA, dcf {¢1¢ E FactsA, and CCA,(¢) < 1} A belief is a fact of whichthe agentis not sure of its truth. Consequently,it cannot be consideredas knowledge. The mental state of an agent is madeup of knowledgeand beliefs. Wedenotewith EmA~ the current mentalstate of the agent Ai, definedas: EmA,da ~’Ai U 13elA~ Wedenotewith Smthe set of mentalstates of the all agents. Negotiating AgentProperties In order to be able to conclude a process of negotiation, an agent has to be equipped witha suitable set of capabilities. Wepaya particular attention to the capability of communication. Besides, an agent must also be able, using its mental state, to generate and evaluate proposals. Finally, an agent has to be autonomous in order to makedecisions concerningthe updating of its ownworking plan. Wedenote with Cap&the set of functionalities that an agent Ai can achieve.A capability is represented, accordingto its nature, by a function, a relation or an elementaryprimitive (operation). Theessential capabilities neededfor the negotiation are presentedformallyas follows: ¯ Communication is the capability of exchanginginformation of various types: data, facts or proposals for global plans, local plans or resourceallocations. This capability is ensured by two primitives, namelySend and Receive: - The primitive Send enables an agent to submit a message of any type to anotheragent. Thesubmittingagent must specify the nameof the receiver agent, the informationtype and the messageto be sent. - Theprimitive Receive allows the reception of messages already sent by another agent. Anagent Ai is said to be Communicating if it has among its capabilities the primitives SendandReceive. ¯ Proposal is the capability to generate proposals or counter-proposals,using the mentalstate andthe already madeproposals. Wedenoteswith 79 the set of all the proposals (PGU PLU PI) and 2;o the of all subsets of 79. Anagent A~is knownas Proposingif has the capabilities Propose, CounterPropose,and Evaluate, defined as follows: dcf Proposing(Ai) 3Propose, CounterPropose,Evaluate E CaPA ~ where Propose: CounterPropose: Evaluate: References gm x (t3G U PG u PL) ~ (e~, Bg/Pg/Pl) ~-~ Pg/Pl/Pi $m x (Bg UPG u PL) x 2 9 ~ 79 (el, Bg/Pg/Pl, Ep) ~-* Pg/Pl/Pi Cm × 79 ~ (Accept, Reject} Proposedefines, using the mental state and the global goal, global plan, or local plan, respectively, a proposal for a global plan, a local plan or a resourceallocation. ¯ Autonomy is the capability to generate and modifyits ownworkingplan. At the end of the process of negotiation, an agent shoulddefine the tasks whichit mustcarry out in order to achievethe goals specified in its role. Accordingto resource allocations over the time axis, it defines the sequenceof tasks to be carried out. Thesetasks constitute the agendawhichis described by a sequenceof pairs (task, time). Anagent is said to be Autonomous if it is able to create andmodifyits agendaaccordingto its mentalstate andto its plan. Formally: Agendaac=r{[(taskl,tl) .... , (task,~,t,~)]l task~ E Tasks and t~ E Time} def Autonomous( Ai ) 3CreateAgenda,UpdateAgendaC CaPA, where CreateAgenda: Plan ×$m ~ Agenda (p, e) ~ UpdateAgenda: This study constitutes a part of a global approachaiming at defining a platformfor designingand implementing multiagent systems. At the current state of this work, a formal validationof the modelsis essential at the semanticlevel as well as at the behavioralone. Thusthe validated modelscan be used in two different ways: - as basis for a detailed analysis for the behaviorof a set of negotiating agents. - as a formalismfor specifying a negotiating agent system. Acknowledgment:The authors are gratefully thankful for MoniaLoulouand AmiraRgaiegfor their valuable contributions concerningthe specification and the implementation of the negotiating agent system. Plan ×$m× Agenda ----* Agenda (p, e, a) H Conclusion This article is centered aroundthe study of the negotiation as a techniqueneededto supportthe cooperativeactivity in a MAS. Thus, wehave initially presented the conceptsrelated to the negotiation modelsuch as proposals,their evaluations as well as the generatedcommitments. In the secondstep, a modelfor negotiating agent is given. This modeldescribes the agent’s mentalstate and the properties neededto support negotiation. It shouldbe noted that a systemof negotiating agent, based on the suggestedmodel,has been implemented. 124 Chainbi, W.; Jmaiel, M.; and Hamadou,A. B. 1998a. Conception, behavioral semantics and formal specification of multi-agent systems. In Proceedingsof the 4th Australian Workshopon Distributed Artificial Intelligence, volume 1544of LectureNotesin Artificial Intelligent. Chainbi, W.; Jmaiel, M.; and Hamadou,A.B. 1998b. A newapproachto the conceptionand the behavioral semantics of a multi-agent system. In Proceedingsof the 2nd IEEEInternational Conferenceon Intelligent Processing Systems. Davis, R., and Smith, R.G. 1983. Negotiation as a metaphorfor distributed problemsolving. Artificial Intelligence 20(1). Decker, K. 1987. Distributed problem-solvingtechniques : A survey. IEEEtransactions on systems, man,and cybernetics 17(5):729-740. Durfee, E. H., and Lesser, V. 1989. Negotiating task decompositionand allocation using partial global planning. DistributedArtificial Intelligence 2(1 ):229-244. Ferber, J. 1995.Les syst~mesmulti-agents. Vers une intelligencecollective. InterEdition. Hadj-Kacem,A., and Jmaiel, M. 1999. Towardsa formal definition of the cooperation amongmulti-agent systems. In Proceedingsof the first IFACWorkshopon Multi-AgentSystemsin production. Hadj-Kacem,A. 1996. Les SBCcoop6ratifs, vers une architecture fond6esur uneformalisationdu contr61e.In Proceedings du lOdmeCongrdsReconnaissancedes Formeset IntelligenceArtificielle. Jmaiel, M., and Hadj-Kacem, A. 2000. A formal definition of cooperation and agencyamongmulti-agent systems. In ProceedingsInternational conferenceon Artificial and ComputationalIntelligence for Decision, Control and Automationin Engineeringand Industrial Applications. Jmaiel, M. 1996. Specification of communicationprotocols using temporallogic. Journalof Systemsand Software, Special Issue on SoftwareEngineeringfor Distributed Systems 33(3):299-312. Lander,S. E., and Lesser, V. 1993. Understanding the role of negociation in distributed search amongheterogeneous agents. In Proceedingsof the thirteen International Joint ConferenceonArtificial Intelligence, 438---444. Mi~ller, H. J. 1996. Negotiationprinciples. England:John Wiley& sons. Park, S., and Birmingham,W.P. 1995. Toward a unified frameworkfor multiagentnegotiation. In Proceedings of the international Conferenceon Intelligent Computer Communication,105-122. S. Cammarata,D. M., and Steeb, R. 1983. Strategies of cooperationin distributed problemsolving. In Proceedings of the IJCAI83. Smith, R. G., and Davis, R. 1980. Frameworks for cooperation in distributed problemsolving. IEEEtransactions on Systems, Manand Cybernetic 11(1). Wooldridge, M., and Jennings, N. R. 1995. Intelligent agents: Theoryand practice. KnowledgeEngineering Review 10(2). 125