An Operational Semantics Dedicated to the Coordination of Cooperating Agents Mohamed Jmaiel, Ahmed Hadj Kacem and Amira Rgaieg LARIS Laboratory FSEG-SFAX B.P. 1088 3018 SFAX - TUNISIA Email:Mohamed.Jmaiel@enis.rnu.tn,Ahmed@fsegs.rnu.tn Abstract. This paper presents a contribution towards rigourous reasoning about coordi- nating agents. First, it defines formal models for coordination and coordinating agents. These models enable to specify the relations between the concepts of: plan, plan proposal and resource allocation, on the one hand, and concepts of: knowledge, belief and capability, on the other hand. Second, it provides a structured coordination language enabling to spec- ify primitives, protocols and processes of coordination. This language is defined by a precise syntax, and it is formally interpreted using a transition system leading to an operational semantics for coordinating agents. Keywords: Operational semantics, Specification language, Formal validation, Transition system, Coordinating agent, Plan, Proposition, Resources allocation. 1 Introduction In contrast to traditional Artificial Intelligence (AI) which models the intelligent behaviour of a single agent, the Distributed Artificial Intelligence (DAI) is interested in intelligent behaviour resulting from cooperative activities of several agents. The transition from individual behaviour to collective behaviour is considered as an enrichment of the AI, from which new properties and activities emerge. The interaction between agents is one of the main results of the collective activity, which tries to increase the performances of the problem resolution at both individual and collective level. Indeed, the interaction constitutes a central concept in Multi-agent Systems (MAS). It occurs due to the inevitable interdependence between agents, and it appears in different forms, namely cooperation, coordination, negotiation and communication. In this paper, we mainly focus on the coordination as a fundamental mechanism to maintain coher- ence between agents and to solve possible conflicts which may occur between them. In general, the coordination may take different forms, such as organizational structuring, contracting, planning and negotiation [HNJ96]. Many models and protocols have been designed to support the coordi- nation activity, such as coordination through coalition formation [SSJ97], coordination by plans conversation, and coordination using contract nets [YHH + 98]. Most of the proposed models are based on the concept of plan. Accordingly, during the coordination process, agents communicate in order to build and to update their individual plans and others plans, while trying to avoid possible conflicts and to enhance performance. This technique was mainly adopted by Durfee [DL89] in its model of Partial Global Planning (PGP), and also by Decker and Lesser [ DEC 94 ] in their model of Generalized Partial Total Planning. In most case, the proposed models are specific to particular domains, and they handle coordination at a low level. This makes them concrete and relatively vague, what complicates reasoning about them. Indeed, neither the process of coordination is formally specified nor the corresponding communication primitives are rigourously clarified. We notice an absence of a methodological approach to specify coordination at both collective (MAS) and individual (agent) levels.