Abstract (EN)

Negotiation is a fundamental mechanism in multi-agent systems since it allows self-interested agents to come to mutual agreements and to distribute resources efficiently and effectively. However, in competitive situations, the agents need to negotiate repeatedly with one another, thus developing mechanisms that are effective over repeated interactions is a key challenge. At present, few research work deals with designing mechanisms for complex multilateral negotiations in which agents’ interactions are fully decentralized. We tackle the problem of multilateral negotiation where a group of agents has to make a joint decision. The explanatory breadth of this mechanism is illustrated by looking at its applicability in situations in which agents are self-interested, have asymmetric information, are motivated by individual interests but they have to converge to an agreement, and behave opportunistically. We search to guide the agents in their process, promote even partial solutions and facilitate the achievement of agreements specially in presence of large sets of agents where negotiation is notably difficult and convergence is hardly achieved. To this end, the protocol we propose leads the agents in efficiently coordinating their negotiations and to coming to an agreement based on provided rules. Negotiation illocutions are formalised to allow interactions between the agents. In addition rules are provided for the agents to manage their interactions. We introduce an illustrative scenario and detail the properties of our protocol. Experimental results show that our protocol guides the agents towards an agreement with reasonable communication and computational complexity.