Multiagent Planning as an Emerging Behavior in Agent Societies

Multiagent planning is the activity of synthesizing a plan of actions for a team of (usually cooperating) agents. The problem is often addressed as a special case of distributed problem solving by assuming that agents are homogeneous and that they can trust on each other. In some cases, agents can even inspect one another their own beliefs. These solutions, however, are not applicable when agents constitute an open society in which agents are heterogeneous and can join/leave at any time. To cope with such a kind of systems, we propose to address multiagent planning problem by means of a social computing approach. The idea is to explicitly model a social state as a set of commitments. In fact, since commitments have a normative power, they enable an agent in the society to (pragmatically) reason about the expected behaviors of the others. We argue that such a pragmatical reasoning can be effectively implemented via classical planning techniques.