Planning has traditionally focused on single agent systems. Although planning domain languages have been extended to multiagent domains, solution concepts have not. Previous solution concepts either focus on planning for teams of agents with a single goal, or on a single agent in an environment containing other agents with unknown and unpredictable behavior. In reality other agents are usually acting to achieve their own explicit goals, which may not be the same or even related. In game theory the notion of an equilibria provides a framework for thinking about self-interested, utility maximizing agents. We define a formalization of the multiagent nondeterministic planning problem and introduce a notion of equilibria inspired by the game theoretic concept. As far as we know, this is the first solution framework that explicitly accounts for the various goals of all agents. In addition to the formalization we also demonstrate how this framework applies in a number of different domains....
Michael H. Bowling, Rune M. Jensen, Manuela M. Vel