Coalition formation among autonomous agents

Abstract

Coalitions of agents can work more effectively than individual agents in many multi-agent settings. Determining which coalitions should form (i.e., what agents should work together) is a difficult problem that is typically solved by some kind of centralised planner. As the number of agents grows, however, reliance on a central authority becomes increasingly impractical. This paper formalises the coalition formation problem in decision theoretic and game theoretic terms and presents a fully distributed algorithm that can efficiently determine coalitions that will be approximately “stable.” Stable coalitions are resistant to attempts of outsiders to break the coalition, because remaining in the coalition maximises the expected reward for each agent in the coalition. The algorithm is a variant of the “stable marriage matching with unacceptable partners” [6]. The Shapley value ([11], [12]) is suggested as a fair method to divide the coalition's utility among the members.KeywordsCoalition FormationStable MatchCustomer OrderCoalition MemberPreference ListThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.