Coalition Formation in Multi-Agent Systems under Real-World Conditions

Contents

The benefits of using multiple agents to tackle difficult problems are well-known. To date, however, most work in multiagent systems has focused on making teams of agents work better under the assumption that those teams have already been hand-constructed. Comparatively little research has been performed on coalition formation — the process of forming groups of agents with possibly conflicting individual goals, in order to improve their collective capabilities. Current research in this area has produced a number of approaches for coalition formation, all of which work under strong assumptions of the nature of a coalition and the domain in which it operates. Unfortunately, these assumptions are generally not true of most real-world environments. This paper describes these assumptions in light of current work in coalition formation, and proposes a new approach.