Coalition Formation Algorithm Taxonomy

Informally, coalition formation is the process of agents forming coalitions (e.g., teams) in order to accomplish tasks, complete goals or in general improve their (individual or collective) situation beyond what each agent could achieve on its own. Much research has been done on this abstract coalition formation problem [12, 15, 14, 16, 17, 4]. However, the informal “coalition formation problem” description encompasses a variety of different situations. For example: Are the agents self interested or group rational? Are their constraints between the individual tasks or goals (e.g., subgoals of a goal)? Are their restrictions on agent communication? The variety of situations encompassed by the “coalition formation problem” makes direct comparison of existing algorithms tedious at best, as it requires the meticulous extraction of the underlying model and assumptions employed by each algorithm. Oftentimes assumptions are left implicit (e.g., communication restrictions if any) or the depth of the models employed is not sufficient to accurately capture the complexity of many real-world problems. All of these factors complicate the comparison between algorithms. We attempt to address this difficulty of comparing coalition formation algorithms and models by: 1. Providing a list of dimensions along which coalition formation algorithms and models may be compared. In doing so we attempt to make explicit many underlying assumptions commonly employed.