Exploiting Opportunities through Dynamic Coalitions in Robotic Soccer

Contents

While the coordination of multiple agents in static networks or hierarchical relationships has been the subject of much research, the relationships between agents in dynamic multi-agent systems has only comparatively recently been the subject of significant investigation. In dynamic multi-agent systems, the nature of relationships between agents can change, possibly along with the goals of a group. In such systems, the environment surrounding the agents may change, altering the optimality of interactions (e.g. a change in opponent agents; a change in optimal strategy leading to different relationships between agents). While we may have preferences for the way in which agents interact (or even some agents in authority to rely upon), there is no hard-and-fast answer to who should optimally interact with whom and under what conditions. Agents may seek out momentary coalitions with others, or form longer-term bonds that are flexible enough to support some degree of interaction under a wide range of circumstances. This paper discusses issues in forming coalitions between agents in robotic soccer, and focuses on recognizing the situations where coalitions are useful and entering into and terminating coalitions both verbally and non-verbally as events unfold.