Reconfiguration of multi-agent planning systems
Diane J. Cook · 1994
Many multi-agent planning systems ca ~ perform more effectively than single-agent systems due to redun-dancy, concurrency, and betemgencous capabilities. However, lack of an effective coordination mecha-nism can diminish the advantages of using multiple agents. Researchers have suggested a variety of coor-dination configurations, such as centralized planning, distributed planning, and local planning. While each configuration offers some advantages, no single plan-ning configuration is optimal for all planning scenar-ios ̄ In this paper, we identify the need for reconfiguration in multi-agent planning systems. Instead of selecting just one planning configuration, we design a system that can make use of a number of planning modes. In order to determine the best configuration for a given problem, a decision-theoretic model is designed. This models selects the configuration that will maximize the expected payoff for the system. P~conflguration of the multi-agent system can be initiated by a single or by multiple agents, and can occur between plans or during plan execution.