COUNTERPLANNING FOR MULTI-AGENT PLANS USING STOCHASTIC MEANS-ENDS ANALYSIS

摘要

We have developed a hierarchical planning method for multiple agents in worlds with significant levels of uncertainty. This has resulted in expert-system tools (MEAGENT) for analysts and planners without background in artificial intelligence. MEAGENT is particularly useful in analysis of counterplanning methods intended to thwart plans in complex situations. We apply heuristics to define experiments involving many runs of carefully modified simulations, use the results to quantify the effects of various counterplanning tactics, and then produce a counterplan. We exemplify our "experimental AI" approach for the domain of firefighting on ships.