A multiagent planner using agent constraints

A planning system for a multiagent environment is presented. Improvements in planning efficiency, cooperation handling, and plan quality are achieved. The restriction of using agents with the same capabilities as imposed by most planners is relaxed. The system has three features: the meta-level planning, the breakable/unbreakable action representation, and the integrated agent assignment scheme. The meta-level planning utilizes domain-related heuristic knowledge to organize a given set of subgoals to avoid an incorrect subgoal achievement sequence. Unlike other planners which create a multiagent plan by integrating multiple single-agent plans, this approach involves less conflict analysis at low-level plans. The meta-level planning process also identifies specific needs of cooperation in subgoal achievement. The breakable/unbreakable action representations provide formal ways to define cooperation and specify agent involvement. With cooperation identification and breakable/unbreakable action representations, low-level action generation is straight forward. The integrated agent assignment scheme provides ways to consider agent constraints in improving plan quality and at the same time generates actions to suit agents.