MACBeth: A Multi-Agent Constraint-Based Planner

摘要: MACBETH is a constraint-based tactical planning engine for multi-agent teams. designed domains in which human user must quickly specify mission to team of autonomous agents. In these domains, "puzzle-mode" thinking come up with novel plans not important; the key task rapidly and accurately tailor existing situations. To this end, combines hierarchical network modern constraint reasoning techniques, into mixed-initiative system. This system driven by graphical interface inspired "playbook" metaphor, generate, check modify teams heterogeneous has been tested two robotics domains: Unmanned Combat Aerial Vehicles (UCAV) sorties Tactical Mobile Robotics (TMR). quickly, do this, be able simply easily high-level "play" perform. The must, however, plan; completely automatic acceptable. every scenario, brings context information process that directly available automation. For example, UCAV domain, wing commander identify objective sortie. some cases, may want add further restrictions on behavior UCAVs. he might indicate should circumnavigate airspace belonging noncombatant nation, or select different attack formation based his own skills preferences. applications, ensure are feasible. An attempt an infeasible plan flagged algorithm as soon possible, so she can retract decision reconcile conflicting objectives early process. initially assign widely separated reconnaissance targets single insufficient fuel reach both locations. planner inconsistency ideallybefore it expands effort details, such determining final approach routes targets, immediately alert need her allocation resources.