ROMA:Multi-AgentReinforcementLearningwithEmergentRolesTonghanWang1HengDong1VictorLesser2ChongjieZhang1Abstract598Theroleconceptprovidesausefultooltode-signandunderstandcomplexmulti-agentsys-162tems...