Equilibrium Selective Role Coordination for Autonomous Driving

Abstract

Role coordination is crucial in multi-agent collaboration because the collaboration may fail if the roles played by agents are inconsistent. In this paper, we present a role coordination method, Equilibrium Selective Role Coordination (ESRC), for decentralized continuous mutual action control in autonomous driving. In ESRC, the roles of agents are represented by game-theoretic equilibrium points that the agents try to achieve. ESRC comprises three hierarchical functions: (1) action due to given dynamics and constraints, (2) prediction of mutual actions, and (3) selection of roles. Corresponding to this functional hierarchy, three-layered mutual belief hierarchy is adopted. Each agent acts to achieve equilibrium with other agents while selecting an equilibrium point as an appropriate role assignment adaptively and online to reduce risk. The results of simulation experiments conducted demonstrate that our proposed method could produce appropriate actions even in complicated situations where several possible collisions needed to be considered. ESRC can be used to model a wide range of decentralized multi-agent based phenomena, such as human-robot physical interactions, dialogues, economic activities, artificial muscles, and neural information dynamics.