Team Agent Behavior Architecture in Robot Soccer

摘要

One of the most challenging goals in artificial intelligence (AI) is to develop a team of autonomous agents with human-performance. In a team-based multi-agent system with cooperative and opponent team agents, the ability to build a coordinated system of game strategies can be useful to provide the agents with team plans to achieve common goals. These team plans are used to make agents decide which behavior is the best in a given situation, specially in a realistic setting of situated agents with local perception and uncertain domains. In this paper we present a team agent behavior architecture (TABA), an approach to coordinate the behavior of a multiagent team. We introduce an autonomous and dynamic leader agent capable of analyzing the current environment, including the opponent team, and select a team plan according to the state of the world; this agent is selected among one of our cooperative team agents. The roles of the robots are not fixed and they can dynamically exchange their roles with each other. The plan selected is pre-defined and is communicated to the rest of the agents who execute it in a distributed fashion. We also propose a soccer strategy description symbols (SSDS) that is very simple and intuitive. TABA is built using a hierarchical architecture. The first stage is responsible for leader agent selection; the second stage is for strategy selection; the third stage is for role assignment and last stage is for tactic execution, in which robots execute their behavior commands based on their roles. We have successfully followed a case-based reasoning approach for strategy selection in the robot soccer domain, a simulation of the RoboCup four-legged league soccer competition. Initial results of TABA are also discussed.