Automated Agent Behavior Construction for Protocol-Based Multiagent Systems

摘要

In this paper we study automated construction of an agent's behavior for a protocol-based multiagent system, where the exact behaviors of the other agents are unknown. This uncertainty makes it challenging to develop an optimal agent behavior that complies with the system protocol, while ensuring achievement of the agent's temporal goals. To solve this problem, we develop an automated method that uses graph search and model checking techniques to construct the optimal agent behavior with respect to a labeled transition systems model of a protocol and linear temporal logic specification of the agent's goals. We show that our algorithm is sound, complete and optimal. We also conducted computational experiments to evaluate the scalability of our method.