Optimal Group Consensus of Multiagent Systems in Graphical Games Using Reinforcement Learning
研究了多智能体系统的最优群体一致性问题，提出基于离策略强化学习的分布式控制策略，通过数据驱动值迭代实现群体一致，并扩展到非完整移动机器人的编队跟踪问题。
This article investigates the optimal group consensus problem (GCP) in multiagent systems (MASs). To address this problem, a novel distributed optimal control policy is designed in the framework of off-policy reinforcement learning (RL). First, a framework for multiagent differential graphical games is formulated. Second, a min-max strategy is then introduced to ensure the achievement of group consensus through a data-driven value iteration (VI) approach. Finally, the presented consensus control policy is extended to address the group formation tracking problem (GFTP) of nonholonomic mobile robots, with a numerical example to illustrate the efficacy of the proposed results. Compared with the existing literature, this article has the following contributions: 1) A group of agents are decomposed into multiple subgroups to accomplish different consensus objectives; 2) the prior knowledge of agents’ dynamics and initial stabilizing control gains can be eliminated; and 3) the performance index function (PIF) for each agent is designed to integrate not only its individual control policy but also that of its neighboring agents.