Scholarlink

<p>This paper is concerned with group consensus problem of unknown multi-agent systems. Both the leader-following group consensus (LFGC) and the leaderless group consensus (LLGC) problems are investigated. For LFGC problem, a novel Q-function-based policy iteration (PI) algorithm with explicit policy expression is developed, accompanied by rigorous convergence analysis. The developed PI algorithm effectively minimizes the performance index and determines the optimal control policy through iterative refinement. Building on this foundation, a novel model-free algorithm is proposed by constructing a simplified critic-only neural network structure to solve the LFGC problem online using only system operation data. Compared to existing LFGC algorithms, the proposed algorithm overcomes the requirement for system model while simplifying the algorithm configuration. Furthermore, it is shown that the model-free algorithm also applies to the LLGC problem. Two simulation analyses are presented to demonstrate the capacity of the proposed model-free algorithm.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Model-free group consensus of multi-agent systems using critic-only reinforcement learning

  • 期刊论文

  • 发表日期：

    2025年9月20日

查看全文
  • Chuanjian Li,
  • Xiaoping Wang,
  • Zhigang Zeng,
  • Tingwen Huang

摘要

This paper is concerned with group consensus problem of unknown multi-agent systems. Both the leader-following group consensus (LFGC) and the leaderless group consensus (LLGC) problems are investigated. For LFGC problem, a novel Q-function-based policy iteration (PI) algorithm with explicit policy expression is developed, accompanied by rigorous convergence analysis. The developed PI algorithm effectively minimizes the performance index and determines the optimal control policy through iterative refinement. Building on this foundation, a novel model-free algorithm is proposed by constructing a simplified critic-only neural network structure to solve the LFGC problem online using only system operation data. Compared to existing LFGC algorithms, the proposed algorithm overcomes the requirement for system model while simplifying the algorithm configuration. Furthermore, it is shown that the model-free algorithm also applies to the LLGC problem. Two simulation analyses are presented to demonstrate the capacity of the proposed model-free algorithm.