Scholarlink

<p>In this paper, a novel distributed finite-time control framework based on reinforcement learning (RL) is proposed to address the optimal time-varying formation (TVF) control problem for nonlinear multi-agent systems (MASs) with model uncertainties under directed network topologies. Initially, each agent is equipped with a newly designed distributed finite-time estimator, enabling it to obtain the desired target states within finite time. This design effectively removes the dependency on global information and decouples the original problem into two independent subproblems: target estimation and tracking control. Subsequently, leveraging the estimated state information, a finite-time optimized control strategy is developed using an RL-based approach. The identifier-critic-actor architecture is employed to seek an approximation to the Hamilton-Jacobi-Bellman (HJB) equation, thereby deriving the optimal control policy. By integrating gradient descent mechanisms with finite-time stability theory, update laws for the weights of neural networks (NNs) are established, thus guaranteeing finite-time convergence while achieving optimal time-varying formation behavior. Finally, the performance and robustness of the proposed approach are verified through numerical simulation studies.</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

Finite-time optimized time-varying formation control for uncertain nonlinear multiagent systems

  • 期刊论文

  • 发表日期：

    2025年7月23日

查看全文
  • Xinyi Wei,
  • Jinxuan Zhang,
  • Fuyong Wang,
  • Zhongxin Liu

摘要

In this paper, a novel distributed finite-time control framework based on reinforcement learning (RL) is proposed to address the optimal time-varying formation (TVF) control problem for nonlinear multi-agent systems (MASs) with model uncertainties under directed network topologies. Initially, each agent is equipped with a newly designed distributed finite-time estimator, enabling it to obtain the desired target states within finite time. This design effectively removes the dependency on global information and decouples the original problem into two independent subproblems: target estimation and tracking control. Subsequently, leveraging the estimated state information, a finite-time optimized control strategy is developed using an RL-based approach. The identifier-critic-actor architecture is employed to seek an approximation to the Hamilton-Jacobi-Bellman (HJB) equation, thereby deriving the optimal control policy. By integrating gradient descent mechanisms with finite-time stability theory, update laws for the weights of neural networks (NNs) are established, thus guaranteeing finite-time convergence while achieving optimal time-varying formation behavior. Finally, the performance and robustness of the proposed approach are verified through numerical simulation studies.