Reinforcement Learning-Based Distributed Event-Triggered Leaderless Consensus for Unknown Linear Multi-Agent Systems
Xiangkai Wu, Long Jiang, Wei Wang, Zhen Han · 2024
In this paper, the leaderless consensus control problem for general linear multi-agent systems with completely unknown dynamics is investigated. Different from the existing results, both event-triggered communication and controller updates are considered under a directed graph condition. The primary challenges addressed in this paper are twofold: the completely unknown system dynamics for each agent and the inherent asymmetry of the Laplacian matrix that characterizes the communication topology. To tackle these challenges, the hybrid iteration reinforcement learning algorithm is utilized to derive the unique solution to the algebraic Riccati equation (ARE). Furthermore, the distributed consensus control schemes and the triggering conditions are proposed without using the precise knowledge of the system dynamics of all the agents. It can be proved that all the closed-loop signals are globally uniformly bounded. Besides, the practical leaderless consensus can be reached with discretized communication and controller updates, effectively avoiding Zeno behavior in all the agents. Finally, we present illustrative simulation results which underscore the effectiveness of the proposed control scheme.