关键词: 强化学习, 固定翼无人机集群, 分布式观测器, 动态事件触发, 编队跟踪控制

Abstract: Aiming at the formation problem of fixed wing UAV under the conditions of limited communication resources, modeling uncertainty and interference, a new hierarchical pilot follow formation control scheme combined with reinforcement learn-ing is proposed. Firstly, a distributed state observer based on dynamic event triggering mechanism is designed to save communication resources and use local information to accurately estimate the leader's state, so as to ensure that Zeno phenomenon will not occur. Then, in order to compensate the system uncertainty and wind disturbance, a sliding mode tracking controller combined with reinforcement learning is proposed, and a lightweight actor critical algorithm is con-structed by using RBF neural network, and the sliding mode surface and convergence law are designed according to the fixed time theory. The convergence of the observer and controller is strictly proved by Lyapunov stability theory. Finally, the effectiveness of the proposed algorithm is verified by simulation.

Key words: reinforcement learning, fixed wing UAV swarm, distributed observer, dynamic event triggering, formation tracking control