Reinforcement Learning-Based Event-Triggered Adaptive Fixed-Time Optimal Formation Control of Multiple QAAVs

초록

This article focuses on addressing the distributed adaptive fixed-time optimized formation control issue of multiple quadrotor autonomous aerial vehicles (QAAVs) based on reinforcement learning (RL). To optimize the control performance, the RL-based backstepping strategy is first introduced under the identifier-critic-actor architecture, where a monotonically decreasing quadratic function is constructed to provide an upper bound for the critic-actor learning law. Then, the fixed-time command filter is devised to overcome the "explosion of complexity" and singularity problem, based on which a distributed adaptive fixed-time controller is designed to guarantee the formation pattern of multiple QAAVs, while the fractional-power errors' compensation mechanism is established to eliminate the impact of filtering error effectively. An improved event-triggered mechanism is further developed and employed in the devised RL optimized controller to balance communication resource utilization and system performance by dynamically adjusting the triggered threshold. Eventually, it is proven that the whole closed-loop signals are fixed-time bounded, and the effectiveness of the developed control protocol is demonstrated through numerical simulation experiments in multiple QAAVs cluster flights.

키워드