Optimal Policy Learning for Distributed Formation Control of Behavior-based UAVs

This work studies the distributed formation control problem of behavior-based unmanned aerial vehicles (UAVs). A novel distributed reinforcement learning behavioral control (DRLBC) is proposed to reduce the cost during distributed formation and obstacle avoidance processes. Firstly, a distributed reinforcement learning task supervisor is designed to prevent undesired priority switching between distributed formation and obstacle avoidance behaviors. Subsequently, a reinforcement learning controller is developed to strike a balance between control efficacy and resource consumption. Furthermore, an adaptive sliding mode control law featuring an unknown control input observer is implemented to enhance system robustness. Finally, the simulation results show that the proposed DRLBC drives a group of UAVs to form a distributed formation while avoiding obstacles near their paths.