Layered Reinforcement Learning Design for Safe Flight Control of UAV in Urban Environments

Unmanned aerial vehicles(UAVs) have gained prominence due to their high maneuverability, robust controllability, and reliability. Nowadays, reinforcement learning has attracted increasing attention among researchers. In this paper, a data-based PID control method is proposed to craft a trajectory tracking controller for UAV to address the problem that adjusting controller parameters can be a complex task. Furthermore, aiming at the issue of sparse rewards in intricate urban environments, the concept of hierarchical reinforcement learning is proposed to decompose the overall task into smaller. The proximal policy optimization algorithm is then proposed to train UAV, enabling it to make informed decisions within urban environments. The experimental results demonstrate the effectiveness of our autonomous flight strategies proposed for the UAV, showing both their efficacy and high performance.