PID with Deep Reinforcement Learning and Heuristic Rules for Autonomous UAV Landing

Unmanned Aerial Vehicle (UAV) landing is a challenging task in dynamic environments. The PID controller can be used, but it suffers from the problem of manual parameter tuning. In this paper, we propose that PID can be combined with deep reinforcement learning to learn the PID parameters autonomous learning. Besides, heuristic rules of how to adjust the PID parameters can be used to further speed up the learning. We demonstrate the effectiveness of the proposed method in a simulated quadrotor UAV landing task.