Drone Landing on Moving UGV Platform with Reinforcement Learning Based Offsets
Abstract: Safe and precise landings of unmanned aerial vehicles (also known as drones) in dynamic platforms such as unmanned ground vehicle (U G V) can significantly enhance the practical applicability and efficiency of operations that benefit from complementing aerial and terrestrial vehicles. In this paper, we propose a Reinforcement Learning (RL) based framework for landing a drone on a moving U G V platform. The trained policy provides position offset commands to an underlying PID position control, which, in turn, converts it into forces and torques to maneuver the drone towards the dynamically moving UGV platform. The policy is trained for three different UGV platform velocities (1 m/s, 2 m/s and 3 m/s) with three runs for each case. The training curves converge to maximal accumulated reward and provide policies that enabled successful drone landings. By simplifying the RL problem and incorporating PID control, we achieve efficient training, and robustness in the landing task, overcoming challenges in generalization across dynamics and avoiding the complexity of more sophisticated control systems.
Loading