Reinforcement Learning Control for Moving Target Landing of VTOL UAVs With Motion Constraints

This article investigates the motion planning and reinforcement learning control method for autonomous landing issue of a vertical take-off and landing (VTOL) unmanned aerial vehicle (UAV) on a moving target. A novel funnel shaped surface is first proposed to maintain the relative position within a preassigned set for precise and safe landing. An orientation constraint planning is designed to avoid the flipping over. Moreover, a data-based reinforcement learning strategy is proposed to update the learning weight online, which not only effectively relieves the typical persistent excitation condition to the finite excitation but also allows the inadmissible initial control condition. A filter-based disturbance estimator is employed to compensate the unknown terms. The stability of the overall closed-loop system is analyzed theoretically and concluded to be uniformly ultimately bounded. Simulation and flight experiment are performed to validate that UAV can land on the moving target along the preassigned trajectory.