Misalignment Tolerance Enhancement and Intelligent Minimal-Activation Pathway Decision for Unmanned Aerial Vehicles WPT Applications

This paper introduces a reinforcement learning (RL)-enhanced wireless charging system for unmanned aerial vehicles (UAVs), addressing critical challenges of misalignment tolerance and energy-efficient coil matching in autonomous landing scenarios. A reconfigurable magnetic array generates a 2D rotating field to accommodate positional/angular deviations while a Q-learning framework maps stochastic UAV positions into grid states via pick-up voltage feedback, optimizing energy transfer unit (ETU) activation through multi-objective rewards (efficiency, switching cost, boundaries). Experimental validation shows 354.4 W output at 87.74% DC-DC efficiency, with 92.2% fewer activation steps than exhaustive methods. The RL strategy achieves average 5-step activations and <4.2 cm coarse positioning errors, enabling 3D spatial awareness without prior models. By synergizing adaptive magnetic structures with model-free RL, this work establishes a scalable UAV charging solution, balancing high misalignment tolerance, and efficient power delivery. This article is also accompanied by a video file demonstrating training experiment in PyCharm.