Hybrid Energy-Time Efficient Reinforcement Learning for UAV Coverage Path Planning Under Wind Disturbance

Abstract

UAV coverage in real environments is challenging because onboard energy limits and spatially varying wind jointly affect motion, safety, and propulsion costs. This paper proposes ETA-PPO, or Energy and Time Aware Prior Guided PPO, within a framework that separates wind aware flight execution from fleet level coverage coordination. At the execution layer, ETA-PPO augments PPO with a deterministic state conditional VAE action prior and a bounded residual policy for closed loop point to point flight under wind disturbed dynamics. At the coordination layer, Heuristic JointETA-PPO-H assigns target, return to base, and hold macro actions above the frozen low level executor. Experiments are conducted in a shared 3D urban simulator with CFD derived time varying wind fields. Across point-to-point navigation, ETA-PPO achieves a 100% success rate on all evaluated tracks while maintaining a practical balance between energy use and flight time against actor-critic baselines. The same executor also completes all four long-horizon single-UAV ROI tours, demonstrating reliable transfer from individual legs to chained coverage execution. In the scaled multi-target, multi-UAV case study, Heuristic JointETA-PPO-H reaches 100% coverage in every tested fleet and launch configuration. The fleet-level results are therefore treated as deterministic systems evidence.