UAV Path Planning for Maritime Search and Rescue Based on Deep Reinforcement Learning

This paper presents a UAV path planning method based on Proximal Policy Optimization (PPO) to address the dynamic drift of targets and strict time-critical constraints in maritime search and rescue (SAR) missions. We integrate the Open Drift framework with ocean environmental data for drift prediction, and conduct discrete probability modeling to construct a gridded search environment. Furthermore, we design a deep reinforcement learning framework that improves the basic PPO algorithm. We introduce parallel interaction sampling (PIS) to boost strategy exploration efficiency and use a shared representation network to reduce parameter redundancy. Experimental results from real-world SAR cases show that our method outperforms existing ones by better escaping local optima, enhancing cumulative search success rates, and achieving faster convergence. This research offers an efficient technical approach for intelligent SAR decision making in complex marine environments.