Optimizing Multi-Agent Coverage Path Planning UAV Search and Rescue Missions with Prioritizing Deep Reinforcement Learning

In this paper, we present a novel multi-agent real-time coverage path planning algorithm designed for Unmanned Aerial Vehicles (UAVs) engaged in Search and Rescue (SAR) missions across complex 3D terrains, particularly alpine environments. The proposed approach leverages Deep Reinforcement Learning (DRL) to dynamically adapt to evolving scenarios. The algorithm significantly reduces search time by prioritizing areas near hiking trails, where the likelihood of locating individuals in distress is higher. The DRL environment for training the model incorporates real-world data, including GPS positions of surrounding hiking paths and high-precision 3D information data from open source point clouds. This integration of diverse data sources enhances the reliability and effectiveness of the system in operational scenarios, ensuring robust performance in real-time SAR missions. SAR scenarios were simulated and the results were benchmarked against actual rescue mission data.