| A collaborative coverage-evacuation decision model is developed to address uncontrollable terminal positions and additional inter-regional transfer distance in coverage path planning. Four directional evacuation boundaries are encoded as conditional observations, and a five-channel observation representation is designed by fusing perceived obstacles, local view, visitation history, agent position, and evacuation target. Proximal policy optimization and two-stage curriculum learning are used to jointly optimize area coverage and directional evacuation. Random-obstacle experiments show that EBCC-PPO improves late-stage average return by 69.70% compared with PPO. In multi-obstacle scenarios, the average coverage rate, evacuation success rate, and comprehensive effectiveness reach 92.30%, 90.64%, and 88.60, respectively. The results indicate that EBCC-PPO achieves high-coverage path planning with controllable terminal evacuation in complex obstacle environments. |
