Multi-objective path planning for unmanned surface vessels based on DQN-NSGA-III algorithm

The rapid expansion of maritime transportation and offshore operations has highlighted the need for efficient and intelligent navigation of Unmanned Surface Vessels (USVs). Traditional path planning methods often struggle to balance multiple conflicting objectives such as route length, energy consumption, and navigational safety in complex and dynamic marine environments. To address this challenge, this study proposes a hybrid path planning framework that integrates Deep Q-Networks (DQN) with an improved NSGA-III algorithm. A grid-based environmental model is constructed to support multi-objective optimization, while a DQN module refines local maneuverability and adaptability. A path repair mechanism ensures route feasibility under uncertain conditions. Comparative simulations on different map scales demonstrate that the proposed method achieves better overall performance than DQN, NSGA-III, A*, and Bidirectional A* algorithms. On a 50×50 map, the model reduces total cost by 12.6% and navigational risk by 39.5%. These results confirm that the DQN-NSGA-III approach effectively improves planning accuracy, safety, and efficiency, showing strong potential for real-world deployment in maritime autonomous navigation.