A real-time decision-making method for USV swarm area coverage empowered by large language models

Unmanned surface vehicle (USV) swarms are increasingly utilized for complex maritime tasks, such as continuous environmental monitoring, island patrols, and strategic surveillance. However, existing methods still face significant challenges in online autonomous task allocation, dynamic replanning, and real-time adaptation to unexpected events in open and uncertain environments. This study aims to improve the decision-making autonomy, robustness, and task continuity of USV swarms during long-duration area coverage operations. To achieve this, we propose a real-time decision-making and cooperative control framework powered by large language models (LLMs).

The proposed method introduces a hierarchical decomposition framework that integrates semantic interpretation with structured task planning. By incorporating LLMs, natural language mission instructions are dynamically translated into executable objectives, enabling USV swarms to autonomously generate efficient, conflict-free navigation paths. A hybrid architecture is adopted, where a centralized planner handles global task allocation, while distributed planners on individual USVs compute local collision-free trajectories in parallel. To manage uncertainty, a dual closed-loop feedback mechanism is implemented: (i) at the task level, dynamic prompt generation enables reallocation when resources change or failures occur; (ii) at the action level, incremental trajectory correction through geometric collision detection ensures safe adaptation without the need for full replanning. This multilayered structure enhances scalability, responsiveness, and resilience for long-term multi-agent operations.

The proposed framework was validated in a simulated maritime island patrol scenario, where multiple USVs collaboratively encircled randomly generated islands and returned to base. Comparative experiments with classical methods, such as A*, genetic algorithms, and ant colony optimization, demonstrate that our approach achieves comparable efficiency in terms of path length and travel time, while uniquely ensuring strict adherence to natural language mission goals. Cross-model evaluation with phi-4, Qwen-3, and DeepSeek R1 further shows that lightweight, reasoning-oriented models strike the optimal balance between planning accuracy and computational efficiency. Experimental results indicate that the framework achieves an average task success rate of over 80% in complex, long-term, multi-task scenarios, while maintaining a success rate above 70% for real-time responses and handling of abnormal events (e.g., random USV failures). Ablation studies confirm that hierarchical task decomposition and multi-level feedback significantly contribute to the overall performance improvement.

This study demonstrates that LLMs can effectively support decision-making for USV swarms in dynamic maritime environments. By integrating semantic instruction understanding, hybrid centralized–distributed planning, and a dual closed-loop feedback mechanism, the proposed framework ensures task continuity, adaptability, and safety. These findings present a promising approach for deploying USV swarms in reliable, long-endurance missions and provide a foundation for extending this method to large-scale, real-world multi-agent systems.