Experimental validation of a modular navigation architecture for marine autonomous surface vehicles with reactive collision avoidance

This paper presents a modular software architecture for the autonomous navigation of surface vehicles, designed around a layered awareness, navigation, guidance, and control structure. The proposed framework separates global path management, reactive local planning for collision avoidance, and control, while situation awareness combines LiDAR perception with INS/GNSS localization to maintain an up-to-date, and realistic representation of the surrounding environment. The architecture is designed around the concepts of modularity and scalability, enabling distributed computation and the flexible integration of modules. The implementation employs a lightweight publish/subscribe protocol to enable efficient real-time communication among modules. The experimental validation of the proposed architecture in a collision avoidance test featuring a research ASV is reported and discussed. The vehicle successfully executed polygonal paths, adapting its trajectory to avoid multiple unexpected obstacles while still reaching its prescribed waypoints. These results demonstrated the reliability of the proposed framework in supporting path following and adaptive collision avoidance under realistic operating conditions.