Domain knowledge embedded anti-disturbance autonomous navigation for marine vehicles.

The complex ocean disturbances in ocean engineering have long constrained the precise autonomous navigation of intelligent marine vehicles, such as surface vessels and underwater vehicles. Nevertheless, the unpredictable wind-wave-current coupling effects pose severe challenges to the safety of autonomous navigation for marine vehicles. Here we introduce a domain knowledge embedded disturbance observation-control framework, fusing real-time observation and compensation for composite environmental disturbances using model-free control. This framework embeds specialized basis functions from domain knowledge into a specialized Kolmogorov-Arnold network and extracts control knowledge therefrom to train a machine learning controller. Our approach achieves better adaptability and robustness, surpassing conventional model-based controllers. It enables more accurate path-following and safer operations under complex ocean disturbances. It is worth noting that this method has been validated to be effective for both surface vessels and underwater vehicles through offshore wind farms inspection task scenarios. It fundamentally extends the adaptive control theory for marine cyber-physical systems and has potential applications in multi-domain oceanographic operations.