14 September 2024 in Ocean Engineering
1 January 2026 in IEEE Transactions on Industrial Electronics
8 June 2026 in IEEE/ASME Transactions on Mechatronics
Autonomous surface vehicles (ASVs) require accurate trajectory planning to ensure safe and efficient navigation in complex maritime environments. Traditional planning algorithms, including search-based and sampling-based methods, often have limitations such as trajectory nonsmoothness, high computational cost, and poor adaptability to environmental disturbances. Reinforcement learning (RL) has emerged as a promising alternative due to its ability to learn optimal policies from interaction data without relying on precise system models. However, most RL-based methods remain confined to simulation due to the challenges posed by sim-to-real transfer and unmodeled hydrodynamic disturbances. This article proposes a novel trajectory planning framework integrating data-driven hydrodynamic modeling and RL. A transformer-based model is employed to learn the coupled hydrodynamic characteristics of ASVs from real-world state data, including ASV states influenced by wind, waves, and currents. The learned model is then incorporated into a RL-based planner to generate safe and dynamically feasible trajectories. Experimental evaluations in simulated and real-world water environments demonstrate that the proposed approach significantly improves trajectory reachability, target-point accuracy, and overall navigation safety.