Enhancing port channel logistics safety: An asynchronous distributed DRL approach for waterborne ASVs trajectory tracking under dynamic disturbances

Port channel safety is essential for reliable vessel movements and efficient logistics operations in intelligent ports. Waterborne ASVs can support inspection, monitoring, and auxiliary transportation in restricted port waters, but their deployment requires accurate trajectory tracking to avoid additional port-channel navigational safety risks. However, reliable ASV tracking remains challenging under dynamic disturbances and strict ship-to-shore communication constraints. This study proposes an Asynchronous Distributed Deep Reinforcement Learning (AD-DRL) control scheme for robust ASV trajectory tracking under dynamic marine disturbances. Unlike traditional centralized training, the proposed framework enables distributed ASVs to collaboratively update policies by exchanging model gradients rather than raw sensor data, thereby reducing communication requirements. To improve decision-making in dynamic waters, a gated hybrid attention mechanism is integrated into the Twin Delayed Deep Deterministic Policy Gradient (TD3) architecture. This mechanism helps the agent capture spatiotemporal disturbance features and delayed vessel responses caused by hydrodynamic lag. Moreover, an ERA5-informed data-driven simulation environment is constructed using real-world ocean current, wind, and wave data to train and evaluate the controller. Experiments across five representative international ports show that the proposed method achieves lower tracking errors and higher success rates than baseline methods under varying environmental disturbances. Additional safety-corridor analysis further confirms its ability to maintain reliable trajectory execution in port channels. These results demonstrate the robustness and practical relevance of AD-DRL, providing technical support for safer and more reliable ASV operations in intelligent port logistics scenarios.