Verbal-command-driven ship trajectory prediction: a context-aware intent injection mechanism for intelligent maritime transportation

Accurate ship trajectory prediction is a prerequisite for collision avoidance and the efficient operation of Intelligent Maritime Traffic Systems and Maritime Autonomous Surface Ships. Current data-driven methods heavily rely on historical AIS data, a satellite-based real-time ship positioning system. Very High Frequency (VHF) voice communications from Vessel Traffic Service (VTS) systems contain valuable, explicitly stated navigational intentions, such as turns and collision avoidance. Despite their high value, this information is not adequately considered in traditional models. To bridge this gap, this paper proposes a novel plug-and-play Intention-Aware Trajectory Modulator (IATM), which utilizes an intention encoder to transform discrete VTS voice commands into semantic vectors. Crucially, the module features an adaptive routing network that intelligently calibrates the confidence weight of these intention features based on the navigational context, allowing the model to adjust its reliance on intention information dynamically. Systematic experiments using simulated intention data demonstrate that integrating intentions can improve prediction accuracy by an average of 30.8%. Furthermore, this study explicitly explores the operational boundaries of intention integration. The study identifies a critical point at 80.8% intention recognition accuracy: above this threshold, intention information provides net benefits, while below it, noise impairs performance. The module also exhibits good adaptability, maintaining predictive performance in low-accuracy environments by learning patterns of noise distribution. This research contributes to the development of intention‑aware Intelligent Maritime Traffic Systems.