0

Multi-vessel Interaction-Aware Trajectory Prediction and Collision Risk Assessment

Accurate vessel trajectory prediction is essential for enhancing situational awareness and preventing collisions. Still, existing data-driven models are constrained mainly to single-vessel forecasting, overlooking vessel interactions, navigation rules, and explicit collision…

Preview
Year
2025
Hosting
Abstract onlyARXIV-DEFAULT

Cite

Notes

Only stored in your browser.

Attribution

Abstract & full text
arxiv.org/abs/2509.01836ARXIV-DEFAULT
TL;DR
Semantic Scholar
Attribution policy →

Abstract

Accurate vessel trajectory prediction is essential for enhancing situational awareness and preventing collisions. Still, existing data-driven models are constrained mainly to single-vessel forecasting, overlooking vessel interactions, navigation rules, and explicit collision risk assessment. We present a transformer-based framework for multi-vessel trajectory prediction with integrated collision risk analysis. For a given target vessel, the framework identifies nearby vessels. It jointly predicts their future trajectories through parallel streams encoding kinematic and derived physical features, causal convolutions for temporal locality, spatial transformations for positional encoding, and hybrid positional embeddings that capture both local motion patterns and long-range dependencies. Evaluated on large-scale real-world AIS data using joint multi-vessel metrics, the model demonstrates superior forecasting capabilities beyond traditional single-vessel displacement errors. By simulating interactions among predicted trajectories, the framework further quantifies potential collision risks, offering actionable insights to strengthen maritime safety and decision support.