Hierarchical Two-Stage Framework for Vessel Trajectory Prediction
A recent study has developed a two-layer framework designed to forecast ship movements over long periods in real ocean settings. This innovative system combines a long-term predictor with a short-term, grid-focused predictor using a hierarchical fusion method. For the short-term predictions, a Spatio-Temporal Graph Transformer analyzes maritime cells to grasp local dynamics, while the long-term aspect focuses on broader navigational goals. Furthermore, an environmental module incorporates oceanographic elements such as surface currents, wind patterns, and significant wave height, employing cross-modal attention to adjust to changing sea conditions. This research aims to improve collision avoidance, traffic management, and route planning by enhancing prediction accuracy in the face of long-term dependencies and variable environments. You can check out the paper on arXiv (2605.16442).
Key facts
- The framework is hierarchical and two-stage.
- It combines coarse long-term and grid-aware short-term predictors.
- Short-term branch uses Spatio-Temporal Graph Transformer on maritime cells.
- Long-term branch encodes navigational intent.
- Environmental module includes currents, wind, and wave height.
- Cross-modal attention and feature-wise modulation are used.
- Aims to improve collision avoidance and route planning.
- Published on arXiv with ID 2605.16442.
Entities
Institutions
- arXiv