NaviAIS Dataset and NaviLane Framework Advance Vessel Trajectory Prediction

Accurate vessel trajectory prediction is crucial for maritime safety and management, but existing datasets often lack standardization and structured representations of navigational lanes. This makes it difficult to develop environment-aware forecasting models. To address this, researchers have introduced NaviAIS, a standardized scenario-level dataset. It organizes multi-vessel trajectories within unified temporal windows and local coordinate systems, crucially providing rasterized navigable maps, vectorized lane priors, lane graphs, and structured map representations. This comprehensive dataset supports reproducible, environment-aware forecasting. Leveraging NaviAIS, the team also developed NaviLane, a hierarchical macro-action framework for map-aware prediction. NaviLane encodes trajectory-map information, generates multimodal candidate trajectories using a discrete codebook, refines them for consistency, and ranks them based on interaction risk and environmental feasibility. Experiments show NaviLane significantly outperforms existing baselines, validating the importance of structured navigational priors for robust vessel trajectory forecasting.