NaviAIS: A Scenario-Level Vessel Trajectory Prediction Dataset with Vectorized Lane Priors and the NaviLane Forecasting Framework

The paper introduces NaviAIS, a scenario-level AIS dataset that standardizes multi-vessel historical and future trajectories within unified temporal windows and local coordinate systems. It includes rasterized navigable maps, vectorized lane priors, lane graphs, and structured map representations. The accompanying NaviLane framework performs trajectory-map joint encoding, generates multimodal candidates through a discrete macro-action codebook, applies residual refinement, and ranks candidates with a world-model-based evaluator considering interaction risk and environmental feasibility. The abstract reports improvements over representative baselines in both single-modal and multimodal forecasting, but provides no metric values or benchmark details.