EnvShip: A Unified Framework for Context-Aware and Cross-Region Vessel Trajectory Forecasting

2026/06/13 by Kun Ma, Qilong Han, Chengjing Song +3 · 1 voice
Computer Science · #cs.LG

paper · pdf

16 pages,8 pages of main text

arxiv published 2026/06/13 · arxiv created 2026/08/02 · arxiv updated 2026/08/02

Abstract

Accurate vessel trajectory forecasting is essential for maritime situational awareness, navigation safety, traffic management, and autonomous navigation. Public Automatic Identification System (AIS) archives have enabled extensive research in this area, yet results remain difficult to compare because existing studies use incompatible preprocessing pipelines, forecasting horizons, data splits, coordinate systems, contextual inputs, and evaluation settings. We present EnvShip, a unified multi-region framework for context-aware and cross-region vessel trajectory forecasting. EnvShip applies a fixed and reproducible pipeline to public AIS data from Denmark, the United States, Greece, and Norway, and defines two standardized forecasting tracks spanning short- and long-horizon settings. From large-scale processed data, we curate 330,000 short-term and 106,857 long-horizon samples through strict motion screening, vessel-category and difficulty stratification, redundancy control, and vessel-disjoint splits. Each sample is aligned with environmental and neighboring-vessel context, together with weather and sea-state variables where available. We evaluate representative methods under in-domain and cross-region protocols, with analyses across prediction difficulty, scene type, and random seeds. Results show that environmental context provides the largest gains in coastline-constrained scenes, whereas neighboring-vessel context primarily benefits interaction-rich cases. Multi-region training improves generalization in most settings but introduces negative transfer for some source combinations. EnvShip provides a common and reproducible testbed for vessel trajectory forecasting. Code and data are publicly available at https://huggingface.co/datasets/mark000071/envship_v2_datasets.

Citations

• MSTFormer: Motion Inspired Spatial-temporal Transformer with Dynamic-aware Attention for long-term Vessel Trajectory Prediction
• Argoverse 2: Next Generation Datasets for Self-Driving Perception and Forecasting
• AgentFormer: Agent-Aware Transformers for Socio-Temporal Multi-Agent Forecasting
• Deep Learning Methods for Vessel Trajectory Prediction Based on Recurrent Neural Networks
• Multi-Modal Trajectory Prediction of NBA Players
• Transformer Networks for Trajectory Forecasting
• Argoverse: 3D Tracking and Forecasting with Rich Maps
• nuScenes: A multimodal dataset for autonomous driving
• Social GAN: Socially Acceptable Trajectories with Generative Adversarial Networks
• An Empirical Evaluation of Generic Convolutional and Recurrent Networks for Sequence Modeling
• FiLM: Visual Reasoning with a General Conditioning Layer
• Attention Is All You Need
• Exploiting AIS Data for Intelligent Maritime Navigation: A Comprehensive Survey
• Crowds by Example
• Long Short-Term Memory

Discussions

• From trajectory prediction to cough diagnostics, our benchmarks are maturing from simple metrics to real-world impact. We’re moving beyond model performance; we’re architecting systemic reliability. E [bsky, 1 points, 0 comments]

Related

• NaviAIS: A Scenario-Level Vessel Trajectory Prediction Dataset withVectorized Lane Priors and the NaviLane Forecasting Framework
• Context-Informed Ship Trajectory Prediction via Conditional Attention
• Unified Multimodal Vessel Trajectory Prediction with Explainable Navigation Intention
• A Multi-Modal Knowledge-Enhanced Framework for Vessel Trajectory Prediction
• Multi-vessel Interaction-Aware Trajectory Prediction and Collision Risk Assessment