M3-Former addresses long-term vessel trajectory prediction challenges

VERA Brief

AI-generated. Grounded in the article and its cited sources.

Researchers have developed M3-Former, a new multimodal Transformer model that uses large language models and a Mixture-of-Experts architecture for improved vessel trajectory prediction. This framework integrates static attributes and navigational intent to enhance accuracy over extended horizons, offering a verifiable approach to maritime navigation safety and efficiency.

Key facts

• M3-Former is a multimodal trajectory prediction framework designed to address challenges in behavioral multimodality, semantic utilization, and long-term error accumulation.
• The model incorporates vessel static attributes and navigational intent as semantic priors for long-term trajectory modeling.
• A dual-granularity Mixture-of-Experts architecture is employed to capture both global route planning and local motion variations.
• Experiments on a Danish AIS dataset show M3-Former surpasses state-of-the-art baselines across prediction horizons of 1 to 4 hours.
• For the 4-hour prediction task, M3-Former reduced Average Displacement Error by 4.4% and Final Displacement Error by 5.1%.