Design and Validation of an Intention-Aware Probabilistic Framework for Trajectory Prediction: Integrating COLREGS, Grounding Hazards, and Planned Routes

2025-04-01

AI Summary

Conventional trajectory prediction methods for autonomous ship navigation suffer from poor generalizability and fail to explicitly model the intentions of other vessels. Method: This paper proposes an enhanced Dynamic Bayesian Network (DBN) model to establish an intention-aware probabilistic trajectory prediction framework. For the first time in DBN-based maritime prediction, it jointly models grounding risk, COLREGs-compliant constraints, and waypoint-level intentions of encountering vessels—overcoming the limitation of single-perspective inference through multi-source intention fusion. The method integrates AIS data-driven validation, maritime geographic risk modeling, and formal rule encoding. Contribution/Results: Evaluated on real-world AIS encounter scenarios, the approach reduces trajectory prediction error by 23.7% and improves intention recognition accuracy by 18.4%, significantly enhancing interpretability, robustness, and decision support for safe collision avoidance in complex waterways.

Technology Category

• Humans and AI: Human-Aware Planning and Behavior Prediction
• Intelligent Robots: State Estimation
• Planning, Routing, and Scheduling: Model-Based Reasoning

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Collision avoidance capability is an essential component in an autonomous vessel navigation system. To this end, an accurate prediction of dynamic obstacle trajectories is vital. Traditional approaches to trajectory prediction face limitations in generalizability and often fail to account for the intentions of other vessels. While recent research has considered incorporating the intentions of dynamic obstacles, these efforts are typically based on the own-ship's interpretation of the situation. The current state-of-the-art in this area is a Dynamic Bayesian Network (DBN) model, which infers target vessel intentions by considering multiple underlying causes and allowing for different interpretations of the situation by different vessels. However, since its inception, there have not been any significant structural improvements to this model. In this paper, we propose enhancing the DBN model by incorporating considerations for grounding hazards and vessel waypoint information. The proposed model is validated using real vessel encounters extracted from historical Automatic Identification System (AIS) data.

Problem

• Improving trajectory prediction by integrating vessel intentions and hazards
• Enhancing Dynamic Bayesian Network with grounding and waypoint data
• Validating model using real AIS data for collision avoidance

Innovation

• Enhanced DBN model with grounding hazards
• Incorporated vessel waypoint information
• Validated using historical AIS data