Human-MASS Interaction in Decision-Making for Safety and Efficiency in Mixed Waterborne Transport Systems

Abstract

This thesis explores the integration of Maritime Autonomous Surface Ships (MASS) into Mixed Waterborne Transport Systems (MWTS), addressing critical challenges in ensuring navigational safety and operational efficiency. Recognising the complexities of interactions in MWTS, especially in scenarios without direct communication between vessels, the research develops a decision-making framework that integrates situational awareness, human-preference-aware navigation, and trust dynamics. These components collectively aim to support seamless interactions between autonomous and manned vessels, ensuring safe and efficient navigation in the MWTS. The proposed framework builds on a systematic exploration of key challenges in MASS operations. For situational awareness, an ontology-driven knowledge maps model is introduced, enabling MASS to integrate multi-source data and maritime regulations. This model is further combined with a Dynamic Window Approach (DWA) path planner, allowing for real-time compliance with COLREGs and proactive collision avoidance. The research also advances human-preference-aware navigation by extracting and modelling navigational behaviours of manned vessels using AIS data. An LSTM-autoencoder with clustering methods is utilised to identify navigational preferences, which are then incorporated into a trajectory prediction model based on Multi-Task Learning Sequence-to-Sequence LSTM with attention (MTL-Seq2Seq-LSTM-Att) architectures. This integration enhances MASS decision-making by aligning manoeuvring strategies with human operators’ expectations, reducing the likelihood of misinterpretation in mixed traffic scenarios.