Date of Graduation
Summer 2026
Degree
Master of Science in Computer Science
Department
Computer Science
Committee Chair
Rahul Dubey
Abstract
The maritime industry is moving toward Autonomous Surface Vessels (ASVs) capable of navigating without onboard crew, creating new challenges for collision-avoidance deci- sion making. To operate safely, vessels must comply with the International Regulations for Preventing Collisions at Sea (COLREGs), which define vessel responsibilities during encounters. Existing approaches, including rule-based planners and deep reinforce- ment learning, encode COLREGs into algorithms but often struggle in situations outside their design assumptions. Large Language Models (LLMs) offer an alternative by interpreting COLREGs in natural language and reasoning about their application. However, many COLREGs provisions rely on contextual reasoning and human judgment, making autonomous implementation challenging. Motivated by these challenges, this thesis pro- poses a novel hybrid agentic framework for COLREGs-compliant ASV navigation in com- plex maritime environments. The research consists of three interconnected studies: (1) examining how zero-shot prompt richness influences collision-avoidance decisions, (2) investigating the impact of temporal-context prompting, and (3) exploring an agentic AI ar- chitecture in which the LLM autonomously manages contextual information. Insights from these studies are integrated into a hybrid framework that combines LLM-based reasoning with rule-based validation. Extensive simulation experiments across diverse encounter scenarios demonstrate that zero-shot prompting alone is insufficient for reliable navigation, while temporal context improves decision quality and consistency. Results further show that the proposed hybrid agentic framework successfully navigates ASVs in complex scenarios while maintaining COLREGs compliance. Overall, this thesis advances agentic AI-driven maritime navigation and provides a foundation for the safe integration of LLMs into future autonomous vessel operations.
Keywords
Agentic AI, LLMs, ASVs, COLREGs, Maritime Autonomy
Subject Categories
Automotive Engineering | Computer Engineering | Digital Communications and Networking | Human-Computer Interaction | Robotics
Copyright
© Arjun Silwal
Recommended Citation
Silwal, Arjun, "Hybrid Agentic AI Framework for COLREGs-Compliant Autonomous Surface Vessel Navigation" (2026). Graduate Theses/Dissertations. 4209.
https://bearworks.missouristate.edu/theses/4209
Open Access
Included in
Automotive Engineering Commons, Digital Communications and Networking Commons, Human-Computer Interaction Commons, Robotics Commons
