Oliver Thompson, Director of Engineering at Marine AI, presented at the 11th MASRWG Conference on a collaborative project with the UK Hydrographic Office (UKHO) that leverages frontier Artificial Intelligence to solve fundamental navigation challenges for autonomous vessels.
The core problem lies in the nature of critical nautical information. Data streams like Sailing Directions and Radio Navigational Warnings are designed for human interpretation. They are rich in context and marine-specific knowledge but are often unstructured, ambiguous, and not readily “machine-understandable.” Traditional computational algorithms struggle with this ambiguity and lack sophisticated methods to interpret natural language or coordinate via VHF radio, which remains essential for port operations and vessel-to-vessel communication.
To close this gap, Marine AI has developed a specialized architecture in partnership with UKHO. The system ingests both offline sources (e.g., Admiralty Sailing Directions extracted from PDFs) and real-time updates (e.g., Notices to Mariners via VHF or web services). This data is fused with the vessel’s traditional sensor suite (AIS, radar, computer vision) and processed through a bespoke, finely-tuned Large Language Model (LLM).
This LLM is the key innovation. It transforms unstructured textual descriptions—such as “several isolated patches on the eastern side” or a VHF warning about an “inoperative fog signal”—into structured, georeferenced data that the autonomous control system can comprehend and act upon. The output is integrated into the vessel’s navigational map, directly influencing safe route planning and hazard avoidance in real-time.
A critical design principle emphasized by Thompson is 100% operational capability without external communications. All AI processing occurs on-board the vessel, ensuring safety and functionality is maintained even in areas with degraded or lost satellite/radio links.
While modern “machine-first” data standards like the IHO’s S-124 for warnings are emerging, their global adoption is slow. In the interim, AI acts as a vital bridge, parsing today’s human-centric data. Looking forward, this natural language interpretation capability has broader applications, such as enabling advanced mission autonomy where complex, high-level verbal or written commands can be decomposed into actionable tasks by the AI.
The partnership demonstrates that frontier AI is not just an add-on but an essential enabler for achieving higher levels of maritime autonomy, allowing systems to safely navigate the ambiguous, context-dependent world that human mariners have operated in for centuries.
Source: Oliver Thompson, Director of Engineering at Marine AI. Presentation given at the 11th Maritime Autonomous Systems Regulatory Working Group (MASRWG) Conference, January 22, 2026.https://marineai.co.uk/