The UK Hydrographic Office (UKHO) and marine autonomy software company Marine AI have launched a new research programme to enable autonomous surface vessels (ASVs) to read, interpret, and act upon official navigational information once only usable by human mariners.
The eight-month project will see Marine AI fine-tune its baseline large language model (LLM) to process admiralty sailing directions (SDs) information and radio navigation warnings (RNWs), currently written in natural language for human interpretation, and feed this structured information into the software suite for autonomous control.
The UKHO said this development could enable uncrewed vessels to make safe, real-time decisions based on the same authoritative data used by professional mariners.
At present, ASVs such as maritime autonomous surface ships rely on human operators to interpret text-heavy navigation data, often described in non-standard nautical language and distributed through legacy systems.
The UKHO said project will address the challenges posed by unstructured text, legacy broadcast formats, and the lack of machine-readability, by retraining a bespoke LLM and developing supporting AI agents to structure the data before it is fed into the AI software's tactical engine and human–machine interface.
The UKHO said that, by solving these challenges, the research will allow autonomous vessels to operate more independently, responding immediately to RNWs and SDs without waiting for human interpretation.
The research programme will culminate in a live on-water demonstration in the spring of 2026, using Zero USV’s 12-metre unmanned surface vehicle equipped with Marine AI’s software suite on Plymouth’s waterways, alongside advanced simulation events.