From Linguistic ambiguity to quantification: Expert-Based Fuzzy representation of COLREG terminology

The International Regulations for Preventing Collisions at Sea (COLREGs) rely on vague terms, which challenges both human navigators and autonomous systems requiring computable representations. This study quantifies what experienced navigators consider a safe distance under the COLREGs, using fuzzy representations derived from expert judgement. In a structured international survey of experienced navigators, 194 of the 235 respondents supplied useable data on Closest Point of Approach (CPA) and Bow Crossing Range (BCR) in open sea conditions. These data were used to derive fuzzy membership functions. Results reveal that navigators require a larger margin ahead of the bow than at the closest point of approach, with a mean Medium BCR of 1.86 NM against a mean Medium CPA of 1.42 NM. The influence of rank, experience, ship type and ship length on the elicited values was tested; only experience produced significant differences, and the model output proved insensitive to them. A weighted fuzzy rule base was built from a separate panel of ten experts, who assessed navigational scenarios on a full mission simulator, and used to construct a fuzzy inference model of safe distance for encounters in which the target crosses ahead of own ship. The model was validated by leave-one-out cross-validation and against data from seven collisions and three close-quarters situations. The novelty lies in deriving the membership functions from a large expert panel rather than assuming them, in building the rule base from assessments of navigational scenarios rather than stated rules, and in validating the model against real collision data. This provides a quantified, human-centred contribution for decision-support systems.