Environment-adaptive automatic ship collision avoidance system
Abstract
Maritime collision accidents remain a critical safety concern despite advances in electronic navigation technology, particularly under adverse weather conditions where reduced visibility and heavy seas substantially degrade a vessel's maneuvering capability. Existing automatic collision avoidance systems typically rely on static ship hazard domains and risk indices that do not incorporate environmental variables, leading to insufficient safety margins in realistic open-sea operation. This paper proposes an integrated environment-adaptive automatic ship collision avoidance system built upon three tightly coupled components: an Automatic Collision Risk Assessment System (ACRAS), a Dynamic Ship Hazard Domain (DSHD) model, and a COLREGs-compliant avoidance algorithm based on a reward-penalty cost function. The ACRAS employs a five-stage hierarchical fuzzy inference architecture containing 160 rules across five functional modules, which simultaneously processes meteorological inputs (wind force, wave height, visibility, and day/night condition) and geometric encounter parameters to produce a normalized Dynamic Risk Factor (DRF) and Final Collision Risk (FCR). The DSHD uses the DRF to dynamically expand the vessel safety boundary with the expansion factor α. The avoidance algorithm employs a three-term cost function comprising course deviation, maneuver stability, and COLREGs penalty terms, combined with a Highest Collision-Risk Ship Candidate priority mechanism to handle multi-vessel encounters. Simulation results demonstrate that the system avoids all five target vessels with a peak rudder deflection of ±23.8° while complying with COLREGs Rules 13, 14, 15, 17, and 8, and adapts autonomously to adverse weather without parameter adjustment. The proposed framework provides a computationally efficient and COLREGs-compliant solution for autonomous ship collision avoidance in variable environments. Highlights A hierarchical fuzzy engine fuses weather and geometric risks into a single index Cascaded inference cuts the rule base from 28,000 to 160 expert-verifiable rules A dynamic risk factor expands the ship hazard domain up to twofold in real time A COLREGs-integrated avoidance algorithm yields smooth and auditable maneuvers Simulations on a KVLCC2 confirm multi-ship avoidance and adverse seas