An Intelligent Decision System for Autonomous Ship Collision Avoidance with Real Ship Validation

Abstract

Practical deployment of autonomous ship collision avoidance requires high-level decisions that can be executed stably by onboard motion controllers. This paper presents an engineering-oriented intelligent waypoint decision system for autonomous ship collision avoidance. The system receives ownship and target-ship states from navigation and perception interfaces, combines current encounter geometry with predicted maneuver intention, and selects the dominant threat. Candidate temporary waypoints are screened using a weighted multiobjective criterion that jointly considers collision safety, COLREGs compliance, route efficiency, and decision continuity. A hysteresis finite-state machine coordinates normal navigation, collision avoidance, and route recovery to reduce frequent target switching and waypoint jumps. The system was integrated with the Chuangzhi experimental vessel. For controlled and repeatable validation, virtual AIS targets were injected into the onboard perception and decision chain, while the generated waypoints were tracked by the real vessel motion controller. Four representative one- and two-target closed-loop tests were conducted. The recorded minimum encounter distances ranged from 471.3 m to 867.6 m and remained above the predefined 400 m safety threshold. The results demonstrate the closed-loop executability and engineering feasibility of the proposed decision system under the tested virtual-target encounter configurations.