COLREG Scenario classification and Compliance Evaluation with temporal and multi-vessel awareness for collision avoidance systems

Abstract

This paper proposes and demonstrates two new approaches for COLREG compliance evaluation and COLREG scenario classification with application to support collision avoidance systems. The COLREGs, defined by the International Maritime Organisation, are a set of regulations that describe how vessel operators should behave to avoid collisions. As automated systems enter our waterways, it is important they adhere to the COLREGs. An overlooked challenge is when multiple vessels need to be considered simultaneously. Additionally, incorporating the scene’s temporal context can improve COLREG classification performance. The presented approaches are: (1) Temporally Aware Geometric COLREG Scenario Classification (TAG-CSC); and (2) COLREG Compliance Evaluation (CCE). TAG-CSC improves established geometry-based COLREG classification methods by proposing algorithms that account for the temporal context of a scene, apply hysteresis to maintain temporal stability of COLREG classification and provide a confidence estimate. CCE improves existing evaluation methods by proposing alternative functions for penalties based on a vessel’s deviation from the COLREGs. We also introduce a vehicle reaction time approach, a collision risk-based method to aggregate CCE scores with greater than two vessels, and uncertainty bounds of the CCE score. Real-world and simulated experiments are presented to demonstrate the efficacy of TAG-CSC and CCE in one-on-one and multi-ship scenarios. • COLREG scenario classification that extend classical methods with temporal context. • Equations to evaluate the COLREG compliance of a marine vessel for use in autonomy. • Results that demonstrate the proposed method via real-world and simulation testing.

Topics & Concepts