A robust COLREG Meta-Classification system for autonomous ships in two-vessel encounter scenarios

The advent of Maritime Autonomous Surface Ships raises many concerns about the safety of navigation, especially concerning how COLREG-compliant collision avoidance activities should be performed. To address these issues, autonomous navigation systems make use of COLREG Classification systems that detect the rules that apply to a given scenario, and proper algorithms that plan and execute evasive manoeuvres. In the literature, there is a third element that determines when the COLREG Classification system should be triggered and how its output should be handled by the collision avoidance algorithm. This paper proposes the term Meta-Classification system to indicate this third element. The approach to Meta-Classification that makes use only of the current features of an encounter scenario is named by this paper Static COLREG Meta-Classification. Recent works in the literature have identified the issue of Classification Instability in these approach. To solve the issue of Classification Instability this paper proposes a novel approach to Meta-Classification named Dynamic COLREG Meta-Classification. This approach introduces a memory effect which increases the robustness of the classification process in two-vessel encounter scenarios. Static COLREG Meta-Classification and Dynamic COLREG Meta-Classification are compared by means of numerical simulations of encounter scenarios.