Scenario-Based Model Predictive Control with Different Numbers of Decision Steps for COLREGS Compliant Ship Collision Avoidance

Abstract

The main question investigated is whether additional decision steps can improve vessel behavior produced by the collision avoidance method scenario based model predictive control (SBMPC). The method, which functions by predicting alternative paths resulting from a finite number of alternative control behaviors, then selecting which behavior to apply by use of a cost function, was originally formulated to allow switching between several behaviors on the prediction horizon. However, current implementations have been limited to a single control step. To compare the single-step and multi-step SBMPC, a simulation study was performed, where different configurations for the number, positioning and possible control actions were tested. In the course of the simulation study it became clear that identifying situations producing a significant difference between the two methods was difficult to identify and the multi-step SBMPC led to only minor improvements in very few scenarios. Nevertheless, multi-step decisions can be visualized to give better situational awareness, and also have additional benefits with other trajectory parameterizations and less uncertain predictions of other ship trajectories.