Reactive Collision Avoidance of an Unmanned Surface Vehicle through Gaussian Mixture Model-Based Online Mapping

With active research being conducted on maritime autonomous surface ships, it is becoming increasingly necessary to ensure the safety of unmanned surface vehicles (USVs). In this context, a key task is to correct their paths to avoid obstacles. This paper proposes a reactive collision avoidance algorithm to ensure the safety of USVs against obstacles. A global map is represented using a Gaussian mixture model, formulated using the expectation-maximization algorithm. Motion primitives are used to predict collision events and modify the USV's trajectory. In addition, a controller for the target vessel is designed. Mapping is performed to demonstrate that the USV can implement the necessary avoidance maneuvers to prevent collisions with obstacles. The proposed method is validated by conducting collision avoidance simulations and autonomous navigation field tests with a small-scale autonomous surface vehicle (ASV) platform. Results indicate that the ASV can successfully avoid obstacles while following its trajectory.