Autonomous Surface Craft Continuous Control with Reinforcement Learning

Abstract

Modern vessel is a sophisticated construction requiring highly prepared and experienced engineers to control it and to handle its underlying systems. Navigation, especially in restricted areas is a tough task which expects from duty navigational officer knowledge of International Regulations for Preventing Collisions at Sea as well as own vessel characteristics, permanent attention to continuously evolving environment and general vigilance. No doubts that autonomous system pretending to “take over the watch” is obliged to be a cut above the human capabilities. In this study we apply a deep reinforcement learning (RL) algorithm to control maneuvering of a surface craft. RL agent controls craft engines power in order to move it in indicated position avoiding objects considered as obstacles.