A Consideration of Unmanned Sailing by Reinforcement Learning with Particle Filter

Abstract

Sailing is a reinforcement learning problem. The only energy source of sailboat is wind. Decision making is need in partially observable and unsteady environment. State and action spaces consist of continuous variables such as appearance wind, target position and sail direction. Some of them are vectors. This paper shows reinforcement learning simulation with particle filter which aims to learn human sailor's knowledge acquisition of predicting and utilizing wind.