Path optimization for a marine vehicle using reinforcement learning

Abstract

This study proposes a global path planning algorithm for a marine vehicle, which considers the dynamic characteristics of the vehicle and disturbance effects in ocean environments. In contrast to path planning on land, various environmental disturbances need to be considered during path planning in ocean environments, such as wind, waves, and currents. In the current study, the effects of ocean currents are the primary consideration in the current study. A kinematic model is used to simulate realistic vehicle motion, which limits the use of conventional search-based optimization algorithms such as A*. Thus, a reinforcement learning algorithm is employed for path optimization. The proposed algorithm determine a near optimal path between the start and goal points when the map and current field data are provided. Conventional path planning algorithms and their limitations in ocean environments are discussed. The new path planning algorithm approach based on reinforcement learning is then introduced. To verify the optimality and validity of the proposed algorithm, numerical simulations are performed in artificial and actual environmental conditions, and their results are presented.