Real-time path planning for unmanned sailboats using wind prediction and reinforcement learning

Le résumé fourni par la source

In most of path planning methods for unmanned sailboats, the wind speed and direction information is obtained from the weather forecast system and the path is planned in advance. However, the rapid change of the environment and the instability of the communication signal make the path planned in advance is not the optimal path. To solve this problem, this paper proposes autonomous wind speed prediction and path planning by unmanned sailboats. As the sailboat travels, wind speed information is updated timely and the path is replanned subsequently. At first, a novel Temporal Self-Attention Network (TAN) is constructed to forecast the wind speed. Then, the wind speed and direction information is decomposed into wind speeds in both directions and predicted separately. In this case, the wind speed at the current location and the historical wind speed at the next location will be used to predict the wind speed at the next location. Subsequently, the Proximal Policy Optimization (PPO) algorithm is used for path planning and incorporates two key innovations, including an aggressive policy for rapid convergence to the destination in early training and a prioritized experience replay mechanism that improves training efficiency by reusing high-quality experiences. To align with the motion characteristics of unmanned sailboats, the reward function and the action space are specifically designed. Finally, the simulation is conducted to validate the effectiveness and superiority of the proposed scheme.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.