Application of reinforcement learning based on curriculum learning for the pipe auto-routing of ships
Youngsu Kim, Kyungho Lee, Byeong-Wook Nam, Youngsoo Han
Abstract
Abstract The pipe routing of ships has been manually performed by experts, and the design quality depends on the competence of the experts. Therefore, studies on pipe-routing automation and optimization are required. In addition, the pipe-routing task in a ship that requires frequent pipe-routing modifications requires a long time to be optimized. In this study, we developed a methodology that enables a rapid response in situations where frequent pipe-routing modifications are required by applying curriculum learning that can be stably learned by gradually solving easy-to-complex problems. In addition, this study aimed to minimize the length of the pipe and number of bends as an objective function. Finally, the proposed methodology was verified by comparing it with existing studies that used the A*, jump point search, and reinforcement-learning algorithms to determine the search speed, number of bends, and length of the path.