Reinforcement Learning-Based Skill Acquisition Study for Linearly Heated Robots

Abstract

With the development of science and technology, the manufacturing industry is undergoing a quiet transformation, with artificial intelligence and intelligent robots gradually entering the smart manufacturing. Currently, the traditional shipboard manufacturing industry is undergoing a baptism of robotics and ushering in profound changes to meet the growing demand for productivity, quality and safety. However, traditional industrial robots are limited to preprogrammed movements and lack the ability to perform multiskilled tasks. In particular, programming industrial robots for complex hull steel plate machining processes, especially multipass machining, is challenging. In order to allow the robot to learn skills like a human apprentice, it can learn the skills related to linear heating of ship plates from experienced workers and make autonomous judgments to complete the hull steel plate machining according to the requirements. This study focuses on enabling machines to acquire workers' skills and plan heating lines for ship hull steel plates. We achieve this through demonstrative learning and maximum gradient-based reinforcement learning. It can be seen through data simulation and experiments that the linear heating robot in this study achieves 83.26% accuracy in planning the heating line. This not only provides support for workers in the complex linear heating process, but also further improves workers' operational efficiency and economic benefits.