Development of Deep Reinforcement Learning Methodology for Co-bot Motion Learning

Today, Korea is facing a time when it is essential to develop new manufacturing technologies and strategies to lead new changes, such as smart factories and manufacturing innovation 3.0, and achieve continuous development of the domestic manufacturing industry. Therefore, many manufacturing companies are promoting process automation using collaborative robots (co-bot) to respond to the paradigm of multi-item, small-volume production. The emergence of co-bots improves the space utilization of production facilities and opens up the possibility of introducing robots without modifying the existing production line. This study aims to conduct primary research on a robot that recognizes and acts on its environment using reinforcement learning to determine its work movements and perform tasks without specific instructions from human experts. In this study, we propose a collaborative robot control methodology using a deep reinforcement learning algorithm. In addition, for the practical application of the HRC system, which is challenging to apply to the production of a single product, the problem of data sharing between collaborative robots and workers based on a process model was addressed. The system proposed in this study is designed to optimize process variables through artificial intelligence-based data learning and is expected to contribute to product and process quality optimization of human-robot collaborative processes in the future.