Human-Robot Collaborative Reinforcement Learning in Semi-Automated Manufacturing Operations

Abstract

The progress in efficient human-robot partnerships has influenced all facets of contemporary industry. A considerable proportion of manufacturing procedures, namely the final assembly and trimming processes, continue to be semi-automated since human agents possess superior skills in performing these tasks. Manufacturers have been progressively integrating augmented reality (AR) into their operations to equip, instruct, direct, and enhance the abilities of their less-experienced staff. As the proficiency of augmented human agents increases, their contact with the robot agents likewise undergoes dynamic modifications. This paper introduces a human-robot collaborative reinforcement learning model (HR-CRL) that enhances the decision-making power of robotic agents. The method considers the changing observable information from the production system environment and the input from human operators. The HR-CRL model is assessed across many scenarios of semi-automated manufacturing activities, which replicate fluctuations in augmented human performance and the environment. The evaluation showcases how robot agents adapt their behavior in response to observable data, enhancing the efficiency of the human-robot collaboration.