Performance Metrics for Human-Robot Collaboration: An Automotive Manufacturing Case

Abstract

A human-robot collaborative system in the form of a power and skill assist robotic system was developed where a human and a robot could collaborate to perform object manipulation for targeted assembly tasks in automotive manufacturing. We assumed such assembly tasks as the representative assembly tasks in automotive manufacturing. We reflected human's weight perception in the dynamics and control of the power and skill assist system following a psychophysical method using a reinforcement learning scheme. We recruited 20 human subjects who separately performed assembly tasks with the system in human-robot collaboration (HRC). We then observed the collaborative assembly tasks, conducted extensive literature reviews, reviewed our previous and ongoing related works and brainstormed with the subjects and other relevant researchers, and then proposed HRC performance assessment metrics and methods for collaborative automotive manufacturing. The proposed metrics comprised of assessment criteria and methods related to both human-robot interaction (HRI) and manufacturing performance. We then verified the proposed performance metrics in pilot studies in the laboratory environment using the same collaborative system and subjects. The verification results proved the effectiveness of the assessment metrics and methods in terms of usability, practicability and reliability. We then proposed to apply classification and regression type machine learning approaches under supervised and reinforcement learning setups to learn different classes and decision-making rules respectively regarding HRC performance. The proposed performance metrics and methods can serve as the preliminary efforts towards developing comprehensive assessment metrics for HRC in general and for human-robot collaborative automotive manufacturing in particular.