A Voice-Controlled Motion Reproduction Using Large Language Models for Polishing Robots

Abstract

In recent years, the shortage of professionally skilled people in industrial fields has been a major social problem. To solve this problem, the transfer of skills to robots has been attracting much attention. However, they are not familiar with robot control, and hard to teach robots their skills by numerical commands or program source code. For more user-friendly human-robot interaction, a lot of studies have been conducted. In previous researches, robot task processes are pre-defined and not changed in task execution. We developed a robot system using the motion-copying system and GPT-3, one of the Large Language Models. This system can not only copy the motion but also modify saved motion in execution by using natural language commands. We evaluated the proposed system by applying it to polishing robots and confirmed that the surface of used workpieces was changed following to input commands.