Acquisition of Human Expertise in Robotic Assembly

Abstract

Robots must be presently taught by human workers to execute given manufacturing tasks. The current problem is that the task of teaching robots is rather time-consuming, especially within the robotic assembly domain. This problem is caused by insufficient accumulation of human expertise that should be reused in this domain. Therefore, a knowledge-intensive method for acquiring human expertise is proposed in this paper. Our method is able to acquire human expertise in the robotic assembly domain by observing robot-teaching demonstrations of human experts. What distinguishes our method from others is that there are two modes of learning: 1. learning from an example directly given by human workers, and 2. learning expertise on error recovery by observing revisions made by human workers in handling execution errors that occur in reusing previously acquired knowledge. The acquired human expertise is required to be represented in a way that satisfies two requirements. The first one is operability so that the representation is easy to transform into robot programs (commands & parameters). The second one is understandability so that the representation is easy for human workers to understand the robot program. A specific robot assembly example is given to illustrate the proposed method.