Human-to-robot skill transfer via teleoperation

Abstract

Programming robots to work in unstructured, changing environments, has proven to be difficult. Humans, however, effectively function in such environments. We propose a framework for programming robotic tasks using teleoperation based human-to-robot skill transfer. We assume that there exists a human expert who can accomplish a task in an unstructured environment solely through teleoperation based feedback. The human expert performs the desired task a number of times while her/his input/output pairs are being recorded. This recorded data is used to construct a mapping between these sensor inputs and actuator outputs. The mapping is then used to autonomously control the robot. In this paper we propose a set of characteristics which a human-to-robot skill transfer system should have. We then summarize the system we have implemented and present the results of some experiments we have done in skill transfer in unstructured environments.