Learning robotic rotational manipulation skill from bilateral teleoperation

Abstract

Bilateral teleoperation is an efficient and powerful solution for conducting manipulation tasks through the robot in remote or hazardous environments such as work in height situations, or radioactive hot cells. However, performing repetitive manipulation tasks through bilateral teleoperation induces heavy human workload. The common repetitive and difficult task in real teleoperation scenario is rotational manipulation task. Therefore, we propose a framework for learning rotational manipulation skill from bilateral teleoperation, with a single demonstration. The skill of conducting rotational manipulation task should include trajectory information, as well as physical interaction information. Therefore, we first present a method to learn trajectory and physical interaction information for the task. Afterwards, we describe a method to successfully perform the task to the new target goal, along with generated trajectory and learned physical interaction information. Additionally, we explain the way to apply human-autonomy shared teleoperation system with proposed framework. The results show that failures caused by estimation error of the target goal can be successfully overcome with minimal human operator intervention through shared teleoperation system with proposed framework.