First-Person Teleoperation of a Bimanual Robotic System

Abstract

First-person teleoperation of robots is a large field of research that could serve many benefits for automation. Teleoperation is a popular method to collect demonstrations for imitation learning that are easily learned by the robot, and thus it’s important to create teleoperation systems that are intuitive and enable human-like perception of a scene. Adding a first-person component to basic teleoperation systems is key to improving operators’ visual perception and making teleoperation possible for extended periods of time. Existing teleoperation systems do not integrate elements that provide the operator with a good perception of the task space, such as a first-person VR view and the ability to leverage the neck to search around the space. They rely on techniques such as third-person view of the space, or provide a first-person view but without the ability to move the neck to look around. This thesis proposes a VR-based teleoperation system with an actuated 5-DoF neck for enabling human-like perception and improving the ability to perform high quality demonstrations for use in imitation learning.