Open-TeleVision: Teleoperation with Immersive Active Visual Feedback

Authors: Xuxin Cheng, Jialong Li, Shiqi Yang, Ge Yang, Xiaolong Wang Submitted: July 2024 Venue: arXiv 2407.01512

Abstract

An immersive teleoperation system that allows operators to actively perceive a robot’s surroundings in a stereoscopic manner. The system replicates the operator’s arm and hand movements onto the robot — “the operator’s mind is transmitted to a robot embodiment.”

System Design

  • Stereoscopic cameras on robot head stream to VR headset → operator sees from robot’s perspective
  • Arm/hand movement captured from operator → retargeted to robot joints
  • Active visual feedback: operator can control robot head orientation for better viewpoint
  • Works across different humanoid robot platforms

Validation

Four complex tasks evaluated:

  1. Can Sorting
  2. Can Insertion
  3. Folding
  4. Unloading

Tested on two different humanoid robots. Data collected via this system used to train imitation learning policies with strong results.

Fully open-source.