OPEN TEACH: A Versatile Teleoperation System for Robotic Manipulation

Authors: Aadhithya Iyer, Zhuoran Peng, Yinlong Dai, Irmak Guzey, Siddhant Haldar, Soumith Chintala, Lerrel Pinto Submitted: March 2024 Venue: arXiv 2403.07870

Abstract

Open-source teleoperation platform built on the Meta Quest 3 VR headset ($500 consumer hardware). Enables intuitive control of various robotic platforms via natural hand gestures at up to 90Hz with visual feedback.

Supported Hardware

  • Arms: Franka, xArm, Jaco
  • Hands: Allegro (multi-fingered dexterous hand)
  • Configurations: single arm, dual arm (bimanual), mobile manipulator

Performance

  • 38 tasks demonstrated
  • 10 contact-rich manipulation challenges validated for policy learning
  • 90Hz control rate with low latency
  • User studies: significant improvements over existing frameworks

Significance

The $500 hardware cost (vs thousands for professional mocap systems) democratizes dexterous teleoperation data collection. NYU group (Lerrel Pinto) is known for accessible, open robotics research.