CLONE

A humanoid whole-body teleoperation system that tracks only the head and hands, correcting drift in closed loop.

CLONE was proposed in June 2025 by teams at the Beijing Institute for General Artificial Intelligence (BIGAI), Peking University, and Beijing Institute of Technology. Earlier humanoid teleoperation often controlled the upper and lower body separately to stay stable, producing uncoordinated motion; robots also drift further from the operator's actual position the longer they walk (accumulated drift). CLONE captures only the operator's head and hand poses with an Apple Vision Pro headset, and a mixture-of-experts (MoE) whole-body policy generates whole-body joint actions for a Unitree G1, with the robot's actual global position fed back to the policy in real time for closed-loop drift correction. The paper reports an average position error of about 5.1 centimeters over 8.9 meters of straight-line walking, and it can perform long-horizon tasks requiring whole-body coordination, such as squatting to pick something up off the floor. Systems like this are also a basic tool for collecting demonstration data on humanoid robots.

ExampleAn operator wearing an Apple Vision Pro walks around a room and bends down; the G1 walks over and squats to pick an object up off the floor, following along.

Also called

CLONE: Closed-Loop Whole-Body Humanoid Teleoperation for Long-Horizon Tasks

Related

Whole-Body Teleoperation · Mixture of Experts · Unitree G1 · Apple Vision Pro · Motion Tracking · OmniH2O

Sources

CLONE (arXiv 2506.08931)
CLONE 论文 HTML 版 (arXiv 2506.08931v2) (Chinese)

As of

2025-08