Meet OpenWBT: Open-source VR teleop for whole-body humanoid control across real and virtual worlds!

In this talk

1. 0:14

   Expand the robot’s workspace

   Jen says upper-body-focused teleoperation limits what humanoids can reach. Squatting and leaning forward could expand their usable workspace and enable more tasks. A key challenge is maintaining dynamic balance as the robot’s center of mass moves.

2. 0:44

   Introducing OpenWBT

   Jang describes OpenWBT as an open-source system for whole-body VR teleoperation across robot types and environments. The setup uses a laptop and VR headset, and she says beginners can build a system within hours. It is intended to collect teleoperation data for training body AI models.

3. 1:23

   Hierarchical control with R2S2

   Jang attributes the system’s robustness to R2S2 whole-body control. Its design combines a low-level library of transferable primitive skills with a neural latent skill space. That space coordinates body parts and transitions between skills.

4. 1:48

   Transfer across humanoid platforms

   Jang says the system runs on the 29-unit G1 and transfers to the adult-height Unitree H1. The H1’s greater size and weight make stability margins smaller and its dynamics more strongly coupled.

5. 2:04

   Open-source collaboration

   Galbot has open-sourced OpenWBT and plans to keep updating it. Jang invites developers, researchers, and robotics enthusiasts to test the system, share feedback, and contribute to humanoid robotics development.