Figure · 2026-05-08 · major
Figure Helix-02 — Two Humanoids Reset a Bedroom in Under Two Minutes With a Single Vision-Language-Action Policy
Figure shows two F.03 humanoids opening doors, hanging clothes, taking out trash, and making a bed together — all driven by one shared Helix-02 vision-language-action net with no central planner, no message passing, just pixels and motion.
Two F.03 humanoids share one neural net and make a bed together — without any messages between them.
Key specs
| Robots | 2 |
|---|---|
| Task time | under 2 minutes |
| Policy | single shared VLA |
What is it?
A new bedroom-tidy demo from Figure starring two Helix-02-equipped F.03 humanoids resetting a staged bedroom together in under two minutes. They open doors, hang clothes on hooks, place headphones on a stand, close a book, push a chair, take out trash, and jointly lift, place, and smooth a comforter to make the bed.
How does it work?
Both robots run the same single learned Vision-Language-Action policy directly from camera pixels to joint actions. There is no shared planner, no message passing, and no central coordinator. Each robot infers its partner's intent purely from visual observation of motion — Figure compares it to two people folding a sheet without speaking.
Why does it matter?
Multi-humanoid manipulation has typically relied on hand-coded coordination or task-specific controllers. Figure claims this is the first demo of a single neural net performing collaborative locomanipulation across two humanoids straight from pixels, which is a structural shift in how household-robot teams could scale.
Who is it for?
Robotics researchers, embodied-AI engineers, and anyone tracking humanoid VLA progress.
Try it
https://www.figure.ai/news/helix-02-bedroom-tidy