Key Findings
GreenWAM demonstrates that a single world foundation model can function as both an action-conditioned world model and a control policy across diverse robot bodies. In benchmarks, it attains top-tier trajectory, depth, and 3D accuracy on WorldArena 1.0, surpassing all compared video generators and world-action models. As a policy, it ranks among the top entries on the RoboTwin 2.0 leaderboard and achieves the highest zero-shot LIBERO-plus success rate among compared methods.
Technical Details
The model is developed using a three-part recipe:
- Shared Anatomical Action Space: A common joint coordinate system that describes 20 robot families, ranging from bimanual arms to dexterous humanoids.
- Joint Training: Training conducted across these families in visual, action-conditioned, and world-action modes.
- Policy Architecture: Policies are created by splitting an action expert off from the generator.
For real-time control on real humanoid robots, the model can be distilled into a single-step policy.
Background & Context
Traditionally, robot world models (which predict scene changes based on actions) and control policies (which choose those actions) are built and evaluated separately. Furthermore, different robots typically record actions in proprietary formats, hindering the creation of a unified model. GreenWAM addresses these silos by unifying the action space and training roles.
Significance & Outlook
Controlled experiments indicate that robot specialization improves both control and world prediction. The research also found that splitting off an action expert at the start of adaptation is beneficial, and maintaining world-model modes during policy adaptation preserves robustness against scene changes. In real-world humanoid applications, GreenWAM succeeds more frequently than policies adapted from Cosmos 3 Edge and $\pi_{0.5}$, suggesting a more robust path toward general-purpose robotic control.
Source: https://huggingface.co/papers/2610.18774
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime
