Sber's GreenWAM world foundation model spans 20 robot families as both predictor and policy
SberRoboticsCenter · hf · 2026-10-14
Sber Robotics Center presents GreenWAM, a world foundation model testing whether one model can serve as both scene predictor and policy across robot bodies.
- Shared anatomical action space covering 20 robot families (bimanual arms to dexterous humanoids) in common joint coordinates, with joint training in visual, action-conditioned, and world-action modes
- As a world model: top-tier trajectory, depth, and 3D accuracy on WorldArena 1.0, beating all compared world-action models and video generators
- As a policy: top RoboTwin 2.0 leaderboard entries, best zero-shot LIBERO-plus success, competitive on CALVIN/LIBERO; adapted to a real humanoid with more successes than π0.5- or Cosmos 3 Edge-adapted policies, distilled to single-step real-time control
- Findings: robot specialization improves prediction and control together; split the action expert at adaptation start; keeping world-model modes preserves scene-change robustness
More from Embodied
- Swaayatt Demos On-the-Fly Camera-LiDAR Calibration Without Markers in the Wild — sanjeevs_iitr · 2026-10-14
- Whimsical robot creatures go viral as commentator urges consumer product version — LinusEkenstam · 2026-10-14
- Stratechery on Apple HomeView: how big is the market for pricey devices built on subpar AI? — Stratechery · 2026-10-14
- Construction workers wear head-mounted cameras to feed first-person data to AI and robots — Far_Process4191 · 2026-10-14
- Nature: smart glasses are heading for a major privacy crisis — marigo · 2026-10-14
- Impatient for JetKVM Mini, dev buys LuckFox KVM to let agents control his Mac Mini — TheZachMueller · 2026-10-14