Distilling World-Model Features into VLAs: 0.8B Policy Hits 97.9% on LIBERO
Trung Dao · hf · 2026-09-22
VLA models map observations to actions with no objective for how the world responds, capping robustness at data coverage; world models carry that objective but cost seconds per roll-out, ruling them out of the control loop. This work shows the two can be decoupled: a world model's scene knowledge lives in its internal features; generating the future is merely the objective that shaped them. The method adds one feature-alignment term to ordinary VLA training — a frozen world model runs over training frames once and is cached, and the student learns to match the cache. No teacher during training, projector discarded after; the deployed policy is identical to the undistilled baseline, running in 32 ms with 1.86 GB on an RTX 5090.
Results: a 0.8B student reaches 97.9% on LIBERO, improves 48.2%→50.5% on RoboCasa-GR1 humanoid manipulation, and transfers to single-arm and bimanual real hardware. Gains survive changes of student scale, backbone, alignment layer, and teacher — a broad representational prior, not a fragile pairwise alignment.
More from Embodied
- Team claims a robot priced like a modern smartphone, presale buyers get the best deal — R4nsack · 2026-11-12
- CoRL results show sim-only training now delivers top real-world robot performance — carlosdponx · 2026-11-12
- Robot fails at folding a towel, reassuring roboticists frontier models won't eat robotics yet — chris_j_paxton · 2026-11-12
- Nuro podcast: manufacturing, charging, and regulation gate robotaxi mass adoption — R_D · 2026-11-12
- Rodeo Robotics demos robot that changes your car tire — ycombinator · 2026-11-12
- Cybercab rear-ended by human driver, researcher quips: the problem is the humans — chris_j_paxton · 2026-11-12