Stanford's Wetzstein endorses finding: web video pretraining scaling boosts real-world robot manipulation
GordonWetzstein · x · 2026-09-11
Gordon Wetzstein endorsed Tongzhou Mu's new research thread, calling it "a big step towards bringing robots outside the lab" into real-world useful tasks.
Key findings:
- Scaling pre-training on general web video substantially improves a complex real-deployment manipulation task
- The team scaled both model size and pre-training compute, testing on one industrial task
- The better a pre-trained model predicts web video, the better its post-trained policy performs
The result suggests web video prediction quality is a usable predictor of downstream robot policy capability.
More from Embodied
- RealSense D585 Pro AI-Powered 3D Stereo Depth Camera Unboxed for Physical AI — chrismatthieu · 2026-10-26
- Physical world agent swarm demo shared on X — pduan · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26
- 'Lamp Swarm': a desk of lamp-shaped robots shows what physical AI work looks like — dee_hw · 2026-10-26
- Tencent's HiSTEVE 1.0: hierarchical VLA agent plays Minecraft with native keyboard and mouse, hits 85% SOTA — _akhaliq · 2026-10-26
- SJTU's VaM lets robots learn online from mistakes, 95%+ success over 4-hour autonomous runs — chris_j_paxton · 2026-10-26