Being-H0.8 adds tactile prediction to robot world models with 500,000 hours of video
量子位 · wechat · 2026-07-28
BeingBeyond released Being-H0.8, an implicit tactile world-action model built from human video data.
Key points:
- The model adds tactile signals to the latent world model so robots can predict contact, not just motion.
- It uses more than 500,000 hours of first-person video data, plus a large data pipeline for filtering, motion recovery, and language annotation.
- TactoHand reconstructs contact and proximity labels from video, while pressure-sensor data adds physical force information.
- A universal tactile encoder normalizes different tactile modalities into shared tokens.
- TopoHand aligns human hands, dexterous hands, and grippers into a unified action representation.
In real robot tests, the system handled bag retrieval, calligraphy, toothpaste squeezing, and snack grasping — all tasks that depend heavily on touch feedback.
More from Embodied
- iPhone Duo preorders sell out in 3 minutes around NYC, deliveries slip to late November — BenBajarin · 2026-10-16
- UMD's PAIR framework bridges perception and action in VLA models via bridge tokens — Kaixi Feng · 2026-10-16
- CI-MSE offline metric hits -0.87 Spearman correlation with real robot task success vs weak raw MSE — chris_j_paxton · 2026-10-16
- List of 15 best robotics researchers on X, from Jim Fan to Chelsea Finn and Fei-Fei Li — FinanceYF5 · 2026-10-16
- Fake Apple listing for dual-screen 'iPhone Duo' jokes it's the best phone for watching your AI agents — AIFlow_ML · 2026-10-16
- Inside Auki's Hong Kong office: a fake grocery store powering robotics data collection — LoganGrasby · 2026-10-16