Success-Guided Sampling: zero real-world data for dexterous robot hands
milesmacklin · x · 2026-10-10
The team releases Success-Guided Sampling (SGS), arguing the missing piece for dexterous manipulation wasn't a better RL algorithm but a small tweak to the task sampler.
- Policies trained with zero real-world data, no demonstrations, no per-task reward tuning — just PPO at scale
- Robots mesh gears, thread nuts, and insert pegs zero-shot from RGB input
- The same recipe works for locomotion across much harder terrains
- Live policies available to try; all demo videos shown at 1x speed
More from Embodied
- NVIDIA releases Hebero: GPU-parallel multi-task robot RL benchmark spanning 40 tasks — nvidia · 2026-10-10
- Artificial Analysis teases AA-Robotics: frontier models zero-shot robot arm control — ArtificialAnlys · 2026-10-10
- Starkey Unveils Omega AI+ Hearing Aids With G4 Gen AI Processor and Quad DNN — BrandonSawalich · 2026-10-10
- Wayve CEO demos AI Driver with Stellantis in a Fiat 500e on Turin's tight streets — alexgkendall · 2026-10-10
- 192GB Framework Desktop Batch 2 sells out; 128GB still in stock — film_girl · 2026-10-10
- Danu Robotics' six-year fight to build a better recycling robot — TechCrunch AI · 2026-10-10