AlphaNIST: one RL policy solves all 20 NIST assembly tasks with 98.2% success in simulation
hbouammar · x · 2026-10-15
- A team that previously showed robots learning precise assembly and hard locomotion purely in simulation—zero real-world data, no demos—has released AlphaNIST: a single policy trained from scratch solves all 20 NIST board assembly tasks at once, with 98.2% success in simulation.
- Highlights: ten different parts assembled in sequence with a single terminal reward; emergent, unprogrammed techniques like varied nut-turning motions in tight gaps, reorienting parts against the environment, and self-chosen task ordering.
- The approach mirrors AlphaZero: no prescribed manipulation stages, just a binary success reward plus vast diverse simulated experience. Parts as small as 4 mm, very tight clearances. Paper is out; code coming soon.
Related event: AlphaNIST: One Policy Masters 20 NIST Assembly Tasks with Binary Rewards(3 posts)→
More from Embodied
- 1X, DeepMind, Physical Intelligence Gather for Humanoid Robot Demos — lukas_m_ziegler · 2026-10-15
- NVIDIA recaps GTC Berlin: open models to physical AI — NVIDIA Developer · 2026-10-15
- Neural field planning for hybrid manufacturing cuts estimated build time 68.3% — jiqizhixin · 2026-10-15
- Leash Bio: DNA-encoded libraries and an RL sandbox for 10,000 drug-discovery campaigns a weekend — Latent Space · 2026-10-15
- Work photos are AI's first mile: if capture is wrong, no model can recover — arnabdotorg · 2026-10-15
- LEGO builds an omnidirectional vehicle for automated warehouses — glenbeer · 2026-10-15