Natura-1: 50B MoE foundation model drives dexterous hands zero-shot from plain language
stingning · x · 2026-10-13
Robotics startup Naturewill introduced Natura-1, billed as its first general foundation model for the physical world:
- 50B MoE, pretrained on 1M hours of egocentric human experience.
- A single model that follows plain-language instructions zero-shot, driving 20-DoF dexterous hands in a 100ms loop.
- Scaling human data from 44K to 1M hours cut action MSE on held-out robot data by 67.7%.
- Built on in-house data and compute infra, it handles fine-grained long-horizon tasks, including a 30-minute continuous zero-shot interaction in one take.
Related event: Nature Will Unveils Natura-1, a Foundation Model for the Physical World(9 posts)→
More from Embodied
- Google's Gemini 4 Argon for robotics: simulation training lifts task success from 56% to 75% — DynamicWebPaige · 2026-10-14
- Founding engineer leaves YC robotics startup to build his own robot from scratch — natesiggard · 2026-10-14
- "GPT-6 Astra" demo claims 25 sim environments, 25 tasks, 5 robot embodiments from 3 prompts in 24h — abhishekunique7 · 2026-10-14
- Dyna-2: million-hour world-action model lets robots learn tasks from 45 minutes of demos — Latent Space · 2026-10-14
- claude-as-cad: how to build real robots and parts without knowing CAD, by having Claude build its own CAD interface — _krishna_murthy · 2026-10-14
- Chronos Policy conditions robot control on real physical timestamps, not sequence indices — SongShuran · 2026-10-14