Natura-1 debuts: 50B-param MoE world model trained on 1M hours of egocentric video
stingning · x · 2026-10-13
naturewill introduced Natura-1, its first general foundation model for the physical world.
- Scale & data: a 50-billion-parameter MoE model pretrained on 1 million hours of egocentric human experience
- Training recipe: large-scale pretraining for world knowledge, followed by efficient post-training on a limited amount of robot data to adapt to dexterous manipulation
- Capabilities: zero-shot natural-language instruction following with reasoning over spatial constraints, multi-object relationships and negative conditions; autonomous long-horizon workflows
- Generalization: the team reports generalization to unseen scenarios and early evidence of emergent behavior
A notable data point for the "video pretraining + robot post-training" paradigm, worth comparing against π and Figure's approaches.
More from Embodied
- Humanoid hand wraps wontons as Nature Will releases Natura-1, a foundation model trained on 1M hours of human data — CyberRobooo · 2026-10-13
- Natura-1 runs multi-minute autonomous dexterous hand workflows with a single policy — stingning · 2026-10-13
- Natura-1 demo: a single-take 30-minute recording of everyday human-robot interaction — stingning · 2026-10-13
- Collov Labs launches NewEyes AI: camera glasses plus open-ear earphones give agents eyes — CodeByPoonam · 2026-10-13
- Surface Ultra unboxed: NVIDIA RTX Spark N1X chip, 20 cores, 64GB unified memory — TinfoilTricorn · 2026-10-13
- Replicating laundry real2sim in MuJoCo and training RL WBC in 3 hours with Opus 5.5 — ZeYanjie · 2026-10-13