Noematrix's Noe-0 trains embodied models with zero teleoperation data
jiqizhixin · x · 2026-09-03
Noematrix (穹彻智能) unveiled Noe-0, an embodied pretraining model whose pipeline uses no teleoperation data at all. Training data instead comes from humans naturally performing tasks in real environments—collected across 50+ cities, hundreds of thousands of hours and task types, departing from centralized data-factory setups.
Results show full-chain pretraining, post-training and real-robot execution work without teleoperation, with robust manipulation on complex, fine-grained real-world tasks—proving unscripted human operation data can ground embodied world models.
More from Embodied
- Sentdex trains robots to go from lying down to a half kneel, eyeing a full stand — Sentdex · 2026-10-21
- NVIDIA open-sources MotionBricks: 350K motion skills at 15,000 FPS, 2ms latency — aftahi_ai · 2026-10-21
- NVIDIA Cosmos3-H-Dreams: Generative Simulation Brings Clinical Surgery to Life — Hugging Face Blog · 2026-10-21
- FLoTS-SLAM uses optical-flow residuals to hit 0.73 DM-IoU in dynamic monocular SLAM — zhenjun_zhao · 2026-10-21
- Plato ring launches at $199 to turn near-silent speech into AI commands — TinfoilTricorn · 2026-10-21
- Dream-1 controlled 31 unseen robot types across 53 countries within 24 hours — gharik · 2026-10-21