NVIDIA open-sources Isaac Video to enable Real-to-Sim-to-Real robot learning workflow
Scobleizer · x · 2026-08-22
NVIDIA has released the open-source NVIDIA Isaac Video project, aiming to bridge the gap between real-world demonstrations and robot training through a Real-to-Sim-to-Real workflow.
Key capabilities include:
- Video Ingestion: Segmenting long videos and retrieving relevant action clips automatically.
- Reconstruction: Reconstructing hand/body motion, depth, meshes, and 6-DoF trajectories.
- Retargeting: Mapping human motions to robot embodiments for RL policy training in Isaac Lab.
- Dexterity: Scaling dexterous manipulation using Contact Wrench Guidance From Human Demonstration (CHORD).
The release also provides access to a library with 1,215 tasks and 121,500 robot episodes.
More from Embodied
- TRIC robots autonomously patrol strawberry fields at night for pest control — TinfoilTricorn · 2026-08-22
- Tesla Cybercab Official Launch Date Set for September 3 — mitchdeg · 2026-08-22
- Tesla Robotaxi removes steering wheel to cut weight and operating costs — Scobleizer · 2026-08-22
- TARS unveils AWE 3.5, an embodied-native model with 'Born as One' architecture — rohanpaul_ai · 2026-08-22
- Enigma Robots Perform Real Chemistry Experiments Online — Scobleizer · 2026-08-22
- comma.ai demo shows large model improving autonomous driving navigation — Scobleizer · 2026-08-22