Nvidia's GR00T 2 robots complete 97/100 toy assemblies, recovering from slips and stalls
mark_k · x · 2026-11-10
Nvidia tested its GR00T 2 system on three different toys, requiring 100 consecutive assemblies each — the robots completed at least 97 per toy, recovering from stalls and failed grasps along the way.
The key mechanism is visual closed-loop planning: GR00T 2 predicts what the robot's cameras should see after its next move, executes it, compares prediction with reality, and uses the actual view to plan the next step. A bad grip becomes a second attempt instead of ending the run.
Demos are published; code is expected in December.
Related event: NVIDIA Unveils GR00T 2 World-Action Model for Robots at CoRL(14 posts)→
More from Embodied
- Will robotics models run 100x faster in months? A 3-part speedup breakdown, and the latency question — micoolcho · 2026-11-10
- 10,000-hour vision-tactile dataset drops as robot data gold rush fuels shovel sellers — CyberRobooo · 2026-11-10
- Team fine-tunes Dream-1 with just 50 examples for a USB-unplugging robot task — IsaiahBallah · 2026-11-10
- Tesla to wind down Cybertruck output at Giga Texas to scale Cybercab robotaxi production — XFreeze · 2026-11-10
- Norway runs 2,000+ underwater AI laser robots delousing 50 million salmon — jjvincent · 2026-11-10
- Stanford tennis courts are so 'bottmaxxed' that bots are lapping each other — jamdac · 2026-11-10