Many MindedNVIDIA opens a new robot model and training toolkit to developers
On November 9, 2026, at the Conference on Robot Learning in Austin, Texas, NVIDIA announced GR00T 2, an updated open world action model, and Isaac Lab 3.0, an open-source robot learning framework. GR00T 2 was trained in part on 100,000 hours of NVIDIA open-source egocentric data.
NVIDIA says GR00T 2 completed 40 of 50 evaluated tasks at least once without task-specific demonstrations. It also reports an average 71% success rate across those 50 tasks with one demonstration per task, rising to 95% with five. NVIDIA says the model ranks in the top three on the RoboCasa365, RoboLab, RoboDojo and MolmoSpaces benchmarks, with several leaderboard wins. Nexuni is testing it for robotic laundry, Foxconn for server-manufacturing automation, and Siemens is evaluating it for industrial uses.
Isaac Lab 3.0 lets users switch between the PhysX and Newton physics solvers and adds a reinforcement post-training library for foundation models such as GR00T. Newton integration lets it model deformable objects like cables and cloth, and multi-joint robotic hands. On one vision-based dexterous manipulation benchmark, it reportedly reached up to 2.5x the training throughput of Isaac Lab 2.3 with PhysX. Separately, AlphaNIST, built in Isaac Lab, learned all 20 NIST Assembly Task Board tasks with one policy, using about 39x less GPU time than per-task policies.
This matters for abundance because the expensive part of general-purpose robots is teaching them new work. Open models that need few demonstrations, plus faster open simulation, could cut that teaching cost for makers of goods. Fewer demonstrations and less GPU time are the mechanisms to watch.
The caveats are real. The performance and ranking figures are NVIDIA's own claims, and no pricing or cost data has been released. The Isaac Lab-Arena 0.4 evaluation framework is only a developer preview. Independent testing will show how well these results hold up.
Source: The Robot Report