Nvidia debuts GR00T 2 robotics model, 71% on 50 tasks
TL;DR
- Nvidia's 95% five-shot success rate covers unnamed tasks pre-selected for 80%+ baseline capability, structurally inflating the headline benchmark figure.
- RoboDojo's independent evaluation scores GR00T 2 at 12.8% real-world success against Nvidia's self-reported 95%, a gap that cannot be audited before December.
- Weights and code ship in December; Foxconn, Nexuni, and Siemens industrial pilots proceed now, running on Nvidia's access timeline, not an open one.
Nvidia says its new GR00T 2 world action model completed "40 of 50 evaluated tasks at least once without task-specific demonstrations," averaged 71% success across those 50 tasks with one demonstration per task, and reached 95% with five. The model was introduced at the Conference on Robot Learning in Austin, Texas. Nvidia's own write-up describes it as an "open frontier world action model built on NVIDIA Cosmos," trained on 100,000 hours of Nvidia open-source human egocentric data.
Alongside the model, Nvidia released Isaac Lab 3.0. In vision-based dexterous manipulation benchmarks the company cites, Isaac Lab 3.0 with the Newton physics solver "achieved up to 2.5x the training throughput of Isaac Lab 2.3 with PhysX." On the new AlphaNIST benchmark, a single policy covering 20 tasks hit 98% training success versus 81% for separate per-task policies, using roughly 39x less GPU time.
Two industrial benchmarks debuted with the release: DexBench, with 15 tasks across four industrial scenes, and VABAR, which evaluates robots across six industrial tasks with standardized variations. Nvidia says GR00T 2 "ranks among top three models across vision language action and world action model categories."
"To me, agentic robotics is the most exciting technological development since the internet in 1995," said Ken Goldberg, a UC Berkeley robotics professor leading the VABAR work. Collaborators listed in the announcement include Ambi Robotics, Bosch Research, Flexion, FieldAI and Siemens.
The release lands in a thick stretch of robot-learning news: our robotics tracker has logged 191 stories in the last 90 days, including Nature Will's 50B model for robot hands the day before. The Nvidia post gives no public release date or license for the GR00T 2 weights or the 100,000-hour dataset.
What others are reporting
-
TechTimes Read →
Flags the unnamed test suite and conflicting DexBench task counts (15 vs 18 across sources); cites RoboDojo at 12.8% vs Nvidia's self-reported 95%.
The 95% figure is NVIDIA's average across 50 tasks after five demonstrations per task, and the announcement does not name the tasks or rollout counts.
-
Unite.AI Read →
Critical pipeline analysis: the 95% five-shot average is high partly by construction, covering tasks pre-selected for 80%+ baseline; median 38% agent-wait in real trash-sorting tasks.
40% zero-shot success on 50 tasks absent from the target robot's training data
-
Humanoids Daily Read →
Frames GR00T 2 as Nvidia's strategic response to OpenAI GPT-6 Astra; details cross-embodiment transfer and the CC BY 4.0 open data release.
Astra didn't shatter our dreams, but it shattered our walled gardens. Robotics is no longer an elitist sport anymore.
-
The Neuron Read →
Surfaces December code delay and absent weights; contextualizes within the robotics scaling debate over demonstrations vs agentic code generation.
Robots could learn much of what work looks like by watching people, reducing how much teaching has to happen on expensive physical machines.
Originally reported by blogs.nvidia.com
Read the original article →Original headline: Nvidia Unveils GR00T 2 Robotics Foundation Model at CoRL, Hits 71% One-Shot Success on 50 Tasks