NVIDIA GR00T 2 was introduced on Monday, November 9, 2026, during the Conference on Robot Learning (CoRL) in Austin, Texas. The open action model for robots learns manipulation skills from videos of people and marks the next generation of Project GR00T, aimed at humanoid and general-purpose robots.
According to NVIDIA, NVIDIA GR00T 2 is a world action model — a model that observes its environment, anticipates the effects of its own actions, and executes complex multi-step tasks — built on NVIDIA Cosmos. A single pretrained model covers parallel-gripper arms, dexterous hands, and a humanoid, and it transfers skills learned from human demonstrations between different robot bodies without needing demonstrations from each robot.
What NVIDIA's tests show
In the controlled tests disclosed by the company, the model completed 40 of the 50 evaluated tasks at least once with no task-specific demonstration. With just one demonstration per task, the average success rate was 71%; with five demonstrations, it rose to 95%. NVIDIA also says GR00T 2 ranks among the top three models across all vision-language-action and world action model categories in the RoboCasa365, RoboLab, RoboDojo, and MolmoSpaces benchmarks.
These numbers should be read with caution: they come from the manufacturer's own controlled tests. NVIDIA's technical paper notes that the tasks in the five-demonstration test were selected from those the model can learn with five episodes at a success rate of at least 80% — in other words, the 95% average is high partly by construction. None of this guarantees the same performance outside the lab.
Training on human video and open data
Pretraining combined 402,730 hours of first-person human video, 10,184 hours of handheld-gripper demonstrations across 2,897 tasks, and 23,500 hours of robot teleoperation. NVIDIA is releasing 100,000 hours of that first-person human material — the Human-1M-100K dataset, with video, language annotations, and head and hand motion data — under a CC BY 4.0 license.
The company presents GR00T 2 as an open model. According to the project's research page, the code is expected to be released in December — the openness has been announced, but access to the code is still to come.
Who is already testing it
Foxconn is applying NVIDIA GR00T 2 to factory automation for server manufacturing, and Siemens is evaluating the model for industrial use cases. Nexuni is testing it for robotic laundry handling, and MicroAGI is using GR00T 2, post-trained with reinforcement learning, to swap Ethernet cables with a YAM robotic arm. None of the companies has announced a production deployment: the disclosed work is testing and evaluation.
The ecosystem around the model
Alongside NVIDIA GR00T 2, NVIDIA introduced Isaac Lab 3.0, an open-source framework for training robots in simulation, available since October 20, 2026, and the developer preview of Isaac Lab-Arena 0.4 for large-scale evaluation. Two new open benchmarks also debuted: DexBench, with 15 tasks across retail, logistics, manufacturing, and hospitality, and VABAR, from UC Berkeley with NVIDIA, with six industrial tasks. The open EgoHand model, which reconstructs 3D hand motion, has already passed 40,000 downloads on Hugging Face in two weeks.
Context
Project GR00T was unveiled by CEO Jensen Huang at GTC 2024. In March 2025, NVIDIA released GR00T N1, an open 2-billion-parameter model, followed by versions N1.5 and N1.6. NVIDIA GR00T 2 is the next generation of that family.
At CoRL, Project GR00T co-lead Jim Fan is delivering the keynote “From Dreaming to Doing: World Action Models and Agentic Robotics” on Monday. Before the event, Fan said the demonstration video shows real humanoid robots operating autonomously, at normal speed and without teleoperation — a statement by him, reported by Crypto Briefing.
Read also: MorphMimic: uma única IA aprende a controlar três robôs humanoides diferentes.
Sources: NVIDIA Blog — Practice Makes Precise: At CoRL, NVIDIA and Partners Reveal How Robots Rehearse Delicate Work in Simulation; Unite.AI — NVIDIA Introduces GR00T 2 for Robots Learning From Human Video; Crypto Briefing — Nvidia unveils GR00T 2 and Isaac Lab 3.0 at CoRL conference; TipRanks — Nvidia Reveals New Robot AI with 95% Success Rate in Tests.
Image credit: Ambi Robotics, via the official NVIDIA Blog. The image shows an industrial automation system featured in the same CoRL 2026 announcement.
Transparency: this article was produced with the assistance of artificial intelligence, cross-checking public posts on X and sources from the internet. We recommend consulting the original sources for more details.
Muse
