What happened
NVIDIA introduced GR00T 2, an open world action model trained on 100,000 hours of egocentric data. It completed 40 of 50 evaluated tasks at least once without task-specific demonstrations, and scored 71% average success across 50 tasks with one demonstration, rising to 95% with five.
Why it matters
A model that can pick up dexterous manipulation skills from a single human demonstration and transfer them across different robot bodies without robot-specific demonstrations could sharply reduce the data needed to train new robots, according to NVIDIA.
WHO IT HITSRobotics researchers and developers building robot-training datasets stand to benefit from GR00T 2's out-of-the-box task performance and skill transfer, which NVIDIA says reduces the need for task-specific demonstrations.
Summaries like this, every morning.
NVIDIA's announcements at CoRL in Austin, Texas, extend its push into robot learning beyond simulation tooling into the models themselves. GR00T 2 is positioned as an open frontier world action model, and NVIDIA says its training mix includes 100,000 hours of open-source egocentric data. The company reports that the model completed 40 of 50 evaluated tasks at least once without task-specific demonstrations, and that success rates across 50 tasks rose from 71% with one demonstration per task to 95% with five. Those figures matter because demonstration data is a bottleneck for robot training, so a model that learns from few examples could lower the cost of teaching new skills.
Alongside GR00T 2, NVIDIA introduced Isaac Lab 3.0, an open-source robot learning framework that adds tighter integration with the Newton physics engine. NVIDIA says the combination enabled up to 2.5x the training throughput of Isaac Lab 2.3 with PhysX in a vision-based dexterous manipulation benchmark. The company also released a developer preview of Isaac Lab-Arena 0.4, a framework for scaled policy evaluations, and highlighted two new open benchmarks, DexBench and VABAR, aimed at measuring robot dexterity and adaptability across industrial tasks.
NVIDIA's partner disclosures suggest early commercial interest. Foxconn is applying GR00T 2 to factory automation for server manufacturing, Siemens is evaluating it across industrial use cases, and Nexuni is testing it for robotic laundry. Separately, NVIDIA announced an investment in South Korean startup Point2 at the MSIT-NVIDIA AI Startup Partnership Forum, with additional investments planned in Motif Technologies, RLWRLD, and Upstage. The launches come as NVIDIA's board investigates co-founder and CEO Jensen Huang over a relationship with a business executive that U.S. intelligence determined to be a Chinese spy; the board said it has "no reason to believe that either national security or NVIDIA secrets were compromised."
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.
Pieter Abbeel's group at University of California Berkeley published SuperMimic, which replaces the physics si…
Aeon, Aeon Global SCM and Mujin began full operations at Aeon Kawaguchi-Higashi XD on October 29, the first Ae…
RobotBank said on November 9, 2026 it will build physical-AI robots under its own brand, beginning with a quad…
The city's demonstration, running since October 26 in Yahatahigashi Ward's Tobata district, checks whether GAM…
Nature Will announced Natura-1, a 50-billion-parameter mixture-of-experts model trained on over 1 million hour…
Xinhua News Agency unveiled two bionic humanoid robot news anchors, Xinyu and Xinyan, developed with UBTECH, o…
