THE CRUNCH
NVIDIA has introduced GR00T 2, an open frontier world action model built on its Cosmos platform, at the Conference on Robot Learning (CoRL) in Austin, Texas. World action models are AI systems that let robots perceive their surroundings, anticipate how their actions will change the environment and carry out precise multistep tasks. Trained on diverse robot and human data, including 100,000 hours of NVIDIA's open source human egocentric footage, the model is aimed at dexterous manipulation such as aligning gears, connecting cables and assembling small components.
NVIDIA reports that GR00T 2 completed 40 of 50 evaluated tasks at least once without task-specific demonstrations, and achieved a 71% average success rate across 50 tasks with just one demonstration per task, rising to 95% with five. It can also transfer skills learned from human demonstrations between different robot bodies without robot demonstrations, and its API lets AI agents run skills in sequence and recover from errors. The company says the model ranks among the top three across vision language action and world action model categories on the RoboCasa365, RoboLab, RoboDojo and MolmoSpaces benchmarks. Nexuni is testing it for robotic laundry handling, Foxconn is applying it to factory automation for server manufacturing and Siemens is evaluating it across industrial use cases.
Alongside the model, NVIDIA released Isaac Lab 3.0, an update to its open source robot-learning framework, with tighter integration of Newton, an open source physics engine. Newton can simulate deformable objects such as cables and cloth, and complex articulated systems like multi-jointed robotic hands, which matters because accurate physics helps skills learned in simulation transfer to real robots. In a vision-based dexterous manipulation benchmark, NVIDIA says Isaac Lab 3.0 with Newton achieved up to 2.5 times the training throughput of Isaac Lab 2.3 with PhysX, allowing thousands of simulated robots to train in parallel.
The announcements sit within a broader push the company calls agentic robotics, where AI agents generate interpretable code for physical robotics systems. NVIDIA also introduced an open Video to Data workflow that turns human video into simulated robot demonstrations, and an EgoHand model that reconstructs 3D hand motion, which it says has passed 40,000 Hugging Face downloads in two weeks. Partners involved include Bosch Research, Siemens, Skild AI, UC Berkeley and the University of Washington, among others.