NVIDIA launches GR00T 2 and Isaac Lab 3.0 at CoRL
NVIDIA unveiled its GR00T 2 foundation model and Isaac Lab 3.0 simulation platform at the Conference on Robot Learning, aiming to accelerate robot training for industrial manufacturing applications.
NVIDIA Corp. introduced its next-generation GR00T 2 foundation model and updated Isaac Lab 3.0 simulation platform at the Conference on Robot Learning in Austin, Texas. The company made these announcements at the Conference on Robot Learning, or CoRL, in Austin, Texas. The event runs from November 9 to 12, with NVIDIA's Jim Fan, co-lead of Project GR00T, scheduled to give a keynote on November 9. The new tools are designed to speed up and improve robot training for manufacturing automation.
GR00T 2 demonstrates human demonstration transfer and task efficiency
GR00T 2 enables dexterous robot manipulation by learning from human demonstrations. The model was trained on robot data and 100,000 hours of open-source first-person human video. In evaluations, it completed 40 out of 50 benchmark tasks at least once without any task-specific demonstrations. With just one human demonstration per task, it achieved a 71% average success rate across 50 tasks. That rate increased to 95% when provided with five demonstrations per task.
GR00T 2 ranks among the top three models in key robotics benchmarks, including RoboCasa365, RoboLab, RoboDojo, and MolmoSpaces. NVIDIA has not disclosed pricing or general availability for the model.
Industry trials begin with Nexuni, Foxconn, and Siemens
Early industrial testing of GR00T 2 is already underway. Nexuni is testing the model for robotic laundry handling. Foxconn is evaluating it for automating server manufacturing processes. Siemens is assessing GR00T 2 across a range of industrial use cases. These trials represent the first wave of real-world manufacturing applications for the new foundation model.
Isaac Lab 3.0 advances training with Newton physics and new capabilities
Isaac Lab 3.0, which became generally available in late October 2026, introduces major upgrades for simulation-based training. A key addition is integration with the Newton physics engine, allowing researchers to model deformable objects like cables and cloth, as well as complex articulated systems such as robotic hands. The platform's modular architecture lets developers switch between the Newton and NVIDIA PhysX solvers.
The performance gains are significant. In a vision-based dexterous manipulation benchmark, Isaac Lab 3.0 with Newton achieved up to 2.5 times faster training throughput compared to the previous Isaac Lab 2.3 running on PhysX. The platform supports scalable training across multiple GPUs and includes deployment tooling like LEAPP for moving trained policies to real hardware.
Ecosystem partners leverage new tools for diverse applications
A broad ecosystem of companies is building on NVIDIA's new robotics stack. Flexion showed how Isaac Lab and Newton train its Reflect model, which combines reasoning and whole-body manipulation. MicroAGI uses a fine-tuned version of GR00T 2, post-trained via reinforcement learning, to perform Ethernet cable swapping with a YAM robotic arm.
Dyna Robotics will demonstrate parallel simulation in Isaac Lab for training whole-body controllers for its Taku robot, intended for warehouses, hotels, and food services. Walden Robotics uses Isaac Sim’s photorealistic rendering to train its general-purpose robot for manufacturing and logistics. Dexmate explores coding agents to rehearse manipulation strategies in Isaac Sim before executing them on its Vega robot.
Skild AI announced STEEL, a rigid-body physics solver built on Newton to accelerate robot training within Isaac Lab. New open benchmarks have also been released. DexBench, developed with NVIDIA, features 15 tasks across retail, logistics, manufacturing, and hospitality scenes. VABAR, from UC Berkeley and NVIDIA, evaluates robots across six standardized industrial tasks. The University of Texas at Austin’s School of Computing and Texas Robotics uses Isaac Sim and Isaac Lab to train robotic skills in simulation for students before physical hardware is available. NVIDIA pairs its software with hardware such as the Jetson Thor computing platform for robots, with partners including Agile Robots and Unitree positioned to build on these releases.
