Research
Nvidia releases Isaac Lab 3.0 and GR00T 2 at CoRL as robot training shifts to dexterity
Nvidia used this week's Conference on Robot Learning in Austin, Texas, to release Isaac Lab 3.0, the GR00T 2 robot model and a set of new benchmarks, all aimed at teaching robots fine manipulation such as aligning gears and connecting cables. The company set out the releases in a blog post on Monday.
Isaac Lab 3.0
Isaac Lab is Nvidia's open source framework for training robots in simulation. Version 3.0 lets researchers use AI agents to set up training environments with simulation ready assets, switch between the PhysX and Newton physics solvers, choose renderers for vision based training and apply a new reinforcement learning post training library to foundation models such as GR00T.
Tighter integration with Newton, an open source physics engine, means Isaac Lab can now model deformable objects such as cables and cloth, and complex robotic hands with many joints. In a vision based dexterous manipulation benchmark, Nvidia says Isaac Lab 3.0 with Newton reached up to 2.5 times the training throughput of Isaac Lab 2.3 with PhysX.
GR00T 2
GR00T 2 is an open world action model built on Nvidia Cosmos and trained on robot and human data, including 100,000 hours of Nvidia's open human egocentric video. Nvidia reports that it completed 40 of 50 evaluated tasks at least once without task specific demonstrations, and averaged 71% success across 50 tasks with one demonstration per task, rising to 95% with five. The research page describes skills learned from human video transferring between different robot bodies. These are Nvidia's own figures.
Nvidia says Foxconn is applying GR00T 2 to server manufacturing, Siemens is evaluating it for industrial uses and Nexuni is testing it for laundry handling.
Benchmarks and partners
Nvidia also previewed version 0.4 of Isaac Lab Arena, a framework for running policies across thousands of test variations to find failure conditions. Two new open benchmarks built on it, DexBench from RLWRLD and VABAR from UC Berkeley, measure dexterity and adaptability on industrial tasks.
Several humanoid and mobile robot makers are using the tools. Dyna Robotics is training whole body controllers for its semi humanoid Taku robot in Isaac Lab, and Flexion uses Isaac Lab and Newton to train its Reflect model, which combines reasoning with whole body manipulation.
Why it matters
Humanoids that walk well are now common; humanoids that can handle small, delicate parts reliably are not. Faster, more accurate simulation is one of the main levers developers have to close that gap before testing on real hardware.