The chipmaker brought a new robot foundation model, an upgraded simulation framework, and an open benchmark to the Conference on Robot Learning in Austin
Nvidia used the 2026 Conference on Robot Learning to show off three new robotics releases: GR00T 2, Isaac Lab 3.0, and a new open benchmark for testing robot behavior.
CoRL 2026 runs from November 9 to 12 in Austin, Texas. Jim Fan, co-lead of Project GR00T, is scheduled to give a keynote on November 9 at 1:50 p.m. CST titled “From Dreaming to Doing: World Action Models and Agentic Robotics.”
What Nvidia brought to Austin #
GR00T 2 is the next generation of Nvidia’s foundation models for robots. A foundation model is a large, general-purpose AI system trained once and then adapted to many tasks, the robotics cousin of the models behind chatbots.
Earlier versions in the line, GR00T N1.6 and N1.7, used a 3B-parameter open vision-language-action model running on a Cosmos backend. A vision-language-action model, or VLA, takes in what a robot sees, understands an instruction written in plain language, and outputs physical movements.
Isaac Lab 3.0 is the training ground. It became generally available in late October 2026, landing roughly between October 20 and 27, a couple of weeks ahead of the conference.
The framework now uses a modular architecture. Developers can plug in different physics engines, including Newton and PhysX, rather than being locked into one way of simulating the world.
It also supports scalable training across multiple GPUs. Isaac Lab 3.0 also includes deployment tooling such as LEAPP, aimed at making it easier to move a trained policy off the simulator and onto real hardware.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
The framework is built to refine those policies through reinforcement learning, where a robot improves by trial and error with rewards, and imitation learning, where it copies demonstrations.
The third piece is a new open benchmark for evaluating robot policies in simulation. It arrives alongside Isaac Lab-Arena, also presented at CoRL 2026.
Why simulation is the whole game #
The core problem in robotics has always been data. Language models could train on the internet. Robots cannot download years of practice folding laundry.
Synthetic data generation and better training pipelines for vision-language-action models have become a growing focus across the robotics industry, and Isaac Lab is Nvidia’s bet on owning that layer.
The “ChatGPT moment” pitch #
Nvidia has framed this period as a turning point for physical AI, suggesting robotics is approaching its own “ChatGPT moment,” where systems gain much higher levels of autonomy and capability.
CEO Jensen Huang has said that general-purpose humanoid robotics is advancing at “extraordinary speed.”
Partners including Agile Robots and Unitree are positioned to build on these releases, and Nvidia pairs its software with hardware such as Jetson Thor, its computing platform for robots.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.