NVIDIA has revealed the Isaac GR00T platform, an open system for humanoid robots built on the company's Jetson Thor hardware. The platform is designed to push forward robotics research and make what NVIDIA calls 'physical AI' more accessible to developers and academics.
An open architecture for humanoid robots
Isaac GR00T is not a finished robot but a full hardware and software stack that researchers can use to build, train, and deploy humanoid machines. By basing it on Jetson Thor, NVIDIA gives the platform a compact but powerful compute package capable of running real-time AI models for perception, locomotion, and manipulation. The company describes the platform as open, meaning it will provide reference designs, APIs, and simulation tools rather than a locked-in ecosystem.
Democratizing physical AI
The underlying goal of Isaac GR00T is to democratize physical AI — the branch of AI that lets machines interact with the physical world. Right now, building a humanoid robot from scratch is expensive and requires deep expertise in hardware, software, and AI. NVIDIA wants to lower that barrier. By offering a standardized platform, the company hopes to let more labs and startups focus on the high-level intelligence rather than reinventing the motor controllers and sensor fusion stack.
What Jetson Thor brings to the table
Jetson Thor is NVIDIA's latest embedded AI computer, designed for robotics and edge applications. It packs a GPU with tensor cores, a CPU cluster, and dedicated accelerators for computer vision and path planning. On Isaac GR00T, Jetson Thor handles everything from running large language models for natural commands to processing depth cameras for navigation. The chip's power efficiency is key for battery-operated humanoids that need to work untethered.
NVIDIA has not yet announced specific partners or customers for Isaac GR00T, but the platform is aimed at research institutions and commercial developers working on general-purpose humanoid robots. The company said the platform builds on years of work in simulation tools like Isaac Sim and the Omniverse, which let developers test robot behaviors in virtual environments before running them on real hardware.
By opening up the platform, NVIDIA hopes to accelerate the day when humanoid robots move from lab curiosities to practical tools in warehouses, factories, and homes.
