This Is NVIDIA’s Most Advanced Robot Yet
NVIDIA is taking another major step in the race to develop intelligent humanoid robots. The company has revealed the NVIDIA Isaac GR00T Reference Humanoid Robot, a new open platform designed to help researchers and developers build more capable robots without starting from scratch.
One of the biggest challenges in humanoid robotics today is that development is often fragmented. Teams must piece together hardware, software, simulation tools, training systems, and deployment workflows from multiple sources. That process can slow innovation and make advanced research difficult for smaller organizations. NVIDIA wants to change that by offering a complete reference design that combines a physical robot with a powerful artificial intelligence development platform.
At the center of the system is the Unitree H2 humanoid robot. Standing nearly 6 feet tall and weighing about 150 pounds, the robot is designed for human scale testing and interaction. It features 31 degrees of freedom throughout its body, allowing it to move in ways that closely resemble human motion. NVIDIA pairs the robot with Sharpa Wave tactile five finger hands, adding another 44 degrees of freedom and bringing the total to 75. This combination gives the robot the dexterity needed for complex tasks such as grasping, manipulating objects, and interacting with the surrounding environment.
The robot is equipped with an impressive array of sensors. A head mounted stereo camera provides a wide field of view, while wrist mounted cameras help with close range manipulation tasks. An inertia measurement unit tracks movement and orientation in real time, giving the robot a better understanding of its position and surroundings.
Physical strength is another key part of the design. The humanoid platform can generate up to 120 Newton meters of torque in its arms and up to 360 Newton meters in its legs. It can carry a rated payload of 7 kilograms and handle peak loads of up to 15 kilograms. These capabilities allow researchers to explore more demanding tasks involving lifting, reaching, and whole body coordination.
Perhaps the most important component is the onboard computing system. The robot is powered by the NVIDIA Jetson AGX Thor T5000, a next generation processor featuring a Blackwell graphics processing unit. The system delivers 2,070 FP4 teraflops of artificial intelligence performance, supported by a 14 core Arm central processing unit and 128 gigabytes of unified memory. This gives the robot enough computing power to process sensor data, make decisions, and execute actions directly on the machine in real time.
The software side is equally important. NVIDIA’s Isaac GR00T platform provides tools for collecting robot training data, creating simulations, training policies, testing behaviors, and deploying artificial intelligence models. Researchers can use Isaac Teleop to gather demonstration data, Isaac Sim and Isaac Lab to build virtual training environments, and Isaac GR00T foundation models to develop reasoning and multitask capabilities. Once ready, those trained behaviors can be deployed onto physical robots using NVIDIA Isaac Robot Operating System technologies and Jetson Thor computing.
A major advantage of the platform is flexibility. Research teams can use the entire stack or integrate individual components into existing projects. This approach helps developers avoid rebuilding the same infrastructure repeatedly and allows them to focus more on advancing robot intelligence.
NVIDIA believes humanoid robots could eventually transform industries ranging from manufacturing and logistics to healthcare and service operations. By providing an open platform instead of a closed ecosystem, the company hopes to accelerate research and encourage collaboration across universities, laboratories, and robotics companies.
Several leading institutions have already committed to using the platform, including Stanford Robotics Center, ETH Zurich, Ai2, and the Advanced Robotics and Controls Laboratory at the University of California San Diego. NVIDIA’s own research teams will also use the system to further develop future Isaac GR00T models and frameworks.
The company plans to make the NVIDIA Isaac GR00T Reference Humanoid Robot available through Unitree in late 2026. Support for the Unitree G1 humanoid robot is also on the way, extending the platform to one of the most widely used humanoid research robots in the world.
As the competition to create general purpose humanoid robots continues to intensify, NVIDIA’s latest platform could become an important tool for researchers looking to move from experimentation to real world deployment much faster than before.
Submit a Comment