NVIDIA, Unitree and Sharpa Turn the Humanoid Platform into a Reference Design

On 31 May 2026, ahead of COMPUTEX and GTC Taipei activity, NVIDIA announced the Isaac GR00T Reference Humanoid Robot, an open reference design intended to help researchers and developers build generalist humanoid systems. The design combines Unitree’s humanoid body, Sharpa’s dexterous robotic hands and NVIDIA’s Jetson Thor compute platform with the Isaac GR00T software stack. NVIDIA said the reference humanoid robot would be available from Unitree in late 2026, while a GR00T reference workflow for the Unitree G1 was expected to become available on GitHub and Hugging Face for developers. Reuters later reported that NVIDIA also planned to work with humanoid robot makers in the United States, Europe and South Korea, not only with Unitree, and that the research robots were intended for academic users including Stanford University and UC San Diego.

This update is important because it shifts the platform discussion from isolated robot demos to shared development infrastructure. NVIDIA is not presenting itself as a complete humanoid manufacturer; instead, it is trying to define a compute, software and workflow layer that other robot makers can build around. The Unitree-Sharpa-NVIDIA structure also shows how the humanoid stack is becoming modular: body, hands, onboard compute, simulation, model training and security are increasingly supplied by different actors. The commercial impact remains early because availability is planned for late 2026, but the strategic signal is strong: humanoid development may accelerate through reference architectures rather than every company building a fully isolated stack.

This item helps separate deployment infrastructure from promotional demonstrations. A reference robot does not prove mass commercial adoption, but it can reduce friction for universities, labs and developers that need standardized hardware for embodied AI research. If widely adopted, it could also make benchmarking, model transfer and component comparison more practical.

Sources & Further Reading: