Nvidia selects Unitree for humanoid research platform
NVIDIA has unveiled an open humanoid robot reference design for academic research, built on its Isaac GR00T platform and using a Unitree humanoid for the body, according to NVIDIA and CNBC. The NVIDIA Isaac GR00T Reference Humanoid Robot pairs a Unitree H2 Plus chassis and Sharpa five-finger dexterous hands with NVIDIA's Jetson Thor onboard compute, which uses the Blackwell GPU, plus Isaac GR00T models and simulation tools. NVIDIA says the roughly six-foot, 150-pound robot has 31 degrees of freedom and will be sold by Unitree in late 2026, with early adopters including Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego. CNBC notes the deal lands as Unitree pursues an IPO. Packaging on-device compute with a ready body and shared software stack lowers integration time and tends to improve reproducibility across humanoid-robotics labs.
What NVIDIA announced
NVIDIA introduced the NVIDIA Isaac GR00T Reference Humanoid Robot, an open reference design aimed at academic robotics research. NVIDIA describes the body as a Unitree H2 Plus humanoid fitted with Sharpa five-finger dexterous hands, and the brain as NVIDIA Jetson Thor onboard compute, which includes the Blackwell GPU, running Isaac GR00T open software, models, and simulation workflows.
Specs and availability
NVIDIA says the robot stands nearly six feet tall, weighs about 150 pounds, and has 31 degrees of freedom. The reference robot will be available from Unitree in late 2026, and NVIDIA says an Isaac GR00T reference workflow for the Unitree G1 is expected on GitHub and Hugging Face. Named research adopters include Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego's Advanced Robotics and Controls Laboratory. CNBC reports the partnership comes as Unitree pursues an IPO.
Analysis
- •Bundling Blackwell-class on-device compute with a ready humanoid body and a shared software stack lowers the integration burden that often slows embodied-AI experiments.
- •Common reference hardware historically improves reproducibility and benchmarking, which can raise the pace of published humanoid-robotics results.
- •The open-design framing matters for research access, but real adoption will track tooling maturity, support, and unit economics once units ship.
Key Points
- 1NVIDIA's Isaac GR00T Reference Humanoid Robot pairs a Unitree H2 Plus body and Sharpa five-finger hands with Jetson Thor (Blackwell) compute and Isaac GR00T models, sold by Unitree in late 2026 (NVIDIA, CNBC).
- 2Named early adopters include Ai2, ETH Zurich, Stanford Robotics Center, and UC San Diego, signaling a focus on academic embodied-AI research.
- 3Industry pattern: vendor reference robots reduce integration overhead and tend to improve reproducibility and benchmarking across labs.
Scoring Rationale
NVIDIA establishing a standardized, open humanoid reference platform with named academic adopters is a notable embodied-AI development that lowers research barriers and could shape humanoid-robotics benchmarking. It is significant for the robotics and physical-AI community but is a reference design shipping later in 2026 rather than a deployed product, keeping it in the notable band.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
