Jensen Huang walked onstage at NVIDIA GTC Taipei on June 1, 2026, and announced what academic robotics labs have waited years for: a single purchasable system that bundles a full-size humanoid robot body with NVIDIA's complete simulation-to-deployment software stack — no vendor assembly required. The NVIDIA Isaac GR00T Reference Humanoid Robot is the first open humanoid reference design built on Jetson Thor silicon and the Isaac GR00T open platform, and it is already committed to by five leading institutions: Stanford, ETH Zurich, the Allen Institute for AI, UC San Diego's Advanced Robotics and Controls Laboratory, and NVIDIA Research itself.
For a university lab that has spent years stitching together hardware from one vendor, simulation software from a second, and training infrastructure from a third, that bundling is the actual product — not the robot.
Open Platform Puts Frontier Humanoid Research Within Reach
Until this announcement, frontier humanoid robotics research required either bespoke hardware that only well-funded corporate labs could afford or negotiated access to proprietary systems with locked-down software. NVIDIA's reference design changes that calculus. According to reporting from GTC Taipei, Rev Lebaredian, NVIDIA's VP of Physical AI Simulation, described the goal as taking humanoid research out of the hands of only the world's largest tech companies and AI unicorns and putting it in reach of every lab.
The hardware centers on Unitree's H2 Plus chassis — standing nearly six feet tall, weighing 150 pounds, with 31 degrees of freedom across the body. Bolted to that are dual Sharpa Wave tactile five-finger hands, which add 22 degrees of freedom each, bringing the total to 75 degrees of freedom across the full system. Each fingertip carries tactile sensors, enabling the manipulation fidelity that tasks like tool use or component assembly require.
The compute brain is an NVIDIA Jetson AGX Thor T5000 module: a Blackwell-architecture GPU delivering 2,070 FP4 teraflops alongside a 14-core Arm CPU and 128 GB of unified memory. That is enough headroom to run language-grounded manipulation commands locally — no cloud roundtrip — at the millisecond latencies that real-time robot control demands. A 15 Ah battery provides roughly three hours of operation.
Sensing includes a head-mounted stereo camera with a 140-degree horizontal field of view, wrist cameras for close-range manipulation, and an inertia measurement unit for motion tracking. Connectivity spans Wi-Fi 6, Bluetooth 5.2, Ethernet, and USB. An on-remote emergency stop lets researchers disengage the robot instantly — essential when prototyping novel behaviors on a six-foot machine.
One Software Stack, Not Seven
The more significant part of the announcement may be the software. The Isaac GR00T development platform integrates tools that research teams typically assemble and maintain independently: Isaac Teleop for capturing high-quality demonstration data from human operators; the Isaac GR00T open foundation models for humanoid reasoning and multitask behavior; Isaac Sim and Isaac Lab for simulating and evaluating robot policies before physical deployment; accelerated Isaac ROS middleware for deploying trained policies onto the robot hardware; and the Jetson Thor runtime for on-device inference and low-latency control.
Critically, researchers retain control of their own robot data, training data, telemetry, and logs. The platform is also modular: teams can adopt the full stack or plug selected components into existing pipelines, lowering the barrier for labs already invested in specific simulation or training infrastructure.
The same Isaac GR00T development workflow will extend to the Unitree G1 — a smaller humanoid already widely deployed in academic labs — through a reference workflow expected soon on GitHub and Hugging Face.
What Stanford, ETH Zurich, and Peers Are Saying
Steve Cousins, executive director of the Stanford Robotics Center, stated that open platforms enabling researchers to share code and test ideas on real machines are the foundation for fast-moving research communities — and that the reference design gives Stanford students and collaborators a shared hardware baseline for comparing and sharing robot behaviors on physical hardware.
Professor Marco Hutter at ETH Zurich's Robotic Systems Lab cited the platform's value for collecting data, testing algorithms, and validating behaviors on a state-of-the-art humanoid body. Hutter's lab is among Europe's strongest in reliable locomotion and manipulation in real-world environments.
Dieter Fox, senior research director at Ai2 and professor at the University of Washington, emphasized that open technology is central to the Allen Institute's mission of broadly competent robotics through open science. Professor Michael Yip at UC San Diego's Advanced Robotics and Controls Laboratory pointed to the platform's integrated approach — connecting hardware, data capture, policy learning, and physical evaluation — as key to accelerating loco-manipulation research in dynamic environments.
NVIDIA Research will use the H2 Plus as its own internal reference system for advancing Isaac GR00T open models, which means academic labs and NVIDIA's research division will share the same platform — an alignment designed to accelerate the pace of open-model improvement.
The announcement landed during ICRA 2026, the flagship annual robotics conference running June 1–5 in Vienna, providing a natural context for academic discussion of what the platform means in practice.
How Does the NVIDIA Isaac GR00T Robot Handle Data and Privacy?
This is the question researchers at federally funded US labs need to answer before procurement. The hardware chassis — Unitree's H2 Plus — is manufactured by Unitree Robotics, a company headquartered in Hangzhou, China. Under Article 7 of China's National Intelligence Law, enacted in June 2017, every Chinese company and citizen is required to support, assist, and cooperate with state intelligence work in accordance with the law. This obligation is not a risk to weigh against the platform's capabilities — it is the operative law of the company's home jurisdiction and applies regardless of where data is stored, where subsidiaries are incorporated, or what any privacy policy states.
The Unitree hardware track record on security is additionally relevant. A critical vulnerability — CVE-2025-2894 — formally catalogued Unitree's pre-installed CloudSail remote access tunnel, which silently connected robots to Unitree servers in China by default on prior models. A separate security assessment published in September 2025 by Alias Robotics documented the UniPwn exploit, which provided full root access to Unitree G1 robots via a Bluetooth provisioning flaw, and found systematic telemetry transmission to servers in China every 300 seconds — including audio, video, and sensor data — without operator notification.
A bipartisan group of 24 House lawmakers sent a letter in May 2025 calling for Unitree's designation as a Chinese military company and its addition to the Commerce Department's Entity List and the Federal Communications Commission's Covered List. The House Homeland Security Subcommittee on Cybersecurity and Infrastructure Protection held a dedicated hearing on Unitree's national security risks on March 17, 2026. Senators Rick Scott and Tom Cotton introduced the Blocking CCP Spy Tech Act of 2026 in May, which names Unitree among six Chinese technology firms and would require a national security investigation that could result in the company being added to the FCC's Covered List — a designation that would restrict its use in US telecommunications infrastructure and could affect researchers receiving federal funding.
NVIDIA addressed these concerns directly. The company confirmed that all software updates destined for the robot's subsystems will flow through NVIDIA's Blackwell chip, where code can be checked for authenticity before execution. NVIDIA is applying secure boot and confidential computing technologies — the same protections used for data center servers — with the explicit goal of preventing the robot from running malicious code and blocking sensitive data from leaving the platform without permission.
Unitree has not confirmed whether the H2 Plus carries the same CloudSail feature as prior models; the company posted a LinkedIn statement in September 2025 acknowledging earlier vulnerabilities and promising patches, but has not published a security audit of the H2 Plus, which is a new product. NVIDIA executives have also confirmed the company plans to extend the Isaac GR00T reference design to humanoid hardware partners in the United States, Europe, and South Korea — though no names have been disclosed. Researchers at US federally funded labs should monitor the status of the Blocking CCP Spy Tech Act before committing procurement budgets.
Unitree IPO Context
The announcement coincided with Unitree clearing its listing-committee review for an IPO on Shanghai's STAR Market on June 1. According to Unitree's IPO disclosures, more than 40% of the company's revenue already comes from customers outside China. Being named by NVIDIA as the hardware partner for a flagship academic initiative is a meaningful credential for a company simultaneously pursuing a domestic capital raise. Pricing for the H2 Plus has not been disclosed.
What Researchers Can Access Now
Researchers who want to begin building immediately do not need to wait for October. NVIDIA is making the Isaac GR00T reference workflow for the Unitree G1 — a smaller humanoid already in widespread use at academic labs — available soon on GitHub and Hugging Face. That release will let existing G1 users immediately adopt the Isaac GR00T development pipeline, providing continuity for institutions that have already deployed G1 hardware, pending whatever additional security guidance emerges for that platform.
The H2 Plus itself is expected to ship from Unitree in October 2026. For university labs with no prior US government funding concerns and a mandate to work at the frontier of humanoid physical AI, NVIDIA's reference design represents the most direct path from lab procurement to deployable policy research that has existed in this field. Whether it proves equally viable for federally funded programs will depend on what the US government determines about Unitree's classification in the months ahead — a question that the pending Blocking CCP Spy Tech Act and ongoing agency reviews may resolve before the October ship date.
Frequently Asked Questions
What is the NVIDIA Isaac GR00T Reference Humanoid Robot?
It is the first open humanoid robot reference design built on NVIDIA's Jetson Thor compute platform and Isaac GR00T open software stack, announced at NVIDIA GTC Taipei on June 1, 2026. The system pairs a Unitree H2 Plus chassis with Sharpa Wave five-finger tactile hands and NVIDIA's complete simulation-to-deployment software pipeline, giving research labs a single integrated starting point instead of assembling hardware and tools from multiple vendors.
How does the NVIDIA Isaac GR00T robot work for humanoid robotics research?
Researchers use the platform's Isaac Teleop tool to capture demonstration data from human operators, then simulate and refine robot behaviors in Isaac Sim and Isaac Lab before deploying trained policies to the physical robot via Isaac ROS middleware. The Jetson AGX Thor T5000 compute module runs inference locally at real-time control speeds, eliminating the need for cloud processing during operation.
When will the NVIDIA Isaac GR00T humanoid robot be available to buy?
The H2 Plus-based reference design is expected to ship from Unitree in October 2026. A reference workflow for the smaller, more widely deployed Unitree G1 is expected to become available on GitHub and Hugging Face before that date. Pricing for the H2 Plus configuration has not yet been disclosed.
Is Unitree Robotics safe to use for federally funded research?
Unitree is headquartered in China and subject to Article 7 of China's National Intelligence Law, which requires Chinese companies to cooperate with state intelligence agencies on demand. The Blocking CCP Spy Tech Act of 2026 — introduced in May and pending in the US Senate — names Unitree specifically and would require a national security investigation that could result in restrictions on its use. NVIDIA has implemented secure boot and confidential computing through its Blackwell chip to address prior security vulnerabilities in Unitree hardware, but no independent security audit of the H2 Plus has been published. Researchers at federally funded US institutions should review the bill's status before committing procurement funding.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
