Agile Robots Builds a Reference Humanoid on Nvidia’s GR00T

Agile Robots, the Munich robot maker, launched Agile ONE SN at Nvidia’s GTC Berlin on 21 October 2026, a humanoid its founder and chief executive, Zhaopeng Chen, calls “a full-stack, open reference humanoid built on NVIDIA Isaac GR00T, powered by Jetson Thor.” Nvidia’s release says the robot is “scheduled to be available in early 2027”; the keynote slide that unveiled it read “Available Q1 2027.”

In plain terms, a reference humanoid is a ready-made robot that labs can share as a common starting point, instead of assembling the body, sensors, computer and software themselves. Nvidia’s release says researchers often struggle to integrate separate simulation, training and deployment tools, and that Agile ONE SN unifies them “on a single platform, cutting setup time from weeks to days.” The same release expands Nvidia’s Isaac robotics platform and names ABB Robotics, KUKA, NEURA Robotics and more than a dozen other companies building on it.

Business Pill · PHYSICAL AI

A short Business Engineer explainer on physical AI: what changes when AI gets a body. It teaches the general idea only and says nothing about Nvidia or Agile Robots.

Watch · The Keynote

NVIDIA GTC Berlin 2026 keynote with CEO Jensen Huang, 21 October, on NVIDIA’s YouTube channel. Quotations in this piece come from the releases and the slides shown, not from the spoken keynote.

What the Robot Is

• Body. The Agile ONE S humanoid body stands 1.40 meters tall, with 31 degrees of freedom and a 3-kilogram payload per arm, according to Nvidia’s release. Nvidia puts the weight at 40 kilograms; Agile Robots’ own release says 43 kilograms.
• Hands. Two WUJI Hand 2 five-finger hands, each with 20 degrees of freedom and tactile sensing. Agile Robots says that makes “a total of 71 degrees of freedom across body and hands.”
• Computer. An onboard Nvidia Jetson Thor, which Nvidia says delivers “2,070 FP4 teraflops of compute and 128GB of unified memory.”
• Software. Nvidia’s Isaac GR00T platform, which Agile Robots says includes the GR00T 1.7 open VLA model and GR00T whole-body control. Nvidia says integration with its Halos safety system is planned; Agile Robots says it is exploring that integration.
• Buyers. Nvidia says the robot is aimed at frontier humanoid research, “with interest from leading institutions such as Stanford University and ETH Zurich.” Neither release gives a price.

On stage, Huang described Agile ONE SN as Nvidia’s reference robot, built in partnership with Agile Robots, and said it will come preassembled with the computer and software inside, according to an automatic transcript of the keynote stream. Agile Robots says skills and data developed on Agile ONE SN can later be transferred to Agile ONE, its full-size humanoid for industry.

What Was Posted on X

Nvidia’s newsroom account posted the announcement at 10:30 UTC on 21 October: new Isaac tools for agents to help developers train, test and deploy robots, as Agile Robots unveils Agile ONE SN, a reference humanoid built on Isaac GR00T and powered by Jetson Thor. It is Nvidia’s own summary of the release this piece is built on.

What Nvidia Added to Isaac

• Data. A SimReady Central Library of “more than 2,000 simulation-ready robotics and factory assets” and a Human 1M Dataset of human activities with three-view video and 3D hand tracking, both on Hugging Face. The full Human 1M Dataset is accessible to Cosmos Coalition partners, the release says.
• Training. Isaac Lab 3.0, an open source learning framework, is now generally available, according to Nvidia.
• Runtime. Isaac OS, an open source runtime that Nvidia says connects AI, sensors, controls and robot hardware for real-time performance.
• Workflows. Three open source reference workflows for Nvidia’s DGX Station, including one that post-trains Nvidia Cosmos world models for synthetic data generation “using ChatGPT Astra.”

Jensen Huang said in Nvidia’s release: “General-purpose robotics is advancing at extraordinary speed — the ChatGPT moment for robotics is fast approaching.”

Who Nvidia Names

The release’s section on European robotics lists 17 companies building on Nvidia’s tools, on our count, including ABB Robotics, KUKA Group, NEURA Robotics, Schaeffler, Hexagon Robotics and Hugging Face with Pollen Robotics. Nvidia says Toyota Material Handling Europe and SoftServe use its simulation tools to develop autonomous material-handling solutions, “reducing deployment time by up to 50%.”

Some names point at deployments rather than research. Nvidia says Bellboy’s Billie robot is “already deployed at hotels operated by Marriott, Hilton, IHG and OYO,” and that Schaeffler is contributing its actuator technology “as an exclusive foundational component for Newton,” a physics engine Nvidia’s tools use.

Hugging Face and Pollen Robotics are using Isaac Lab, Isaac Sim and Newton to build Microduck, “a small, open source robot,” according to the release. Small Pollen robots walked on stage during the keynote, and Huang urged viewers to order one from Pollen’s website, according to an automatic transcript of the stream.

Agile Robots shows up more than once. Its own release lists Franka Robotics and idealworks among its subsidiaries, and both appear separately in Nvidia’s release: Franka as a simulation-asset contributor and as a user of Nvidia’s tools, and idealworks as a builder of memory capabilities for factory robots. Agile Robots itself is also named among the companies integrating Halos safety and among those backing the DGX Station workflows.

Connecting the Coverage

Our Germany pact piece reported on 19 October that Nvidia announced “new planned investments” in four German startups, Agile Robots among them, as part of what Nvidia called a nearly $10 billion commitment across Europe; it gave no amount for any of the four. Two days later, Agile Robots is the company whose humanoid Nvidia put on stage as its reference robot.

Our Eka piece reported on 20 October that Eka Robotics closed a $150 million Series A at $1.45 billion, led by a16z and F-Prime, for a force-sensing approach to robot manipulation. Our SoftBank piece reported Bloomberg’s account on 16 October that SoftBank was weighing an acquisition of Allied Universal at an equity value of as much as $18 billion, as Masayoshi Son explores deploying AI-powered robots.

Today’s other GTC Berlin pieces cover the rest of the stack: our Vera Rubin piece on Nvidia’s data-centre chips, and our OpenShell piece on running AI agents under enforced limits. What is new here is the body: a named humanoid, a ship window and a list of companies training robots on Nvidia’s tools.

The Structural Read

As we read it, the release is as much about data as about hardware: Nvidia lists more than 2,000 simulation-ready assets, a Human 1M Dataset and open source workflows for generating synthetic training data, alongside the robot itself.

As we read it, Agile Robots is the European anchor of the announcement: its humanoid is on the keynote slide, its subsidiaries Franka Robotics and idealworks appear in Nvidia’s list, and Nvidia announced a planned investment in it two days earlier.

As we read it, the named deployments are still few: Nvidia says Bellboy’s Billie robot is already deployed at hotels and that Toyota Material Handling Europe and SoftServe are reducing deployment time by up to 50%, while the humanoid itself is described as a research platform.

Three Implications

FOR ROBOTICS RESEARCH LABS Nvidia says Agile ONE SN cuts setup time from weeks to days by putting simulation, training and deployment tools on a single platform, with interest from Stanford University and ETH Zurich.

FOR EUROPEAN INDUSTRIAL ROBOT MAKERS Nvidia names ABB Robotics, KUKA Group and NEURA Robotics among the companies building on its simulation and training tools, and Schaeffler as the supplier of an exclusive actuator component for the Newton physics engine.

FOR ROBOT INVESTORS Nvidia announced a planned investment in Agile Robots on 19 October without giving an amount, and neither 21 October release gives a price or unit volume for the new humanoid.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Four Intelligence Moats.

The framework opens: “Every AI paradigm produces a different kind of moat.” It argues that what changes between paradigms is “where intelligence accumulates and who can capture it.”

As we read it, Nvidia’s release is a bid to sit where robot data accumulates: it supplies the datasets, simulation assets and training workflows that robot makers use, as well as the reference robot. The framework was published in July 2026 and does not discuss this release.

What Is Not Established

Neither release, as we read them on 21 October, gives a price for Agile ONE SN, a number of units, or the terms of any customer agreement; Stanford and ETH Zurich are described as showing “interest,” not as buyers. The two releases differ on the robot’s weight. Neither release mentions the planned Nvidia investment in Agile Robots announced on 19 October. The counts of named companies in our chart are our arithmetic.

The Bottom Line

Nvidia used GTC Berlin to give its Isaac robotics platform a body: Agile ONE SN, a reference humanoid from Agile Robots that runs GR00T on Jetson Thor and is due in early 2027. As we read it, Nvidia is offering robot labs a standard machine to build on, with the software, data and simulation tools around it, much as it pitched Hyperion to carmakers the same day.