Munich-based Agile Robots will build the Agile ONE SN, a compact humanoid that Nvidia presented as a reference platform for robotics research at its GTC Berlin conference on October 21, 2026. The robot runs Nvidia’s Isaac GR00T models on a Jetson Thor computer and is due to be available in early 2027. Agile chief executive Zhaopeng Chen told Reuters the company expects to make 1,000 to 2,000 of its small humanoids next year.

Why it matters: Nvidia does not build robots, but it is now specifying one. A reference design brings labs onto its compute, models and simulation tools before they have picked anything else.

Key Facts

  • Announced October 21, 2026 at GTC Berlin; availability early 2027 (Nvidia)
  • Nvidia spec: 1.40 m, 40 kg, 31 degrees of freedom, 3 kg payload per arm; two WUJI Hand 2 hands with 20 DoF each, 1 kHz force control and tactile sensing
  • Compute: Jetson Thor, 2,070 FP4 teraflops, 128 GB unified memory
  • Volume: 1,000–2,000 small humanoids in 2027; ONE S and ONE SN production to start by end-2026 (Chen, to Reuters)

What is in the reference humanoid?

Nvidia lists head and wrist cameras, lidar and inertial measurement units alongside the hands and Jetson Thor, with integration of its Halos safety stack planned. Reuters reported that the “N” stands for Nvidia, and that the SN is a jointly specified configuration of Agile’s ONE S. Reuters put the robots at roughly 34 kilograms; Nvidia’s spec sheet says 40. Neither company has published a price.

The buyers are research labs. Nvidia named Stanford University and ETH Zurich among interested institutions. Spencer Huang, Nvidia’s product director for robotics, told Reuters a pre-configured robot could save researchers two or three months of integration work: “If we can bring researchers something that’s already 80% of the way there, it accelerates their development.”

What else did Nvidia ship in Berlin?

The software around the robot. Isaac Lab 3.0, Nvidia’s open-source robot-learning framework, became generally available with support for the Newton and PhysX physics engines. Nvidia also introduced Isaac OS, an open-source runtime built on Holoscan that links AI, sensors, controls and hardware; a SimReady library of more than 2,000 simulation-ready robot and factory assets on Hugging Face; and a Human 1M dataset of synchronised three-view video with 3D hand tracking, whose full version goes to Cosmos Coalition partners.

The partner list is broad. ABB Robotics, KUKA, NEURA Robotics, Universal Robots and the British start-up Humanoid were named among European adopters, and Agility and FORT Robotics among Halos safety partners. Jensen Huang also showed Microduck, a 25-centimetre open-source biped from Hugging Face and Pollen Robotics.

Why does a lab robot matter commercially?

Because labs set defaults. A researcher who trains a policy on GR00T in Isaac Lab and deploys it on Jetson Thor carries that stack into later projects. Nvidia already sits under several humanoid makers, including LG’s GR00T-based humanoid; the ONE SN gives it a robot it can point labs to directly.

Agile’s target is modest by Chinese volumes, where the largest makers count output in the tens of thousands. It is large for a European humanoid. Reuters listed NEURA and RobCo in Germany and Humanoid in Britain as Europe’s other early contenders. Nvidia’s release did not address US availability; our FCC report explains why foreign-produced robots now need specific approval there.

Frequently Asked

What is the Agile ONE SN?

A compact humanoid built by Agile Robots that Nvidia presented as a reference platform for robotics research at GTC Berlin on October 21, 2026. Nvidia lists it at 1.40 m with 31 degrees of freedom, running Isaac GR00T on a Jetson Thor computer.

When can labs get the Agile ONE SN?

Nvidia says early 2027. Agile Robots says production of its ONE S and ONE SN models starts by the end of 2026 and expects to make 1,000 to 2,000 small humanoids next year. No price has been published.

Who is the Agile ONE SN for?

Universities and corporate research labs. Nvidia named Stanford University and ETH Zurich among the institutions interested in it.