Agile Robots and Nvidia launch Agile ONE SN humanoid with plan for up to 2,000 units

At Nvidia GTC Berlin on October 21, 2026, Munich based Agile Robots introduced Agile ONE SN, a 1.4 metre research humanoid built around Nvidia Isaac GR00T software and the Jetson Thor onboard computer. The machine is a reference platform, meaning it arrives with a jointly specified set of hands, cameras and computing aimed at university and corporate research labs rather than factory floors at first. It is scheduled to be available in early 2027. No price was disclosed.

The launch places Agile, an industrial robot maker, in Europe's early humanoid race alongside NEURA and RobCo of Germany and Humanoid of Britain. Nvidia used the same Berlin event to expand its Isaac robotics platform with a reference humanoid, an Isaac OS runtime and Isaac Lab 3.0, a combination the company says unifies simulation, training and deployment. Jensen Huang, Nvidia's chief executive, said general purpose robotics is advancing at extraordinary speed and that the ChatGPT moment for robotics is fast approaching.

Agile ONE SN is not a standalone gadget. It is designed to feed skills and data into Agile ONE, the company's full size industrial humanoid, and to give labs a standard starting point for collecting demonstrations and testing manipulation. Reuters described the robots as smaller versions of Agile's full size models that are easier and safer for researchers to handle. The timing also follows Nvidia's October 19, 2026, announcement at the Invest in Germany Summit in Berlin, where Huang outlined a European investment push alongside Chancellor Friedrich Merz.

Key Facts

Nvidia Newsroom reported on October 21, 2026 that the Agile ONE S body stands 1.40 metres (4.59 feet), weighs 40 kilograms (88 pounds), has 31 degrees of freedom and carries a 3 kilogram payload per arm. The dual WUJI Hand 2 five finger hands add 20 degrees of freedom each, 1kHz force control and tactile sensing. Agile Robots reported in its own release on October 21, 2026 that the body weighs 43 kilograms (95 pounds) and that body and hands together provide 71 degrees of freedom, with joint torque up to 140 Nm. Reuters put the robot at about 34 kilograms. The differences show that the published specifications are not yet fully harmonised.

The compute package is central to the reference design. Nvidia says Jetson Thor delivers 2,070 FP4 teraflops and 128GB of unified memory for onboard workloads. Nvidia says the preconfiguration cuts setup time from weeks to days. Agile says the robot ships with preconfigured Isaac GR00T software, including the GR00T 1.7 open VLA model, GR00T whole body control and an end to end reference workflow that spans demonstration capture, data generation, simulation, training, evaluation, deployment and real world testing. The package includes URDF and USD assets for Isaac Sim and Isaac Lab, so labs can move from a demo to simulation without rebuilding the stack.

Reuters reported on October 21, 2026 that Agile will start building the Agile ONE S and Agile ONE SN by the end of 2026 and expects to produce 1,000 to 2,000 units next year. CEO Zhaopeng Chen gave the figures in an interview, and Reuters named Stanford University and ETH Zurich among potential customers. Agile's release says availability is scheduled for the beginning of 2027 and that no price was disclosed. The robot is on show at the Nvidia booth and at Agile Robots booth 3007 at GTC Berlin.

Nvidia Newsroom reported on October 21, 2026 that the company also made Isaac Lab 3.0 generally available, introduced an open source Isaac OS runtime built on Holoscan, released three DGX Station reference workflows for GR00T 1.7 and Cosmos post training, and published a SimReady library of more than 2,000 simulation ready robotics and factory assets plus a Human 1M dataset on Hugging Face. European partners named include ABB Robotics, KUKA, NEURA Robotics, Sereact, Humanoid, Hexagon Robotics and Universal Robots. Separately, Tech.eu reported on October 20, 2026 that Nvidia is investing in four German startups, including Agile Robots, with undisclosed amounts, as part of a commitment of nearly $10 billion across European tech.

Analysis

The strategic logic is straightforward. Nvidia supplies compute, simulation and foundation models, while Agile supplies a physical platform that researchers can touch and program. What this really means is that Nvidia is trying to make its Isaac stack the default development environment for humanoid robotics before the market settles, with Agile ONE SN acting as a ready reference design rather than a mass market product. Spencer Huang, Nvidia's product director for robotics, said standardised preconfiguration could save researchers two or three months spent getting separately purchased components working together, and that a system already 80 per cent of the way there accelerates development.

The production target is modest by industrial standards but meaningful for a research category. A run of 1,000 to 2,000 units would be a notable output for early humanoid hardware. Yet the spec discrepancies are a warning. Reuters described about 34 kilograms, Nvidia listed 40 kilograms and Agile cited 43 kilograms, while the companies agree on 1.4 metres and 3 kilograms per arm. Such differences are common in early hardware, but they complicate procurement for labs that need repeatable safety and payload assumptions.

Europe's competitive position is also at stake. Agile is one of the region's early humanoid contenders, alongside NEURA and RobCo of Germany and Humanoid of Britain, and Nvidia has named several of them as partners. A standardised Nvidia reference platform could raise the floor for everyone, giving smaller labs access to a capable base. It could also concentrate value in Nvidia's software and chips while hardware makers compete on cost, reliability and local support. The bigger picture here is that the humanoid race is becoming an ecosystem contest, not just a robot contest.

Why It Matters

For research labs, the appeal is time. A preconfigured robot with dexterous hands, cameras, lidar, inertial measurement units and onboard AI can remove months of integration work, letting teams focus on data collection and skill learning. Agile says the skills and data developed on the SN model can later transfer to Agile ONE, its full size industrial humanoid, which gives the research purchase a path to factory and warehouse applications. The 1.4 metre size and roughly 34 to 43 kilogram weight make it easier to handle than a full size humanoid, a practical safety consideration in crowded labs.

For the wider market, the GTC Berlin package shows how fast the humanoid toolchain is maturing. Simulation assets, open source runtimes, post training workflows and datasets are as important as motors and hands, because they determine how quickly customers can turn a robot into a useful system. Nvidia's nearly $10 billion European commitment and its startup investments add financial weight to that push. The risk is dependence: if one vendor supplies the chips, the model, the simulation and much of the investment, labs may gain speed but lose bargaining power and control over their own roadmaps.

Next Up

The next milestones are concrete. Agile plans to begin production by the end of 2026 and to become available in early 2027, so the coming months will show whether the 1,000 to 2,000 unit target holds and whether Stanford, ETH Zurich or other labs sign firm orders. Pricing remains undisclosed, a key missing detail for research budgets. Nvidia's planned integration with the Halos safety stack and the adoption of Isaac Lab 3.0 will also be watched by European partners such as ABB Robotics, KUKA, NEURA Robotics and Universal Robots.

GTC Berlin also provides a test of messaging. Agile and Nvidia are presenting a reference humanoid as a shared platform for data and skills, not as a finished industrial product. If researchers adopt it, the Agile ONE SN could become a common baseline for humanoid experiments in 2027. If they do not, the gap between demonstration and deployment will remain the story.