Most humanoid robot headlines promise a robot that will one day do everything. Agile Robots is trying something narrower, and arguably smarter: build a small, standardized humanoid that researchers can use today, prove its skills in a real factory, and only then scale up. In an announcement dated October 27, the Munich-based company detailed its Agile ONE SN research humanoid and a new pilot with Audi that puts that strategy to a very concrete test.
The pilot, run with Audi, Dassault Systèmes, and Nvidia at Audi’s Ingolstadt plant, will have mobile humanoids pick parts from containers and sequence them for final assembly. That is a task performed around one million times a day, according to the announcement. It is repetitive, high-volume, and unforgiving of errors, which is exactly why it makes a better proving ground than a staged demo.
A 1.4-Meter Robot Built for the Lab, Not the Stage
The Agile ONE SN is deliberately petite. It stands 1.4 meters tall and, according to the launch announcement, weighs 43 kilograms. It has dual WUJI tactile five-finger hands and 71 degrees of freedom in total, runs on Nvidia’s Jetson Thor computer, and ships with the GR00T 1.7 open vision-language-action model.
The “SN” stands for Nvidia, and the point of the joint specification is standardization. Reuters reported that production of the ONE S and ONE SN models will begin by the end of this year, with 1,000 to 2,000 units expected next year, and Stanford University and ETH Zurich among the potential customers. Nvidia’s robotics product director told Reuters that a standardized, pre-configured robot could save researchers two or three months otherwise spent getting separately purchased components working together.
Skills developed on the smaller robot are designed to transfer to the full-size industrial Agile ONE, so lab work is not a dead end. Availability of the ONE SN is planned for early 2027.
Audi’s Million-a-Day Test, Rehearsed in a Virtual Twin First
The Audi project will not start on the factory floor. Agile Robots says the workcell will first be tested in a DELMIA virtual twin accelerated by Nvidia Omniverse libraries, letting engineers rehearse the picking and sequencing task in simulation before any robot touches a real production line.
That sequencing matters. Humanoid pilots that skip the simulation step tend to look impressive for a day and stall on edge cases. Testing in a virtual twin first is slower theater but faster deployment, and it mirrors how Agile’s own Diana 7 and Diana 3 robot arms, also announced this week, share one control interface so code written for a research arm moves directly to industrial payloads. The Diana 7 offers a 7-kilogram payload and 923-millimeter reach, while both arms claim repeatability within 0.05 millimeters.
Nvidia’s role goes beyond supplying chips. The company announced plans to invest in Agile Robots at the Invest in Germany Summit in Berlin, two days before the GTC Berlin launch, deepening a partnership that pairs Nvidia computing and GR00T AI models with Agile hardware.
Why This Matters
The humanoid race is usually framed as a sprint to the biggest, flashiest robot. Agile’s angle is different, and it is the element most coverage misses: the bottleneck is not building a humanoid, it is getting thousands of researchers and factory engineers to develop reliable skills on one. A standardized 1.4-meter platform with tactile hands, a shared software stack, and a transfer path to a full-size industrial model attacks that bottleneck directly, and the Audi pilot gives it a ruthless scoreboard — one million sequencing tasks a day, with no room for demo-day excuses.
If the Ingolstadt workcell clears its virtual-twin rehearsal and holds up at real production pace, Agile will have something its bigger rivals mostly lack: proof from a working car plant, not a keynote stage. If it stalls, it will still have shown the industry a cheaper, faster way to find out. Either way, the next humanoid breakthrough may come from a lab-sized robot, not a life-sized one.
Meanwhile in Robotics: Boston Dynamics keeps pushing dexterity with Atlas and its four-finger hand, while another German contender, Neura, is building its own humanoid gym in Munich, as we covered in Neura’s 4NE1 Gen 4 reveal.
