Prince Mario-Max Schaumburg-Lippe: A Reference Humanoid Arrives for Every Lab

Building a humanoid robot used to mean building everything. The body, the hands, the computers, the training software, the safety system – a lab could spend years on plumbing before it taught the machine a single useful skill. On Wednesday, October 21, 2026, in Berlin, Agile Robots and NVIDIA launched a shortcut: Agile ONE SN, a compact humanoid that arrives as one integrated reference system, ready for researchers to build on.

The launch came at NVIDIA’s GTC event in Berlin, where the company expanded its Isaac robotics platform. Agile ONE SN is the hardware face of that expansion – a full-stack reference humanoid preconfigured with the NVIDIA Isaac GR00T development platform, so skills built and trained in the lab can move toward real-world deployment on the same machine.

The machine, by the numbers

Agile ONE SN uses the Agile ONE S body. It stands 1.40 meters tall and weighs 40 kilograms, with 31 degrees of freedom and a payload of 3 kilograms per arm. The hands are the headline. Its dual WUJI Hand 2 five-finger hands each provide 20 degrees of freedom, with 1 kHz force control and tactile sensing for delicate manipulation – the difference between a robot that can grip a box and one that can learn to handle objects the way a person does, by feel.

Onboard computing comes from NVIDIA Jetson Thor, delivering 2,070 FP4 teraflops of compute and 128 GB of unified memory for reasoning and real-time control. A multimodal sensing suite combines head and wrist cameras, lidar sensors and inertial measurement units. NVIDIA says the integrated system cuts setup time from weeks to days, because the simulation, training and deployment tools arrive unified rather than assembled piece by piece.

The Isaac GR00T platform supplies the foundation model and a workflow that runs from data capture to deployment, including whole-body control for coordinated movement. Agile Robots also plans to integrate on-robot safety protocols with NVIDIA Halos, the company’s full-stack safety system for robotics.

Why a reference robot matters

Every maturing technology gets its reference design moment. Personal computers had one. Smartphones had one. A reference platform does not end competition – it moves competition up a level, from wiring and drivers to skills and applications. A university team that wants to test a new way of teaching a robot to fold, sort or assemble should not first have to become a hardware company.

Agile Robots founder and CEO Zhaopeng Chen put it in terms of usefulness. “The breakthrough for humanoids is usefulness at scale: one adaptable machine that learns many skills and works across environments built for people,” he said, describing Agile ONE SN as a full-stack, open reference humanoid that gives developers a ready system to create, train and validate the capabilities needed to put humanoids to work.

The timing connects to what factories are already doing. In China this month, humanoid robots at Schaeffler started building parts for other robots on a production line. Machines in factories are no longer waiting for perfect research platforms. A reference humanoid that shortens the path from experiment to deployment lands in a market that is already moving.

The software around the robot

Agile ONE SN did not arrive alone. NVIDIA made Isaac Lab 3.0 generally available as an open-source robot learning framework, introduced Isaac OS as an open-source runtime connecting AI, sensors, controls and hardware, and released simulation-ready assets and a human activity dataset covering diverse movements with synchronized video and 3D hand tracking. Reference workflows cover refining GR00T models with reinforcement learning, training dexterous manipulation skills, and generating synthetic training data.

That ecosystem approach mirrors what smaller, stranger research machines are exploring from the other direction, like one robot that changes shape for every surface, launched this week for warehouses, roads and rough ground. The field is splitting into specialists and generalists. Agile ONE SN is a clear bet on the generalist – one body, many skills, shared tools.

What it means for labs, builders and investors

For research labs, availability is scheduled for early 2027, and NVIDIA says institutions including Stanford University and ETH Zurich are among those interested. A shared platform also means shared results: a skill trained at one university can, in principle, be validated at another on identical hardware. Reproducibility is an unglamorous word. In robotics, it is gold.

For builders and startups, the calculation changes. Instead of raising money to construct a humanoid from scratch, a small team can buy the reference machine and spend its budget on the one skill that makes its business different. For investors, that shifts where value collects – toward skills, data and deployment services, and toward the platform owners supplying the common base.

None of this puts a humanoid in every workplace next spring. Hands still need training, safety cases still need proof, and early 2027 hardware will go to frontier research teams first. But the direction is set. Humanoids are becoming a platform business, and this week the platform got a standard model that any serious lab can order, open up, and start teaching.

Also published on: