Agile Robots launches Agile ONE SN, a small humanoid for research labs

It is a humanoid reference platform, which means a ready-made robot for people who want to build and test robot skills. Agile Robots says it gives researchers "one integrated system" to build, train and validate those skills in the lab.

The skills can then move into real-world work. That is the pitch: less time wiring tools together, more time on experiments.

Why it matters: Research teams often lose weeks on setup. Agile Robots says hardware, data collection, simulation, training and deployment usually rely on separate tools. NVIDIA says the new platform cuts setup time "from weeks to days."

Here are the main specs from Agile Robots and NVIDIA:

• Body: the Agile ONE S body, 1.4 meters (4.59 feet) tall, with 31 degrees of freedom.
• Hands: two Wuji five-finger hands with 20 degrees of freedom each and tactile sensing, for 71 degrees of freedom in total.
• Brain: NVIDIA Jetson Thor onboard compute for real-time sensor processing and robot inference.
• Eyes: RGB-D cameras on the head, wrist cameras for close-range work and 3D LiDAR.
• Strength: joint torque of up to 140 Nm and a payload of up to 3 kilograms per arm.
• Connections and safety: Wi-Fi 7, Bluetooth and an emergency stop button.

The two companies list slightly different weights. Agile Robots says 43 kilograms (95 pounds). NVIDIA's announcement says 40 kilograms (88 pounds).

Degrees of freedom are the number of ways a robot can move its joints. More of them means more human-like movement, and more to control.

In real life Think of a student lab that wants a robot to learn how to pick up a screw. With separate tools, someone first has to make the camera, the simulator and the robot agree with each other. Here they come in one package.

The robot ships with preconfigured NVIDIA Isaac GR00T software. That includes the GR00T 1.7 open model, whole-body control and an open workflow that covers the full development cycle.

Agile Robots lists the steps: demonstration capture, data generation, simulation, training, evaluation, deployment and real-world testing.

URDF and USD assets are included, so teams can bring the robot straight into NVIDIA Isaac Sim and Isaac Lab. In simple terms, there is a digital twin of the robot ready to practice in before it touches anything real.

NVIDIA's Deepu Talla, vice president of robotics and edge AI, said Agile ONE SN combines GR00T and Jetson Thor "in a single system that unifies demonstration capture, simulation, training, validation and deployment."

The platform is open and modular. Teams can use the full workflow or take selected parts into their own software. Agile Robots says that makes it easier to share and compare research across labs, and that researchers keep full control of their data.

NVIDIA says Jetson Thor delivers 2,070 FP4 teraflops of compute and 128 GB of unified memory. That lets the robot do its reasoning and control on its own body instead of relying on a distant server.

For a walking robot, that matters. A delay of even a moment between seeing and moving is a bad idea when you have legs.

Agile ONE SN is the compact reference platform in Agile Robots' humanoid line. Agile ONE is its full-size counterpart for industrial work.

Agile Robots says skills and data developed on the SN can later be transferred to the bigger robot. So the small one is the practice room, and the big one is the workplace.

The Robot Report covered the full-size robot when it was introduced in November 2025. Agile Robots said then that it planned to build Agile ONE in Bavaria starting in early 2026, and that it had designed the motors, actuators and control systems in Germany.

At that time, the company also said it focused on safe work next to people. CEO Zhaopeng Chen told The Robot Report: "We want to build a co-worker that humans can trust and that people are comfortable to work around."

Agile Robots was founded in Munich in 2018 by robotics researchers from the German Aerospace Center (DLR). It says it now employs more than 3,500 people worldwide and has one of the largest research and development teams in AI and robotics.

Its portfolio spans robot arms, mobile robots and AI software. Its subsidiaries include BÄR Automation, Franka Robotics, idealworks and Krause Automation.

When The Robot Report spoke with the company in 2025, Chen said it had installed more than 20,000 systems, mostly in the automotive and consumer electronics sectors. That is the base the new humanoid grows from.

The big picture: Chen has said that data is the heart of the company's approach. He described a "data pyramid" that starts with real-world teleoperation, adds simulation and tops it with video and images. The new research robot is a way to bring more labs into that kind of work.

NVIDIA says Agile Robots plans to integrate on-robot safety protocols with NVIDIA Halos, its full-stack safety system for robotics. Agile Robots says it is exploring that integration across design, validation and software.

Both companies use careful wording here: it is planned, not finished. Safety also matters more for a robot that labs will test in unpredictable ways.

Agile Robots says the platform draws interest from leading institutions such as Stanford University and ETH Zurich. It is scheduled to be available at the beginning of 2027.

Neither company has given a price. Visitors to GTC Berlin can see the robot at the NVIDIA booth and at Agile Robots' booth 3007.

Agile ONE SN was part of a larger NVIDIA Isaac announcement at GTC Berlin. NVIDIA also said that Isaac Lab 3.0 is now generally available, and that it is releasing a new open source runtime called Isaac OS.

It also released a Human 1M Dataset and a SimReady Central Library with more than 2,000 simulation-ready robotics and factory assets on Hugging Face. NVIDIA founder and CEO Jensen Huang said "the ChatGPT moment for robotics is fast approaching."

We have covered other humanoids heading toward work, such as Agility's Digit 5 battery deal and Tesla's Optimus ramp.

• Not for homes: this is a lab tool for researchers and developers, and it is due in early 2027.
• Faster robot skills: a shared, open toolkit could help labs test ideas sooner, which feeds into the robots that reach workplaces later.
• Watch the ecosystem: NVIDIA's platform is now the base for several humanoids, so progress in one lab can help many.
• Price still unknown: neither company has said what it will cost.