Agile Robots and Nvidia team up to produce 2,000 humanoid robots

A standardized research platform

Germany’s Agile Robots is preparing to manufacture up to 2,000 compact humanoid robots by the year 2027. The company unveiled its Agile ONE SN model at Nvidia’s GTC event in Berlin to target universities and corporate research labs. These institutions are currently working to develop the skills robots will need for future industrial use.

The primary goal of this initiative is to provide a ready-to-use platform for researchers who want to focus on robot behavior rather than basic setup. As eweek.com reported, it often takes months to sync hardware, sensors, simulation tools, and AI software before developers can actually begin testing new skills. Agile and Nvidia want to eliminate this hurdle by providing a preconfigured system.

The Agile ONE SN stands 1.4 meters tall and features a humanoid body equipped with five-finger tactile hands. It includes 3D lidar, cameras, and 71 degrees of freedom across the body and hands. This configuration allows for coordinated movement and more dexterous manipulation of objects compared to simpler robotic designs.

The Nvidia software stack

The robots are powered by Nvidia’s Jetson Thor onboard computer and the Isaac GR00T robotics software. This software includes whole-body control tools and a specific workflow for capturing demonstrations and generating data. These tools allow researchers to train models and test new skills in a virtual environment before moving to physical hardware.

The platform supports Nvidia Isaac Sim and Isaac Lab, which are tools for developing and evaluating robot behaviors in digital worlds. By using these simulations, teams can refine how a robot moves or interacts with objects without the risk of damaging physical equipment. This creates a bridge between digital training and real-world deployment.

Nvidia product director Spencer Huang said that a standardized configuration could save researchers two or three months of integration work. By providing a unified stack, Nvidia is positioning its technology as the backbone for physical AI. This allows Agile to access a high-end computing and software suite that it would otherwise have to build from scratch.

From lab to factory floor

Agile Robots is designing the ONE SN as a smaller version of its full-size Agile ONE industrial humanoid. The company intends for skills and data developed on the smaller research models to eventually transfer to the larger industrial units. This creates a potential pathway for moving laboratory experiments into practical factory applications.

The company has already seen interest from major institutions like Stanford University and ETH Zurich. While the announcement does not confirm specific orders from these schools, the interest highlights the demand for accessible humanoid hardware. The robots are not intended for immediate mass deployment in factories, but rather for the research that precedes it.

The company plans to start manufacturing the ONE S and ONE SN models by the end of 2026. Agile is targeting the production of 1,000 to 2,000 units in the following year. This expansion of hardware will give more research teams the ability to test and compare robot behaviors across different institutions.

Challenges in industrial readiness

While the partnership speeds up development, it does not guarantee that the robots are ready for immediate industrial use. Transferring a skill from a small research robot to a large industrial one does not ensure identical performance. Factors like different factory layouts, operating conditions, and hardware variations may require significant additional training and validation.

Agile is also exploring the integration of Nvidia’s Halos safety platform, though this work is still in the exploration phase. The company notes that this exploration should not be viewed as proof that the platform is currently validated for working around humans. Safety remains a critical hurdle for any humanoid robot intended for human-centric environments.

A major commercial question remains regarding whether the productivity gains from these robots will cover the costs of maintaining and supervising them. For enterprise buyers, the focus will need to stay on task-success rates and total operating costs. For now, the project serves as a way to build a pipeline of skills for a future where humanoids handle complex manual tasks.