Agile Robots and Nvidia launch a humanoid platform for AI research

A bundled research platform

Agile Robots and Nvidia recently introduced the Agile ONE SN, a humanoid robot designed specifically for research environments. Unlike previous systems where labs had to piece together separate hardware and software components, this platform arrives as a preconfigured package. It combines Agile Robots' physical body with Nvidia's high-performance computing tools.

The robot is built using the smaller Agile ONE S body, which stands about 4 feet, 7 inches tall. It is designed to be a compact workspace for researchers to teach machines new skills. While the company also produces a full-sized humanoid for industrial use, the ONE SN focuses on the development phase of robotics.

By bundling these technologies, the companies aim to simplify the development cycle for academic and private labs. The goal is to reduce the time spent on technical integration, which the companies claim could turn weeks of work into just a few days. This streamlined approach allows researchers to spend more time on experimentation and less on troubleshooting basic connectivity.

Hardware and sensor specs

The Agile ONE SN is equipped with a high degree of mobility to mimic human movement. It features 31 degrees of freedom in its body and 20 in each hand, totaling 71 points of movement. Its hands are five-fingered Wuji hands that include tactile sensing and force control for grasping objects.

To interpret its surroundings, the robot uses a variety of sensors. These include head-mounted RGB-D cameras, wrist cameras, 3D lidar, and motion sensors. These tools allow the robot to see and map its environment while performing tasks. It also includes standard connectivity like Wi-Fi 7 and Bluetooth, along with a physical emergency stop button.

The robot is capable of carrying a payload of up to 3 kilograms per arm. While Agile Robots lists the weight as 43 kilograms, Nvidia lists it as 40 kilograms. The companies did not provide an explanation for the discrepancy in these figures. This physical setup is intended to help researchers test how the robot responds to different physical tasks.

The Nvidia AI stack

At the core of the robot is Nvidia's Jetson Thor platform, which handles the heavy lifting for AI processing. This system provides 128GB of unified memory and significant computing power for AI-driven movement. It allows the robot to process complex sensor data in real time to coordinate its actions.

The robot comes preconfigured with Isaac GR00T, which includes a vision-language-action model. This software allows researchers to move through a four-stage development process. They can capture demonstrations of tasks, create digital simulations, train and evaluate control policies, and finally deploy them on the physical hardware.

To make this easier, Agile Robots provides simulation assets so developers do not have to build digital models from scratch. Researchers can use these tools to recreate the robot and its environment in a virtual space. This allows them to test behaviors in a safe, simulated setting before trying them on the actual robot.

From lab to industrial use

A major goal for Agile Robots is the transfer of skills between different sizes of robots. They intend for the skills developed on the compact ONE SN to eventually work on their larger industrial humanoid. However, the companies have not yet published independent results to prove how reliably this transfer works.

The robot is positioned differently than some other recent humanoid releases. While some competitors focus on showing off physical feats like martial arts, the ONE SN is marketed as a development platform. It is meant to be a tool for creating repeatable performances rather than just a demonstration of what a robot can do today.

The platform is modular, meaning researchers can use the entire workflow or just specific parts. They can combine individual components with their own existing software if they prefer. This flexibility is designed to help labs move from laboratory experiments to the kind of repeatable performance required for real-world tasks.

Timeline and availability

The Agile ONE SN is not expected to be available until early 2027. Neither Agile Robots nor Nvidia has announced a specific price for the research robot yet. Several institutions, including Stanford University and ETH Zurich, have expressed interest in the platform, though no official customers have been confirmed.

The companies are also exploring integration with Nvidia's Halos safety platform. This work is still in development, and the announcement does not state that the robot is certified for unsupervised industrial use. Researchers will maintain control of their data and can choose which parts of the platform to utilize.

As the robotics industry works to move humanoids into practical roles, the ONE SN represents an early step. It provides a connected set of training tools to reduce the time spent on hardware and software integration. Its 2027 release will be a test of whether this integrated approach actually speeds up the development of robotic skills.