Agile Robots Unveils Agile ONE SN as NVIDIA Expands Its Isaac Robotics Platform

Agile Robots has introduced Agile ONE SN, a compact research humanoid built around NVIDIA's robotics software and onboard computing. Announced at GTC Berlin on October 21, the platform is scheduled for availability in early 2027, offering labs an integrated starting point for developing robot skills.

Key Takeaways

Show all

• Agile ONE SN combines a 1.4-meter humanoid with tactile hands, Jetson Thor and preconfigured Isaac GR00T software.
• It is a research platform, with a proposed path for transferring skills to the larger industrial Agile ONE.
• NVIDIA's accompanying announcement expands Isaac with a new runtime, learning tools and datasets.
• Availability is planned for early 2027; Halos safety integration remains in development.

A compact platform for humanoid research

According to Agile Robots' announcement, the robot uses an Agile ONE S body with 31 degrees of freedom and two Wuji tactile hands with 20 each, bringing the total to 71. Payload is rated at up to 3 kilograms per arm. The release lists a weight of 43 kilograms, while NVIDIA's announcement gives 40 kilograms.

The product page describes head and wrist cameras, additional depth cameras and waist-mounted lidar. Jetson Thor handles onboard AI processing, complemented by an Intel base computing module.

The workflow starts with recorded demonstrations or simulated data, moves through training and virtual evaluation, and ends with testing skills on the physical robot. GR00T 1.7 and whole-body control come preconfigured, with robot description assets for Isaac Sim and Isaac Lab. Researchers can adopt selected components rather than the entire software stack.

Agile says skills and data developed on SN can later transfer to its full-size Agile ONE. That makes the smaller machine a potential development tool for industrial applications, although transfer between different bodies still needs to be demonstrated for particular tasks.

NVIDIA expands the tools around the robot

The launch is part of a broader NVIDIA Isaac announcement. Isaac OS, a new open-source runtime built on Holoscan, connects models, sensors, controls and hardware. Isaac Lab 3.0 is generally available for robot learning across multiple GPUs, supporting Newton and PhysX physics engines.

NVIDIA also announced a library of more than 2,000 simulation-ready assets and the Human 1M Dataset, which combines human activity footage with multiple camera views and hand tracking. Access to the full dataset is reserved for Cosmos Coalition partners.

Three reference workflows for DGX Station cover reinforcement learning for GR00T, dexterous manipulation and synthetic data generation using Cosmos and ChatGPT Astra. Together, these additions address the infrastructure needed to train and test robots, beyond the hardware itself.

Halos integration is planned for Agile ONE SN. The announcements should not be read as confirmation that the robot already incorporates the complete safety platform or has achieved a particular safety certification.

Another route to an integrated research robot

The approach follows the NVIDIA and Sharpa reference humanoid we covered in June, which combined a Unitree H2 Plus body, Sharpa tactile hands and NVIDIA computing and software. Agile ONE SN offers another hardware configuration around the same ambition: reducing the integration work laboratories face before experiments can begin.

For Agile Robots, it adds a research-focused offering alongside its industrial push. Our previous Agile coverage includes the original Agile ONE launch and collaborations with Google DeepMind and Hitachi. The new platform gives that story a further route from laboratory development toward factory use.