The clip is NVIDIA’s two-minute, three-second GTC Berlin explainer, and it centers the Agile ONE SN, the reference humanoid Agile Robots announced at the same event on October 21, 2026. The description says frontier AI agents build the training worlds, replay real-world motion and tune the physics before the robot ever moves. The robot itself stands 4.59 feet (1.40 m) tall and weighs 88 pounds (40 kg). The open question is what this pipeline changes for a robot that most developers have not yet held.
Prompts, training worlds and the Agile ONE
The video runs 2 minutes 3 seconds and has no narration, so the description carries the account. Frontier AI agents build training worlds, replay real-world motion and tune physics using, in the description’s words, NVIDIA open libraries and models. The skills are trained with NVIDIA Isaac Lab and Newton. They are evaluated with NVIDIA Isaac Lab-Arena. All of it runs locally on an NVIDIA DGX Station. Then the Agile ONE brings those skills into the physical world. The description says the robot is built on NVIDIA Isaac GR00T and powered by NVIDIA Jetson Thor. The press release adds the hardware the clip is about: a 31-degree-of-freedom body, dual WUJI Hand 2 five-finger hands, head and wrist cameras, lidar sensors and inertial measurement units.
A second reference humanoid in eight months
This is the second reference humanoid design tied to the Isaac platform in eight months. NVIDIA announced its own Isaac GR00T Reference Humanoid Robot on May 31, 2026, a full open hardware stack assembled from a Unitree H2 Plus chassis, Sharpa Wave hands and a Jetson AGX Thor T5000. The Agile ONE SN is different: it is a machine Agile Robots built, preconfigured with Isaac GR00T and a Jetson Thor, with planned integration of the NVIDIA Halos safety stack. The press release says the platform cuts setup time from weeks to days. The GTC Berlin release also names ABB Robotics, Flexion, Hexagon Robotics, Humanoid, KUKA, NEURA Robotics, Sereact and Universal Robots as Europe-based builders on the Isaac platform.
What a follow-up clip could show
No task is named in the title or description, so a follow-up clip of the Agile ONE doing a specific job would be the natural next piece. The hands are the most interesting part on paper: 20 degrees of freedom each, 1kHz force control and tactile sensing. A clip of the WUJI Hand 2 closing on a thin object would tell more about the pipeline than any walkthrough of the training stack. The description says skills are evaluated with Isaac Lab-Arena, but no success rate or benchmark figure appears. A number attached to a named task would close that gap.
Stated and not stated
- Stated: the video is 2 minutes 3 seconds long, from NVIDIA’s channel, published October 23, 2026, with no spoken narration.
- Stated: the description names Isaac Lab, Newton, Isaac Lab-Arena, DGX Station, Isaac GR00T, Jetson Thor and the Agile ONE.
- Stated: the press release gives the body specs: 1.40 m (4.59 ft), 40 kg (88 lb), 31 degrees of freedom, 3 kg (6.6 lb) payload per arm.
- Not stated: the playback speed, whether the run was teleoperated or autonomous, or which task the robot performs.
- Not stated: any price, availability date or unit count for the Agile ONE SN.
Who else is on the same platform
NVIDIA does not sell robots; it sells the Jetson computers, the Isaac training stack and the GR00T open models, and the robots shown in its releases are built by partners. That pattern is visible here: the hardware is Agile Robots’, the platform is NVIDIA’s. Rivals working on the same general-purpose problem include Google DeepMind, Physical Intelligence, Skild AI, Figure AI and Tesla. For the platform to matter beyond a demo clip, a developer who buys an Agile ONE SN or the Unitree-built reference humanoid would have to train a useful skill without waiting weeks for setup. If the press release’s weeks-to-days claim holds, that is the number to watch next.
Watch next
Kamino Simulates 4,096 Robots on One GPU · Agility Robotics Introduces Digit 5 Humanoid Robot
Source: Accelerating Robot Development With AI Agents and NVIDIA Isaac by NVIDIA , published 2026-10-23. This article was written automatically from the video above and has not been independently verified.