Nvidia robots: from a Star Wars droid to a $150 school kit

Nvidia does not sell a robot you can buy in a shop, and it is still the company most robots are now built on. At GTC Berlin on 21 October it laid out the rest of the stack: Isaac Lab 3.0 for teaching robots in simulation, an open-source runtime called Isaac OS, a library of more than 2,000 simulation-ready parts from makers including FANUC and Universal Robots, and a dataset of a million human activities recorded with synchronised video and 3D hand tracking.

The showpiece is the Isaac GR00T reference humanoid, announced in June: a Unitree H2 body just under six feet tall and 150 pounds, with 31 joints, five-fingered Sharpa hands and a Jetson Thor computer carrying 128GB of memory inside its chest. Ai2, ETH Zurich, the Stanford Robotics Center and UC San Diego are building research on that exact design. At the same Berlin event Agile Robots showed the Agile ONE SN, 1.40m and 40kg, with two hands of twenty joints each, due early 2027. What both of them run on is a foundation model, Isaac GR00T N1.6, trained in simulation before it ever moves a real joint.

Where a child actually touches it

None of which answers the question a ten-year-old asks, which is whether any of it is reachable from a bedroom desk.

“General-purpose robotics is advancing at extraordinary speed, the ChatGPT moment for robotics is fast approaching.”

— Jensen Huang, founder and CEO of Nvidia, GTC Berlin, 21 October 2026, https://nvidianews.nvidia.com/news/nvidia-advances-isaac-platform-for-general-purpose-robotics-with-new-reference-humanoid-frameworks-simulation-pipelines-and-data

Two of these robots are, in different ways. The first is the one with a name children already know. Nvidia built an open-source physics engine called Newton together with Google DeepMind and Disney Research, and the first robot shown running on it was Blue, a BDX droid from the Star Wars films, walking on stage beside Jensen Huang. Those droids are going into Walt Disney World, Tokyo Disneyland and Disneyland Paris. The engine that makes Blue stumble and catch itself convincingly is the same one researchers use, it simulates cloth, sand and food, and anyone can download it.

The second is smaller and much closer. JetBot is a two-wheeled robot with a camera, built on a Jetson Nano, open-sourced by Nvidia and costing under $150 on top of the board. You program it from a browser, the tutorials run from making the wheels turn to teaching it to avoid a collision by showing it photographs, and Carnegie Mellon's Robotics Academy trains teachers on it. Nvidia's Deep Learning Institute gives schools an Edge AI and Robotics teaching kit free, built with Oxford and UMBC.

What it costs to start

The honest caveat is that the simulator wants a decent graphics card, so the free part is free only if the family computer already has one. What a child gets for the price of a bicycle, though, is the same software name on their screen as the one in the Stanford lab, and we have watched that stack closely enough to say the continuity is real rather than marketing. Start with the wheels turning. The humanoid can wait.

Informational material, not investment advice. Prices and availability dates are the manufacturers' own, checked on 25 October 2026, and the Agile ONE SN is a 2027 announcement rather than a shipping product.