NVIDIA Just Built the Android of Robotics
NVIDIA released a complete open platform for physical AI — training tools, a reference humanoid robot, and over 2,000 simulation assets — cutting robot setup time from weeks to days.
- 1.40 m
- height of the Agile ONE SN humanoid
- 31
- joints on the Agile ONE SN
- 2,000+
- simulation-ready factory assets released
- 50%
- deployment time cut reported by Toyota Material Handling
Most news about NVIDIA focuses on chips. But what the company unveiled at GTC Berlin on October 21 is bigger. It released Isaac Lab 3.0, Isaac OS, and a reference humanoid robot — a complete, open-source toolkit for building robots that learn and work in the real world. NVIDIA is positioning itself as the Android of physical AI: an open platform, a reference device, and a data ecosystem for anyone building on top.
The new Agile ONE SN humanoid — built by Agile Robots on top of NVIDIA's tools — is the reference device for the platform. It stands 1.40 metres tall, has 31 joints, and five-fingered hands sensitive enough to handle fragile objects. It ships pre-loaded with NVIDIA's GR00T AI model. More important than the specs: NVIDIA says robot setup time drops from weeks to days. That is the real signal — the friction of starting a humanoid robotics project just dropped sharply.
The software side matters even more for most builders. Isaac Lab 3.0 is now openly available for training robot policies, from a laptop to a full data centre. NVIDIA also released more than 2,000 simulation-ready factory assets — robot arms, conveyor belts, shelves — so teams can teach a robot in a virtual world before it ever touches real equipment. Simulated training is fast and cheap; real-world testing is slow and expensive. Shifting that work into simulation is one of the most practical moves a robotics team can make right now.
Jensen Huang, NVIDIA's CEO, called what is coming 'the ChatGPT moment for robotics.' He picked that phrase carefully. ChatGPT did not invent AI — it made AI easy enough that any business could use it overnight. This week's release is the infrastructure layer that could do the same for physical robots. Hotels and warehouses are already testing it: Marriott and Hilton are piloting a robot concierge on the platform, and Toyota Material Handling reports deployment time cut by 50 percent. If you build software for physical environments, is your product ready when robots arrive?