NVIDIA uses GPT-6 Astra AI agents in Omniverse to automate complex simulations

NVIDIA announced that its GPT-6 Astra model can be directed with natural‑language prompts inside the Omniverse platform to create simulation applications.

In one demo, Astra turned panoramic photos and point‑cloud measurements into a clean building geometry, stored it in OpenUSD and rendered it with the ovrtx library for inspection.

Another demo built a digital twin of a gas turbine, letting users change load, temperature and fuel composition. Astra used Cantera and SciPy for the physics and linked the results to animated 3D visualizations streamed via ovstream.

A warehouse stress test used Astra to scale a scene to 80 conveyors, 40 robot arms and 800 boxes, keeping 30 fps at full HD for a two‑minute run and completing 640 box placements before performance dropped when the scene doubled.

Using the SimReady Central Library, developers accessed more than 2,000 USD assets; a kitchen scene was assembled with 180 placements from 63 unique assets and 18 custom geometry pieces, rendered with Isaac Sim.

Across these examples, Omniverse libraries such as ovstage, ovrtx and ovstream handled physics, rendering and streaming while Astra managed high‑level orchestration, reducing manual coding effort.