Nvidia developers are combining frontier AI agents with Omniverse simulation libraries to build editable digital environments from natural-language instructions. The examples focus on reducing the setup work required before engineers can test warehouse layouts or industrial operating scenarios.
In one prototype, a developer directed GPT-6 Astra to interpret panoramic photographs, create building geometry and align it with point-cloud measurements. The resulting OpenUSD model removes existing contents so shelving and machinery can be repositioned before physical changes are made. Omniverse libraries supply rendering and physics capabilities for inspection.
A second project wrapped a simulated gas turbine and generator in a what-if application. Users can vary load, ambient temperature and fuel composition, then compare power, efficiency, fuel use and equipment response. Six procedures cover conditions including a stuck fuel valve and an exhaust-sensor error.
These are developer demonstrations, not evidence that an agent can certify a building or operate industrial equipment. People supplied the objectives, selected tools and reviewed outputs, while the results remain simulations whose assumptions need validation. The useful change is narrower: agents can help assemble assets, code and interfaces that previously made exploratory simulation slow to start, giving engineers more scenarios to examine before real-world testing.