NVIDIA developers used AI agents with Omniverse libraries to build simulation applications. Agents created editable building models from site scans for warehouse planning. They built a digital twin to test gas-turbine operating scenarios. Developers assembled simulation scenes with over 2,000 open-source SimReady assets. A warehouse simulation scaled to 80 conveyors, 40 arms, and 800 boxes at 30 fps. Agents tested robot control for flexible cable routing with reinforcement learning. CAD robot designs were converted into SimReady assets for pick-and-place testing. A humanoid simulator for warehouse tasks was built with first- and third-person views. An autonomous-driving testing workflow connected asset creation, traffic, and sensor simulation. Agents compared simulated and recorded sensor outputs to refine digital twins. Robo Olympics tested simulated humanoids performing sports movements. A car suspension was modeled and disassembled in simulation to inform robot tooling. An International Space Station model with telemetry was assembled and rendered in a browser.
What changed
Developers manually assembled scenes, configured physics, and iterated on simulations without AI assistance.
What it unlocks
Using natural-language instructions to direct AI agents that build, test, and refine simulations with NVIDIA Omniverse libraries.
- 2,000+ SimReady USD assets
- 80 conveyors, 40 arms, 800 boxes at 30 fps
- 640 successful box placements in 2 minutes
- 64 of 100 hurdle-clearing trials
What you need to act on it
- access to NVIDIA Omniverse libraries
- frontier AI models like GPT-6 Astra or Claude Fable 5
- GPU-accelerated hardware
Sources