NVIDIA Opens Omniverse Libraries to Speed Physical AI Simulation
The open-source release supplies GPU‑accelerated sensor simulation, physics, and CAD-to-OpenUSD conversion so developers can test robots and industrial AI before real‑world deployment.
Overview
- NVIDIA expanded its Agent Toolkit by publishing three Omniverse libraries on GitHub: ovrtx for RTX-powered sensor outputs, ovphysx for GPU-accelerated physics, and CAD-to-SimReady to convert CAD files into OpenUSD simulation assets.
- The company also released a Blender integration blueprint and a SimReady sample workflow, and demonstrated a SimReady Blender pipeline at SIGGRAPH using NemoClaw to show how assets move from creation to simulation.
- The new libraries are open-source to encourage wider adoption and let developers embed sensor, physics, and validation steps into cloud and local workflows without rebuilding core simulation code.
- Several partners and vendors are already integrating the libraries into authoring and CAD tools, with early adopters reported to include SideFX (Houdini workflows), PTC (Onshape), ForgeCAD, Lightwheel, Moonlake AI, and Palatial.
- Realistic simulation is essential for safe physical AI development because virtual sensors and GPU physics let teams validate robot and machine behavior before field tests, and OpenUSD provides a common 3D interchange format for those simulations.