Integrate Physical AI Capabilities into Existing Apps with NVIDIA Omniverse Libraries
NVIDIA is turning core Omniverse capabilities into modular libraries: ovrtx for RTX rendering, ovphysx for PhysX-based simulation, and ovstorage for data pipelines. They’re exposed as headless-first C APIs with C++ and Python bindings, so existing apps and services can integrate physical AI features without a full platform rewrite.
The practical upside is better fit for large robotics and industrial workflows: headless clusters, deterministic stepping, decoupled sensor rates, and easier CI/CD integration. The libraries are in early access on GitHub and NGC now, with API changes still possible; NVIDIA says production release and API stability are planned later this year. Isaac Lab 3.0 Beta is already moving to this modular model, including a choice between ovphysx and a Kit-less Newton/MuJoCo-Warp backend.
“These libraries reduce the need for major architectural rewrites.”
- what
- NVIDIA is adding modular Omniverse libraries: ovrtx, ovphysx, and ovstorage, instead of requiring the full Kit/container stack.
- who
- NVIDIA; internal users include Isaac Lab and Omniverse DSX Blueprint, with partners like ABB Robotics, PTC, Siemens, Adobe, Cadence, and Synopsys adopting or piloting it.
- when
- The libraries are in early access now on GitHub and NGC; production release and API stability are planned later this year. Isaac Lab 3.0 Beta is also out now, with GA later this year.
- impact
- Game and simulation teams can embed rendering/physics into existing tools, run headless on clusters, and avoid major replatforming or framework lock-in.
Useful modularization with clear integration wins
Follow AI updates
See relevant stories in your personalized news feed.
Discussion