NVIDIA Enhances Robot Simulations with Omniverse Libraries and OpenUSD
Tony Kim Aug 11, 2025 13:18
NVIDIA unveils updates to its Omniverse libraries and Cosmos world foundation models, enhancing robot simulations with OpenUSD technology for more accurate and reliable interactions.
NVIDIA has announced significant updates to its Omniverse libraries and Cosmos world foundation models (WFMs), aiming to enhance the development of robot simulations. These innovations, unveiled at the SIGGRAPH conference, are powered by OpenUSD technology, which facilitates the creation of physically accurate virtual environments and AI agents capable of interacting with real-world scenarios, according to NVIDIA's official blog.
Advanced Omniverse Libraries
The new Omniverse libraries are set to advance world composition applications by integrating 3D data for detailed simulations. This development is particularly beneficial for sectors such as robotics, autonomous vehicles, and industrial systems. The OpenUSD Exchange SDK 2.0 introduces modules for UsdPhysics authoring and a layered asset structure, enabling seamless integration of physics into robot models. This SDK, available on GitHub and PyPi, streamlines 3D data workflows, thereby accelerating robotics development.
Interoperable Data Pipeline
NVIDIA, in collaboration with Google DeepMind, has enhanced data interoperability between MuJoCo (MJCF) and USD formats. A new 'mujoco-usd-converter', utilizing OpenUSD Exchange SDK 2.0, allows for interoperable conversion of MJCF data to USD format. This tool is available for early release on GitHub and PyPi, offering developers a seamless way to integrate robot data into simulations.
Isaac Sim 5.0 and NVIDIA Isaac Lab 2.2
The early developer preview of Isaac Sim 5.0, available on GitHub, includes OpenUSD-based robot and OmniSensor schemas that standardize robot and sensor definitions. The PhysX extension now supports a new joint friction model, developed with Hexagon’s Robotics division and maxon, enabling simulated robot movements to more closely match real-life behavior.
Real-World Scene Reconstruction
The NVIDIA Omniverse NuRec libraries combine RTX ray tracing with 3D Gaussian splatting to convert real-world sensor data into interactive OpenUSD simulation environments. This innovation allows for high-fidelity world reconstruction, enhancing the realism and efficiency of simulations.
These updates to NVIDIA's Omniverse platform mark a significant step forward in the development of robot simulations, providing developers with the tools needed to create more accurate, reliable, and efficient virtual environments that closely mimic real-world interactions.
Image source: Shutterstock.jpg)
