NVIDIA Unveils Omniverse Libraries and Cosmos Models to Accelerate Robotics
NVIDIA announces new Omniverse libraries and Cosmos world foundation models to help developers build physically accurate simulations and generate synthetic data for robotics. Major physical AI leaders are already adopting these advanced tools.
NVIDIA announces new Omniverse libraries and Cosmos world foundation models at SIGGRAPH to accelerate robotics development. These tools allow developers to build physically accurate digital twins, capture the real world in 3D simulation, and generate synthetic data for training physical AI models. The company highlights that the convergence of computer graphics and AI fundamentally transforms how developers build robots and autonomous vehicles.
New Omniverse SDKs enable seamless data interoperability between MuJoCo and Universal Scene Description, giving over 250,000 robot learning developers cross-platform simulation capabilities. The new NuRec libraries introduce ray-traced 3D Gaussian splatting, which lets developers reconstruct real-world environments using sensor data. Additionally, NVIDIA releases Isaac Sim 5.0 and Isaac Lab 2.2 as open-source frameworks that integrate this neural rendering to bridge the gap between simulation and reality.
Physical AI leaders like Amazon Devices & Services, Boston Dynamics, Figure AI, and Hexagon embrace these new simulation and synthetic data generation tools. Developers run these demanding workloads using new NVIDIA RTX PRO Blackwell servers and NVIDIA DGX Cloud. Autonomous vehicle toolchain leader Foretellix and the popular CARLA simulator are also integrating NuRec, Omniverse Sensor RTX, and Cosmos models into their platforms.