Inbizzy, SIGGRAPH — NVIDIA has officially introduced a suite of new technologies designed to accelerate the development and deployment of robotics solutions powered by physical AI. The innovations include Omniverse™ NuRec 3D Gaussian Splatting libraries, Cosmos™ world foundation models (WFMs), and the latest AI computing infrastructure such as NVIDIA RTX PRO™ Blackwell Servers and NVIDIA DGX™ Cloud.
Industry leaders including Amazon Devices & Services, Boston Dynamics, Figure AI, and Hexagon are among the first adopters, using these tools for large-scale simulation and synthetic data generation.
Merging AI and Physical Simulation for the Robots of Tomorrow
According to Rev Lebaredian, NVIDIA’s Vice President of Omniverse and Simulation Technologies, computer graphics and AI are converging to fundamentally transform robotics.
“By combining AI reasoning with scalable, physically accurate simulation, we’re enabling developers to build tomorrow’s robots and autonomous vehicles that will transform trillions of dollars in industries,” he said.
Omniverse Libraries for 3D World Reconstruction
The updated NVIDIA Omniverse SDKs now enable data interoperability between MuJoCo (MJCF) and OpenUSD, allowing more than 250,000 robot learning developers to run cross-platform simulations seamlessly.
Omniverse NuRec introduces ray-traced 3D Gaussian splatting, enabling developers to capture, reconstruct, and simulate the real world using sensor data.
NVIDIA also released Isaac Sim 5.0 and Isaac Lab 2.2 on GitHub, both featuring NuRec neural rendering and new OpenUSD-based robot and sensor schemas to bridge the simulation-to-reality gap.
NuRec is now integrated into the CARLA simulator, with adoption already underway by Foretellix, Voxel51, Ford, and Porsche.
Cosmos: World Generation and Spatial Reasoning
The Cosmos WFMs, downloaded over 2 million times, receive a significant upgrade with Cosmos Transfer-2, which speeds up the creation of photorealistic synthetic data from 3D simulation scenes or spatial inputs.
A distilled version of Cosmos Transfer compresses the 70-step distillation process into one, allowing unprecedented speed on RTX PRO servers. Early adopters include Lightwheel, Moon Surgical, and Skild AI.
Meanwhile, Cosmos Reason, a 7-billion-parameter vision language model (VLM), introduces advanced reasoning capabilities to robots and AI agents. Its applications include:
- Automated data curation and annotation
- Robot planning with common-sense reasoning
- Large-scale video analytics and root-cause analysis
Uber, Magna, VAST Data, Milestone Systems, and Linker Vision are among the early users.
AI Infrastructure for Robotics Workloads
NVIDIA introduced RTX PRO Blackwell Servers as a unified platform for training, synthetic data generation, robot learning, and simulation.
In addition, NVIDIA DGX Cloud, now available on Microsoft Azure Marketplace, offers Omniverse developers a fully managed platform for streaming OpenUSD- and RTX-based applications at scale. Early adopters include Accenture and Hexagon.
Boosting the Developer Ecosystem
To further strengthen robotics and physical AI adoption, NVIDIA announced:
- OpenUSD Curriculum and Certification, in collaboration with Adobe, Amazon Robotics, Autodesk, Pixar, Siemens, and others.
- An open-source collaboration with Lightwheel to integrate robot policy training and evaluation frameworks into Isaac Lab, including parallel reinforcement learning capabilities and simulation-ready assets.
With its combination of precise simulation libraries, world reasoning AI models, and high-performance AI infrastructure, NVIDIA is positioning itself as a driving force in robotics innovation.
These advancements not only target industrial developers but also pave the way for the next generation of robots and autonomous systems ready to tackle real-world challenges.
