Nvidia Unveils New AI & Robotics Tools at SIGGRAPH
Nvidia has unveiled a significant expansion of its technology stack aimed at accelerating the development of advanced robotics, blending the realms of computer graphics and artificial intelligence. Announced at the SIGGRAPH conference in Vancouver, these innovations include new Omniverse libraries, advanced Cosmos Physical AI Models, and robust AI computing infrastructure, all designed to make robots and autonomous systems more capable, adaptable, and commercially viable.
At the core of Nvidia’s robotics push are new Omniverse libraries, which enhance simulation capabilities crucial for robot design and deployment. Among these is the Omniverse NuRec 3D Gaussian Splatting library, a powerful tool that enables developers to capture, reconstruct, and simulate real-world environments in high fidelity using sensor data. This is vital for creating accurate “digital twins” where robots can learn and operate safely before real-world deployment. Furthermore, new Omniverse Software Development Kits (SDKs) are fostering greater interoperability by supporting data exchange between MuJoCo (MJCF) and Universal Scene Description (OpenUSD) formats, opening up seamless simulation possibilities for a vast community of robot learning developers. The company has also updated its open-source robot simulation and learning frameworks, NVIDIA Isaac Sim 5.0 and NVIDIA Isaac Lab 2.2, now available on GitHub, to integrate these enhancements and help bridge the “simulation-to-reality” gap.
Complementing the simulation advancements are the new Cosmos Physical AI Models, critical for imbuing robots with intelligence and reasoning abilities. The standout is Cosmos Reason, a 7-billion-parameter vision-language model (VLM) specifically designed for physical AI and robotics. This model empowers robots and vision AI agents to reason like humans, leveraging prior knowledge, physics understanding, and common sense to interpret and act in the real world. This capability is set to transform tasks like data curation, robot planning, and video analytics. Another key addition is Cosmos Transfer-2, which streamlines the generation of synthetic datasets from 3D simulation scenes, drastically reducing the time and cost associated with producing realistic training data for robots.
To support these demanding AI and simulation workloads, Nvidia has introduced new AI computing infrastructure. This includes the NVIDIA RTX PRO Blackwell Servers, offering a unified architecture for every stage of robot development—from training and synthetic data generation to robot learning and simulation. Additionally, NVIDIA DGX Cloud is now available on Microsoft Azure Marketplace, providing a fully managed platform that simplifies the streaming of OpenUSD and RTX-based applications from the cloud at scale, minimizing the complexities of infrastructure management for developers.
Nvidia’s overarching strategy centers on “Physical AI,” a term that signifies the convergence of artificial intelligence and computer graphics to create systems capable of interacting intelligently with the physical world. Rev Lebaredian, vice president of Omniverse and simulation technologies at Nvidia, emphasized that this blend of AI reasoning and scalable, physically accurate simulation is poised to fundamentally transform robotics and autonomous vehicles across multi-trillion-dollar industries. This vision aligns with Nvidia CEO Jensen Huang’s declaration that the 2020s will be the “robot decade,” where AI-driven machines will revolutionize various sectors from manufacturing and logistics to healthcare.
The industry’s embrace of these new technologies is already evident, with leading companies like Amazon Devices & Services, Boston Dynamics, Figure AI, and Hexagon adopting Nvidia’s simulation and synthetic data generation tools. Other notable adopters include Uber, which is utilizing Cosmos Reason for annotating autonomous vehicle training data, and Foretellix, integrating NuRec to enhance synthetic data generation for autonomous vehicles.
Nvidia’s latest announcements at SIGGRAPH 2025 underscore its strategic positioning as a foundational technology provider for the burgeoning field of advanced robotics. By offering a comprehensive ecosystem of simulation tools, intelligent AI models, and powerful computing infrastructure, the company is enabling developers to build the next generation of robots and autonomous systems that can perceive, reason, and act with unprecedented capability in the physical world.