On Monday at SIGGRAPH 2025, NVIDIA's research division revealed a comprehensive suite of robotics development tools. The company introduced new Omniverse libraries, Cosmos world foundation models, and AI computing infrastructure designed to accelerate robotics development and deployment.

The standout release is Cosmos Reason, an open 7-billion-parameter vision-language model that enables robots to understand and reason about and act in the physical world using their existing knowledge, the understanding of physical laws, and common sense. Cosmos Reason can assist in curating and annotating training datasets, act as the decision-maker in a vision language action (VLA) model, and serve as the basis for video analytics AI agents capable of video search and summarization.

Other highlights in the announcement include the New Omniverse NuRec libraries, which introduce 3D Gaussian splatting technology for large-scale world reconstruction, and Cosmos Transfer-2, which accelerates the generation of photorealistic synthetic data from inputs, such as 3D simulation scenes or spatial control inputs. Notably, NVIDIA's new models and libraries are powered by new RTX PRO Blackwell servers and DGX Cloud on Microsoft Azure.

Industry adoption is already underway, with Boston Dynamics, Figure AI, Amazon Devices & Services, and Uber integrating these tools into their workflows. The comprehensive platform addresses the entire robotics development pipeline—from simulation and training to deployment—marking a significant step toward mainstream physical AI applications across industries worth trillions of dollars.