Nvidia has introduced a fresh lineup of AI models, libraries, and infrastructure for robotics developers, headlined by Cosmos Reason, a 7-billion-parameter vision-language model designed for “reasoning” in physical AI applications and robotics.
Part of the expanded Cosmos world models family, Cosmos Reason is built to give robots and AI agents memory and physics-based understanding, enabling them to plan next steps for tasks such as robot planning, data curation, and video analytics. Joining it are Cosmos Transfer-2, which speeds up synthetic data generation from 3D simulations or spatial controls, and a distilled version optimized for faster performance.
Related: DOJ Arrests Two Over Nvidia H100s Smuggling to China-Nvidia Rejects Kill Switch Proposal
The models are intended to help create synthetic text, image, and video datasets for training embodied AI systems. Nvidia also announced neural reconstruction libraries, including new rendering techniques that simulate the real world in 3D from sensor data. These capabilities are being integrated into CARLA, the open-source simulator, alongside updates to the Omniverse SDK.
For hardware and infrastructure, the company debuted the RTX Pro Blackwell Server for robotics workflows and DGX Cloud, a cloud-based management platform for AI development.
The launch, announced at SIGGRAPH, signals Nvidia’s deeper push into robotics as it looks to extend its AI GPU dominance beyond data centres to the next major AI frontier.
