Nvidia unveiled a new collection of AI models, libraries, and infrastructure for robotics developers on Monday. The most notable addition is Cosmos Reason, a 7-billion-parameter vision language model designed for physical AI applications and robots. This model enables robots and AI agents to “reason” by leveraging memory and physics understanding, making it useful for data curation, robot planning, and video analytics.
Joining the existing lineup of Cosmos world models are Cosmos Transfer-2, which speeds up synthetic data generation from 3D simulation scenes or spatial control inputs, and a distilled version optimized for faster performance. Nvidia announced these models at the SIGGRAPH conference, emphasizing their role in creating synthetic text, image, and video datasets for training robots and AI agents.
In addition to the models, Nvidia introduced new neural reconstruction libraries, including one for a rendering technique that allows developers to simulate the real world in 3D using sensor data. This capability is being integrated into the open-source simulator CARLA, a widely used developer platform. The company also released an update to the Omniverse software development kit.
Nvidia expanded its hardware offerings with new servers tailored for robotics workflows. The RTX Pro Blackwell Server provides a unified architecture for robotic development, while DGX Cloud serves as a cloud-based management platform. These developments highlight Nvidia’s growing focus on robotics as it explores new applications for its AI GPUs beyond data centers.
The announcements underscore Nvidia’s commitment to advancing AI-driven robotics, offering tools that enhance synthetic data generation, simulation, and real-world applications.