On Monday, Nvidia introduced new world AI models, libraries, and infrastructure for robotics developers, with Cosmos Reason standing out as a 7-billion-parameter reasoning vision language model for physical AI and robots.
The new additions to the existing Cosmos world models include Cosmos Transfer-2, which accelerates synthetic data generation from 3D simulation scenes or spatial control inputs, and a faster, distilled version of Cosmos Transfers.
At the SIGGRAPH conference, Nvidia announced that these models are intended to create synthetic text, image, and video data sets for training robots and AI agents.
According to Nvidia, Cosmos Reason enables robots and AI agents to reason due to its memory and physics understanding, allowing it to serve as a planning model to determine the next steps for an embodied agent. It is intended for data curation, robot planning, and video analytics.
Nvidia also introduced new neural reconstruction libraries, including a rendering technique for simulating the real world in 3D using sensor data. This rendering is being incorporated into the open-source simulator CARLA, a popular developer platform, along with an update to the Omniverse software development kit.
New servers for robotics workflows were also revealed. The Nvidia RTX Pro Blackwell Servers provide a unified architecture for robotic development workloads, while Nvidia DGX Cloud serves as a cloud-based management platform.
These announcements come as the semiconductor giant deepens its focus on robotics, looking to expand the use of its AI GPUs beyond AI data centers.
Techcrunch event
San Francisco
|
October 27-29, 2025