On Monday, NVIDIA launched a suite of global AI models, libraries, and infrastructure tailored for robotics developers. Among these innovations is Cosmos Reason, a 7-billion-parameter vision-language model designed specifically for physics-based AI applications and robotics. This "reasoning" model incorporates memory and physics understanding capabilities to enable advanced planning functions for robotic agents.

The new models expand NVIDIA's existing product lineup. Cosmos World Models include Cosmos Transfer-2, which accelerates synthetic data generation from 3D simulations or spatial control inputs. A streamlined version, Cosmos Transfers, has also been optimized for faster performance.

Announced during the SIGGRAPH conference on Monday, NVIDIA emphasized these models will be used to create synthetic text, image, and video datasets for training robots and AI agents. According to NVIDIA, Cosmos Reason enables robotic systems to "reason" by serving as a planning model that predicts potential next actions in physical environments. The technology can be applied to data curation, robotic planning, and video analysis tasks.

The company also introduced new neural reconstruction libraries featuring rendering technology that allows developers to simulate real-world environments in 3D using sensor data. These rendering capabilities have been integrated into CARLA, an open-source simulation platform widely used by developers. Updates were also made to the Omniverse SDK.

NVIDIA unveiled new server solutions for robotics workflows, including the RTX Pro Blackwell server which provides a unified architecture for robotics development workloads. DGX Cloud, a cloud-based management platform, was also announced as part of this expansion.

These announcements come as the semiconductor giant continues its strategic push into the robotics market, seeking new application scenarios for its AI GPUs beyond traditional data center environments.