Nvidia unveils new Cosmos world models, infra for robotics and physical uses

In a significant development for the robotics sector, Nvidia has introduced a new suite of AI models and infrastructure aimed at empowering robotics developers. Among the highlights is Cosmos Reason, a sophisticated 7-billion-parameter vision language model designed for physical AI applications and robotic systems. This latest addition joins the existing Cosmos lineup, including Cosmos Transfer-2, which enhances the speed of synthetic data generation from 3D simulations and spatial control inputs. Nvidia has also released a more streamlined version of Cosmos Transfers, optimized for quicker operations. During the unveiling at the SIGGRAPH conference, Nvidia emphasized that these models are intended to facilitate the creation of synthetic datasets—including text, images, and videos—that are crucial for training robots and AI agents. According to the company, Cosmos Reason enables machines to perform reasoning tasks by comprehending memory and physics, which helps plan the next steps for an embodied agent. Beyond AI models, Nvidia showcased new neural reconstruction libraries that feature advanced rendering techniques, allowing developers to create 3D simulations using sensor data. This rendering capability is set to be integrated into CARLA, a well-regarded open-source simulator. The company also announced updates to the Omniverse software development kit and introduced new servers designed specifically for robotics workflows. The Nvidia RTX Pro Blackwell Servers provide a unified architecture for robotic development, while the Nvidia DGX Cloud offers a comprehensive cloud-based management solution. These advancements signify Nvidia's commitment to expanding its footprint in robotics, looking to leverage its AI GPUs for innovative applications beyond traditional AI data centers.