From NVIDIA: 2025-01-22 09:00:11

The next frontier of AI is physical AI, which can power autonomous machines like robots and self-driving cars by understanding instructions, perceiving, interacting, and performing complex actions in the real world. Global leaders are using NVIDIA Omniverse and OpenUSD to accelerate AI development and create true-to-reality virtual worlds known as digital twins.

At CES, NVIDIA announced generative AI models and blueprints, including NVIDIA Cosmos, to accelerate physical AI development. Cosmos provides world foundation models for generating photoreal, physics-based synthetic data to train AI for robotics and autonomous vehicles. When paired with Omniverse, Cosmos can exponentially increase training data for physical AI systems like autonomous vehicles and robots.

Humanoid robots benefit from the NVIDIA Isaac GR00T Blueprint, which generates synthetic motion datasets to train robots using imitation learning. Autonomous vehicle developers can use Cosmos to generate synthetic driving scenarios and improve AI model training efficiency. Mega, an Omniverse Blueprint, enables physical AI and robot fleet optimization in industrial settings before deployment, enhancing training efficiency and robustness.

To learn more about Cosmos and physical AI workflows, watch NVIDIA CEO Jensen Huang’s CES keynote and access Cosmos WFMs under an open model license. Join upcoming livestreams and explore self-paced learning resources for OpenUSD to optimize workflows for 3D developers and practitioners. Meet experts at NVIDIA GTC, subscribe to NVIDIA news, and follow NVIDIA Omniverse on social media for updates.



Read more at NVIDIA: OpenUSD Workflows Advance Physical AI for Robotics, Autonomous Vehicles