NVIDIA Unveils Safety Frameworks for Physical AI with OpenUSD Core Specification 1.0

Details

• NVIDIA introduced new safety frameworks and technologies for developing physical AI systems, built on the newly released OpenUSD Core Specification 1.0 from the Alliance for OpenUSD.
• The framework integrates real-world data, advanced simulation, and robust AI models through OpenUSD, allowing standardized and interoperable development pipelines for robotaxis and autonomous vehicles.
• Key technologies include NVIDIA Omniverse for digital twins, Isaac Sim for robotic simulation, Cosmos world foundation models for synthetic data, and the Halos safety validation and certification framework.
• Supporting resources include the open-source Learn OpenUSD curriculum on GitHub and the Omniverse NuRec Fixer tool, helping teams develop simulation-ready assets and scale robot training efficiently.
• Early partners include Bosch, Nuro, Wayve, and Onsemi, with Onsemi being the first to earn accreditation from the NVIDIA Halos AI Systems Inspection Lab for sensor and AV stack validation.

Impact

NVIDIA’s announcement moves the industry closer to safer and more standardized physical AI solutions as autonomous systems expand beyond test environments. OpenUSD democratizes access across developers, while the Halos framework brings scalable validation and certification. This strategy reduces the need for extensive real-world testing, streamlining regulatory approval and fostering global alignment in AV safety.