BREAKING
NVIDIA agents turn USD spec into runtimes
Spec-first, agent-assisted workflow
1USD Core Spec
2Port codebase
3AI agents generate
4Conformant runtime
Built for physical AI
Old approachLegacy
Port or trim OpenUSD
Laborious and error-prone
One-size-fits-all stack
nanousd-labsNew
Generate from spec
Custom runtimes
Tuned for physical AI
Dual Apache 2.0 and CC-BY-4.0
nanousd: portable C-ABI core
1nanousd C-ABI
2nanousd-python
3nanousdview
4Vulkan/GL/Metal
A proof of concept, not a product
AI NEWS BLITZ
NVIDIA's Omniverse Labs shows AI agents building 3D scene runtimes straight from spec.