NVIDIA Demonstrates AI Agents Preparing 3D Scenes for Simulation
NVIDIA’s technical blog outlines an agentic workflow that uses Omniverse Libraries, Codex (or Claude), and NemoClaw subagents to convert Blender scenes into simulation‑ready USD for Isaac Sim or Isaac Lab.
Original source published: September 16, 2026
On September 16, 2026 NVIDIA published a guide showing how AI agents can automate the preparation of Blender scenes for robotics simulation. The workflow relies on a coordination agent such as Codex or Claude, while specialized Hermes subagents deployed through NVIDIA NemoClaw use Omniverse tools—ovphysx for physics authoring, ovrtx for visual preflight rendering, and SimReady validation—to add semantic labels, collision meshes, sensor configurations, and simulation‑aware materials. OpenUSD serves as the shared contract layer, preserving hierarchy and metadata so agents can iteratively enrich the scene without flattening the original creative work.
The process begins with a Blender Model Context Protocol server that inventories the scene, followed by USD authoring, labeling of prims (e.g., shelf, bin, grabbable_object), physics configuration, rendering of review views, and final SimReady validation before handoff to Isaac Sim or Isaac Lab. The post points readers to the Omniverse Labs GitHub repo, SimReady Foundation, and NemoClaw documentation for further exploration.