glTF Assets in Nvidia Omniverse USD Pipelines

This article provides an overview of how Nvidia Omniverse integrates glTF (GL Transmission Format) assets into Universal Scene Description (USD) pipelines. It details the mechanisms of asset conversion, material translation, referencing workflows, and the non-destructive composition methods that allow lightweight transmission models to function seamlessly within complex, multi-user simulation environments.

The Roles of glTF and USD

To understand how Omniverse handles glTF files, it is essential to distinguish between the two formats:

  • glTF: Often referred to as the "JPEG of 3D," glTF is an open, compact delivery format optimized for efficient transmission and rapid runtime loading on the web and real-time engines.
  • USD: Developed by Pixar, Universal Scene Description is an extensible, non-destructive interchange and scene-composition architecture designed for large-scale production pipelines and collaborative authoring.

Omniverse uses OpenUSD as its native foundation. Rather than replacing USD, glTF serves as an ingestion format, bringing lightweight 3D assets directly into industrial and creative USD-based workflows.

Omniverse glTF Asset Ingestion and Conversion

Omniverse incorporates a dedicated glTF converter service and runtime importer. When a .gltf or binary .glb file is loaded into an Omniverse stage, the platform automatically parses the data and maps it into corresponding USD schemas:

  • Geometry and Hierarchy: Nodes, meshes, and transform hierarchies defined in the glTF file are mapped directly to UsdGeomXform and UsdGeomMesh primitives.
  • Skeletal Animation and Morphs: Joint structures, skin weights, and blend shapes are mapped into the UsdSkel schema, preserving rigging and performance capture data.
  • Cameras and Lights: Supported camera parameters and light extensions (such as KHR_lights_punctual) translate into their respective UsdLux and UsdGeomCamera equivalents.

PBR Material Translation to MDL and UsdPreviewSurface

The standard glTF specification relies on a physically based rendering (PBR) metallic-roughness workflow. Omniverse preserves visual parity by converting glTF material definitions through two primary methods:

  1. UsdPreviewSurface: For lightweight previewing, glTF materials are mapped to the baseline UsdShade network using UsdPreviewSurface, ensuring compatibility across the broader USD ecosystem.
  2. NVIDIA Material Definition Language (MDL): For high-fidelity path tracing and real-time simulation within the Omniverse RTX Renderer, the platform translates glTF PBR parameters (including roughness, metallic, normal, occlusion, and emissive maps) into native MDL representations. Extensions such as KHR_materials_clearcoat, KHR_materials_transmission, and KHR_materials_ior are mapped to corresponding MDL attributes to guarantee accurate optical behavior.

Non-Destructive Referencing and Layering

Omniverse treats glTF assets as referenceable components rather than static, one-time imports. Through USD's composition arcs, glTF files can be referenced directly into an existing USD stage:

  • Payloads and References: Users can attach a glTF model using a standard USD sublayer or reference. Omniverse reads the underlying glTF asset and displays it in the viewport without requiring the user to permanently overwrite the source file.
  • USD Overrides: Because USD operates via non-destructive layering, artists can add "overs" (overrides) to the glTF asset. This means materials, physics properties (via UsdPhysics), and transform coordinates can be modified or animated on a distinct USD layer while leaving the source .gltf or .glb untouched.

Pipeline Efficiency and Interoperability

By facilitating direct ingestion of glTF into USD pipelines, Nvidia Omniverse removes the barrier between web-ready asset libraries and enterprise-grade simulations. Assets sourced from web-based 3D repositories, CAD-to-glTF exporters, and mobile scanning applications can be instantly deployed into digital twin environments, robotics training simulations, and high-end visual effects pipelines without extensive manual cleanup or re-authoring.