Menu
GitHub ↗

USD VS RGD

Scene composition and semantic embodiment should stay distinct.

OpenUSD is designed for scalable, composable scene description. RGD focuses on what an embodied AI or runtime needs to know about a specific robot body, its provenance, capabilities, constraints and lifecycle.

USD / OpenUSD+RGD
CITATION RECORDCanonical source ↗
MATURITYREFERENCE COMPARISON
LAST VERIFIED22 Sep 2026
SOURCE VERSIONOpenUSD / UsdPhysics documentation + OpenRGD Import Guide
EVIDENCE BOUNDARYComplementary-layer comparison; OpenRGD does not replace OpenUSD composition or UsdPhysics.

RESPONSIBILITY MATRIX

Different centers of gravity.

DIMENSIONOPENUSDRGD
Primary responsibility

Composable scene description across complex assets and layers.

Semantic embodiment and robot grounding context.

Core representation

Hierarchical scenegraph with composition arcs and authored opinions.

Modular semantic graph/profile with domain authority.

Scene composition

First-class references, variants, payloads and layered composition.

Not a replacement for USD composition.

Physics

UsdPhysics schemas for bodies, colliders, joints and simulation scenes.

Can carry physical evidence and constraints without being a physics engine.

Robot-specific identity

Possible through authored scene data, but not the central standard purpose.

Canonical robot/profile identity is explicit.

Capabilities / governance

Not intrinsic robot capability/governance semantics.

Agency and Volition domains make those responsibilities explicit.

Provenance

USD layers preserve authored scene opinions and composition provenance.

Adds evidence provenance specific to embodiment and enrichment boundaries.

Runtime authority

Scene description is not hardware permission.

Semantic context is also not hardware permission.

OVERLAP

Where both may describe the same robot facts.

  • Physical joint/body facts may exist in both representations.
  • Units and source-authored physical properties can inform RGD Foundation.
  • Robot and environment assets can be linked to the same real physical system.
  • Simulation pipelines may consume scene data while AI pipelines consume RGD context.
  • Both benefit from deterministic, inspectable source boundaries.

WHAT RGD ADDS

Context beyond the source description.

  • Robot-centric semantic identity across non-scene domains.
  • Explicit imported-evidence versus enriched-knowledge distinction.
  • Operational safety and runtime constraint context.
  • Capability and action semantics for AI/planning systems.
  • Lifecycle and continuity semantics.
  • Canonical cognitive-to-physical execution boundary.

RGD DOES NOT REPLACE

Keep ownership explicit.

  • USD references, payloads, variants and layer composition.
  • Full OpenUSD SDK semantics.
  • UsdPhysics simulation schema implementation.
  • Rendering, asset pipelines or DCC scene workflows.
  • A simulator, middleware layer or runtime.

USE BOTH WHEN

Complementary layers are stronger than forced convergence.

  • The robot already exists as a USD/OpenUSD asset in simulation or digital-twin workflows.
  • The same robot needs machine-readable semantic context for AI systems.
  • You want scene composition to remain in USD while embodiment semantics remain inspectable elsewhere.
  • A full OpenUSD adapter can extract authoritative body evidence before semantic enrichment.
  • Simulation state and semantic profile need independent versioning and ownership.

OPENRGD FLOW

Source evidence can become semantic context without erasing provenance.

01OpenUSD scene→
02Full/narrow source adapter→
03Foundation evidence→
04RGD semantic enrichment→
05Validated context→
06Simulator / runtime

PRIMARY SOURCES

01OpenUSD — Introduction↗02OpenUSD — USD FAQ↗03OpenUSD — UsdPhysics↗04OpenRGD — Import Guide↗
← All description formatsRGD reference →