What problem does it solve?
Building robot and environment assets for Isaac Sim often results in monolithic USD files that are slow to load, hard to edit, and wasteful of VRAM during reinforcement learning. This Skill explains NVIDIA's recommended composition pattern so assets load faster, remain hand-editable, and support headless RL training.
Core Features & Use Cases
- Layered Asset Structure: Splits a robot into a binary geometries.usdc crate plus USDA layers (base, instances, materials, physics, physx, mujoco, robot) composed through an interface.usda entry point.
- Headless RL Optimization: Uses USD payload lazy loading to skip appearance payloads (materials, textures) during RL training, reducing startup time and VRAM usage.
- Asset Transformer Pipeline: Applies GeometriesRoutingRule, MaterialsRoutingRule, SchemaRoutingRule, and InterfaceConnectionRule to convert URDF/MJCF imports into this layout.
- Use Case: When physics edits in a USDA file do not take effect, use the composition arc precedence rules and the property-stack debugging snippet to find which layer overrides your opinion.
Quick Start
Restructure my imported URDF robot into the Isaac Sim layered USD layout with a binary geometry crate and separate physics and materials USDA layers.
