AIAny
Icon for item

SimReady GLB

Packages 2,595 NVIDIA SimReady 3D assets as optimized, browser-ready GLB files, retaining rigid-body and joint data for interactive physics and robotics simulation. It reduces Blender exports from 228.1 GB to 3.2 GB without mesh simplification.

Introduction

A simulation asset is only useful to an AI or robotics workflow if it can load quickly and still behave like a physical object. This collection tackles that practical bottleneck: it makes a large set of SimReady props light enough for web-based viewing while preserving much of the physics metadata needed to animate and simulate them.

What Sets It Apart
  • The 2,595 assets shrink from 228.1 GB of Blender exports to about 3.2 GB, making the collection easier to browse and integrate into web-based tools.
  • Geometry is compressed with meshopt and textures are converted to WebP; no mesh simplification is applied, so the triangle reduction comes from removing duplicate and helper geometry rather than reducing model detail.
  • Most files retain rigid-body properties, and 929 include joints. This makes doors, drawers, wheels, and other moving parts useful for interactive simulation and embodied-robotics scenes, rather than merely static visualization.
Who It’s For—and the Trade-Offs

A good fit for developers building browser-based 3D viewers, robotics simulations, or embodied-AI environments that need varied, ready-to-load props with physical metadata. Look elsewhere if you need the original USD and its full simulation settings, signed-distance-field collision for hollow objects, or MDL-specific material effects: this conversion uses convex-hull colliders and glTF PBR materials, and some physics extensions may not be supported by every loader.

Information

More Items

Builds executable professional tasks by mapping required information dependencies and repairing gaps revealed during test runs. It produces 112,686 tasks across 612 occupations for collecting teacher demonstrations and training working agents.

Hugging Face

Tests whether video world models can follow evolving instructions, control camera motion, and preserve scene details across revisits. Its 280 cases cover four interaction patterns; scoring is left to evaluators.

Hugging Face

Provides transcripts, translations, and multi-task text labels for speech across 48 languages and about 463,000 hours. The release contains manifests only; audio must be obtained from the original data providers.