Lightwheel-YCB is a high-quality simulation asset benchmark built upon the YCB Benchmarks - Object and Model Set. It features 125 meticulously crafted simulation-ready assets across three categories: Deformable, Articulated, and Rigid. Each asset is available in both MJCF and USD formats, ensuring compatibility with popular simulation frameworks.
Lightwheel-YCB aims to provide the embodied AI community with an easily accessible simulation asset benchmark, enabling researchers to:
- Build realistic simulation scenes using top-quality SimReady assets
- Design complex tasks based on accurate physics interactions
- Develop robust simulation frameworks
- Test sim-to-real transfer capabilities (all assets are digital twins of real YCB objects)
- Establish standardized benchmarks for evaluating AI models across simulation and real-world environments
Example: The original YCB contained 106 objects, with 30+ objects either missing scanned models or having low-quality scans. Lightwheel-YCB recreated high-quality SimReady assets for all 106 objects. In addition, Lightwheel-YCB introduces 19 new objects (block cubes) together with their corresponding SimReady assets.
Example: The original YCB scanned models had significant visual defects, such as broken or uneven meshes and missing PBR materials. Lightwheel-YCB comprehensively enhanced them to achieve photo-level realism.
Example: The original YCB scans contained no physical parameters other than dimensions. Lightwheel-YCB supplements each asset with mass, friction, density, friction loss, damping, stiffness, and armature parameters, tailored to the object’s mode of interaction.
4. Dynamic Behaviors – Created deformable and articulated variants based on real object characteristics
Example: Lightwheel-YCB goes beyond rigid bodies by including 2 deformable assets and 16 articulated assets, all exhibiting physically accurate dynamic behaviors during interaction.
5. Optimized Collisions – Manually crafted collision bodies balancing accuracy, computational efficiency, and simulation stability
Example: The original YCB scans used visual meshes directly as collisions. Lightwheel-YCB rebuilt all collisions from scratch, balancing accuracy, computational efficiency, and stability. Notably, Lightwheel-YCB created collisions for the bottle and cap threads, enabling realistic unscrewing behaviors.
Example: All assets have been validated through teleoperation in Isaac Lab and MuJoCo to ensure correct interaction with robots, making them suitable for teleoperation data collection and synthetic data generation.
7. Standardized Formats – Unified USD and MJCF formats for plug-and-play use in Isaac Sim and MuJoCo
Example: The original YCB assets were limited to basic 3D formats (OBJ, PLY, STL). Lightwheel-YCB converts them into fully SimReady formats: USD for Isaac Sim and Isaac Lab, and MJCF for MuJoCo. Among them, 123 assets support both formats, while 2 deformable assets are USD-only.
- Transformed 106 original YCB objects (30+ previously unusable) into 125 SimReady assets
- Bridged the sim-to-real gap through enhanced visual fidelity and accurate physics modeling
- Enabled seamless integration across major simulation platforms
👉 Download Lightwheel-YCB Assets
Visit our website to download the complete collection of 125 simulation-ready assets in both USD and MJCF formats.
| Total Assets | Categories | Formats | Base Dataset |
|---|---|---|---|
| 125 | Deformable / Articulated / Rigid | USD (Isaac Sim) & MJCF (MuJoCo) | YCB Benchmarks |
This work is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License.
For questions or suggestions, please open an issue on GitHub or contact us:
Note: Star ⭐ this repository to stay updated with the latest improvements!