language:
- en
license: mit
pretty_name: GAUGE
tags:
- robotics
- embodied-ai
- physics
- simulation
- world-model
- motion-capture
- sim-to-real
- deformable-objects
- benchmark
task_categories:
- other
GAUGE Dataset
GAUGE: A Measurement-Grounded Benchmark for Physical Fidelity in Simulation Engines and Video World Models
GAUGE is a real-world-grounded benchmark for evaluating the physical fidelity of numerical physics engines and generative video world models. The benchmark contains 22 controlled task families spanning rigid bodies, flexible cables, textiles, and volumetric deformable objects.
Task Overview
GAUGE contains 22 controlled task families spanning rigid bodies, flexible cables, textiles, and volumetric deformable objects.
| Category | Task | Video | Description | Evaluation | Subtasks | Material |
|---|---|---|---|---|---|---|
| Rigid Body | Slope Contact |  |
Fall of a trirectangular tetrahedron | Slope collisions | 1 | Wood, Plastic |
| Rigid Body | Nonsmooth Contact | Fall of wedges and pyramids | Codimensional collisions | 3 | Wood, Plastic | |
| Rigid Body | Slope Slider | Slide of a cube on a slope | Static and kinetic friction | 3 | Wood, Plastic, Metal | |
| Rigid Body | Turntable | Slide of a cube on a turntable | Static and kinetic friction in a non-inertial frame | 3 | Wood, Plastic, Metal | |
| Rigid Body | Bouncing Ball | Fall and bouncing of a ball | Rapid impact | 1 | Rubber | |
| Rigid Body | Newton's Cradle | Collision of closely fitted balls | Momentum transfer | 1 | Metal | |
| Rigid Body | Wall Breaking | Collision of a wrecking ball with a wall | Large-scale dense collisions | 1 | Wood | |
| Rigid Body | Pendulum | Oscillation of a ball | Periodic motion | 1 | Metal | |
| Flexible Cable | Rope Winding | Winding of a rope | Self-collision and stretch modulus | 1 | Rubber | |
| Textile | Textile Stretching | Tensile deformation of a textile | Stretch modulus | 1 | R, S, U, O, L, N | |
| Textile | Textile Bending | Natural sagging of a textile | Bending modulus | 1 | R, S, U, O, L, N | |
| Textile | Textile Flinging | Fluttering of a textile | Motion under locally high acceleration | 1 | R, S, U, O, L, N | |
| Textile | Funnel | Passage of textiles through a hole | Collision, friction | 1 | R, S, U, O, L, N | |
| Textile | Rotating Ball | Motion of a textile driven by a rotating ball | Static and kinetic friction | 1 | R, S, U | |
| Textile | Tablecloth Pulling | Pulling a textile from beneath rigid objects | Static and kinetic friction | 1 | Wood, Plastic, Metal | |
| Volumetric Deformable Body | Foam Stretching | Stretching of an elastic cuboid | Stretching modulus | 1 | Soft, Hard | |
| Volumetric Deformable Body | Foam Compressing | Compression of an elastic cuboid | Compression modulus | 1 | Soft, Hard | |
| Volumetric Deformable Body | Foam Shearing | Shearing of an elastic cuboid | Shear modulus | 1 | Soft, Hard | |
| Volumetric Deformable Body | Foam Twisting | Twisting of an elastic cuboid | Twisting modulus | 1 | Soft, Hard | |
| Volumetric Deformable Body | Foam Bending | Bending of an elastic cuboid | Bending modulus | 1 | Soft, Hard | |
| Volumetric Deformable Body | Stick-Stack | Sliding of an elastic rod on a plane | Static and kinetic friction | 1 | Soft, Hard | |
| Volumetric Deformable Body | Cantilever Beam | Free hanging of an elastic cantilever beam | Coupling with a large stiffness ratio | 1 | Soft, Hard |
Textile material abbreviations: R = Rayon, S = Satin, U = Uniform cloth, O = Oxford fabric, L = Synthetic leather, N = Nylon taslan.
Physical Regimes
| Physical Regime | Number of Task Families | Representative Properties |
|---|---|---|
| Rigid Body | 8 | Collision, friction, momentum transfer, impact, periodic motion |
| Flexible Cable | 1 | Self-collision, stretching |
| Textile | 6 | Stretching, bending, friction, high-acceleration dynamics |
| Volumetric Deformable Body | 7 | Stretching, compression, shear, twisting, bending, friction |
Dataset Contents
GAUGE provides real-world experimental measurements and task-associated physical metadata for standardized physics-fidelity evaluation. The released dataset is organized around:
- real-world motion-capture trajectories and measurements;
- simulation-ready geometry and scene assets;
- calibrated physical-property metadata;
- standardized first-frame and text-prompt inputs for video world models;
- evaluation scripts and task-specific physical metrics.
Citation
If you find GAUGE useful in your research, please consider citing:
@article{Wang2026,
title={GAUGE: A Measurement-Grounded Benchmark for Physical Fidelity in Simulation Engines and Video World Models},
author={Wang, Shuai and Feng, Yaxin and Jiang, Xuekun and Tian, Shihan and others},
year={2026},
copyright = {arXiv.org perpetual, non-exclusive license},
doi = {10.48550/ARXIV.2608.05948},
keywords = {Artificial Intelligence (cs.AI), Computer Vision and Pattern Recognition (cs.CV), Robotics (cs.RO), FOS: Computer and information sciences},
publisher = {arXiv},
}
Please replace this entry with the final BibTeX from the released paper if needed.