Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Real Garment Benchmark (RGBench): A Comprehensive Benchmark for Robotic Garment Manipulation featuring a High-Fidelity Scalable Simulator

Wenkang Hu, Xincheng Tang, Yanzhi E, Yitong Li, Zhengjie Shu, Wei Li, Huamin Wang, Ruigang Yang

TL;DR

This work tackles the sim-to-real gap in robotic garment manipulation by introducing Real-Garment Benchmark (RGBench), a comprehensive framework that pairs a diverse, physically-grounded garment dataset with GarmentDynamics, a high-fidelity, GPU-accelerated continuum-FEM cloth simulator. The framework provides a rigorous ground-truth protocol and evaluation tasks (Grasp, Fling, Fold) to quantify fidelity using bidirectional metrics $CD$ and $HD$ across real and simulated domains. GarmentDynamics achieves state-of-the-art accuracy and robustness, reducing the sim-to-real gap by over 20% on average and up to 77% for topologically complex garments, while delivering 3x faster performance than the closest competitor. The work enables scalable, realistic policy learning for garment manipulation and establishes RGBench as a platform for future advances in deformable-object robotics and fabric realism.

Abstract

While there has been significant progress to use simulated data to learn robotic manipulation of rigid objects, applying its success to deformable objects has been hindered by the lack of both deformable object models and realistic non-rigid body simulators. In this paper, we present Real Garment Benchmark (RGBench), a comprehensive benchmark for robotic manipulation of garments. It features a diverse set of over 6000 garment mesh models, a new high-performance simulator, and a comprehensive protocol to evaluate garment simulation quality with carefully measured real garment dynamics. Our experiments demonstrate that our simulator outperforms currently available cloth simulators by a large margin, reducing simulation error by 20% while maintaining a speed of 3 times faster. We will publicly release RGBench to accelerate future research in robotic garment manipulation. Website: https://rgbench.github.io/

Real Garment Benchmark (RGBench): A Comprehensive Benchmark for Robotic Garment Manipulation featuring a High-Fidelity Scalable Simulator

TL;DR

This work tackles the sim-to-real gap in robotic garment manipulation by introducing Real-Garment Benchmark (RGBench), a comprehensive framework that pairs a diverse, physically-grounded garment dataset with GarmentDynamics, a high-fidelity, GPU-accelerated continuum-FEM cloth simulator. The framework provides a rigorous ground-truth protocol and evaluation tasks (Grasp, Fling, Fold) to quantify fidelity using bidirectional metrics and across real and simulated domains. GarmentDynamics achieves state-of-the-art accuracy and robustness, reducing the sim-to-real gap by over 20% on average and up to 77% for topologically complex garments, while delivering 3x faster performance than the closest competitor. The work enables scalable, realistic policy learning for garment manipulation and establishes RGBench as a platform for future advances in deformable-object robotics and fabric realism.

Abstract

While there has been significant progress to use simulated data to learn robotic manipulation of rigid objects, applying its success to deformable objects has been hindered by the lack of both deformable object models and realistic non-rigid body simulators. In this paper, we present Real Garment Benchmark (RGBench), a comprehensive benchmark for robotic manipulation of garments. It features a diverse set of over 6000 garment mesh models, a new high-performance simulator, and a comprehensive protocol to evaluate garment simulation quality with carefully measured real garment dynamics. Our experiments demonstrate that our simulator outperforms currently available cloth simulators by a large margin, reducing simulation error by 20% while maintaining a speed of 3 times faster. We will publicly release RGBench to accelerate future research in robotic garment manipulation. Website: https://rgbench.github.io/

Paper Structure

This paper contains 62 sections, 8 equations, 14 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deformable Object Manipulation Benchmark
  4. Physics-Based Cloth Simulation
  5. The RGBench Framework
  6. RGBench Dataset
  7. Diverse Garment Assets
  8. Ground Truth Acquisition
  9. RGBench Evaluation
  10. Eval Tasks: Robotic Manipulation
  11. Eval Preprocess
  12. Eval Metrics
  13. Simulator
  14. GarmentDynamics Design
  15. Core Components
  16. ...and 47 more sections

Figures (14)

  • Figure 1: Robotic Manipulation of Garments and Fabrics with Diverse Materials in RGBench
  • Figure 2: Overview of RGBench framework
  • Figure 3: RGBench garment dataset
  • Figure 4: Simulator efficiency comparison (left) Average Initial Time; (right) Average Step Time;
  • Figure 5: Real-to-Sim Comparison for Folding (left column) Real-world robotic folding sequence. (middle column) Point cloud comparison: real-world (white), our simulator (blue), best of other simulators (Grey). (right column) Garment folding state in all simulators.
  • ...and 9 more figures