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^†

Shanghai Jiao Tong University, ² Style3D

Style3D, ³

Nanjing University
AAAI 2026 Accepted
^*Indicates Equal Contribution
^†Corresponding Author

chn.h.w.k@sjtu.edu.cn, tangxincheng@sjtu.edu.cn, ryang2@sjtu.edu.cn

Download PDF Code arXiv(TODO)

Robotic Manipulation of Garments and Fabrics with Diverse Materials in RGBench.

Introduction Video

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.

Overview

Overview of RGBench framework, which integrates a diverse garment dataset, dual-arm robotic setups, and the GarmentDynamics simulation system, covering three core tasks to bridge real-world and simulated garment interaction research.

Dataset

RGBench garment dataset

Results

Simulator efficiency comparison

Simulator efficiency comparison (left) Average Initial Time; (right) Average Step Time
3X faster than Isaac Sim, reducing 90% initialization time

Benchmark in cloth simulation accuracy

It reduces the sim-to-real gap by over 20% on average and by as much as 77% for topologically complex garments.

Qualitative results of different actions for a T-shirt

For grasp and fold tasks, error reduction up to 35% and 58%. For dynamic fling task, improving metrics by over 20%.

Results of different garment types in grasp task

GarmentDynamics achieves the lowest sim-to-real gap across different garment types, with error reductions of over 37% compared to baselines. For complex garments like Cakeskirt, it reduces errors by up to 44% in pseudo mode and 77% in robot mode.

-->