ClothDojo: Learned Data Generation and Benchmarking for Bimanual Garment Flattening and Folding
Sim platform where a fine-tuned VLA generates nearly 8,000 successful garment trajectories, re-rendered for RGB policies. Those beat policies trained on the original human data, but unseen garments stay hard.仿真平台：用微调后的 VLA 生成近 8000 条成功的衣物操作轨迹，并重新渲染供 RGB 策略训练，效果优于直接用人类数据训练，但泛化到未见衣物仍是难题。
BibTeX
@misc{huang2026clothdojo,
title = {{ClothDojo: Learned Data Generation and Benchmarking for Bimanual Garment Flattening and Folding}},
author = {Zilong Huang and Zipeng Ye and Caicheng Wang and Yuchen Xie and Jiaxiang Luo},
year = {2026},
eprint = {2611.01190},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2611.01190},
}Datasets & Benchmarks数据集与基准Manipulation & Dexterous Hands操作与灵巧手Sim-to-Real & Simulators仿真与 Sim-to-RealRobot arm机械臂benchmarkgarmentdeformabledata-generationsim