ICRA 2024 - Cloth Competition

9 January, 2024

The ICRA 2024 Cloth Competition, held as part of the 9th RGMC, focused on grasp point localization for robotic cloth unfolding. Eleven teams competed with their perception systems on a standardized dual-arm robot setup, with evaluations conducted live at ICRA 2024 in Yokohama, Japan.

This competition resulted in a comprehensive dataset of robotic cloth unfolding attempts, capturing valuable real-world data across diverse garments. The dataset is a key resource for developing and evaluating new algorithms for cloth manipulation.

A paper describing the competition and the dataset was submitted to the International Journal of Robotics Research (IJRR) for review.

Dual arm setup — Left: Dual UR5e setup we will provide at ICRA 2024. Right: Grasp point localisation on a hanging shirt.

Important links

📦 Dataset on Zenodo: https://zenodo.org/records/14621179
👩‍💻 Code: https://github.com/Victorlouisdg/cloth-competition
📝 Rules: https://cloud.ilabt.imec.be/index.php/s/AfiiKxyCQ3nESBs
📦 Dataset on our IDLab Cloud (old host): https://cloud.ilabt.imec.be/index.php/s/Sy945rbamg8JMgR

Highlights

Competition Ranking 🏆

Congratulations to all 11 teams who successfully completed their 16 live evaluation trials in the ICRA 2024 Cloth Competition! We’re proud of the hard work and dedication each team put into this challenge. The final rankings and scores are listed below.

Rank	Team Name	Score (Average Coverage)
🥇 1st	AIR-JNU	0.60
🥈 2nd	Team Ljubljana	0.57
🥉 3rd	Ewha Glab	0.55
4th	SCUT-ROBOT	0.53
5th	Team Greater Bay	0.53
6th	Samsung Research China - Beijing	0.48
7th	Shibata Lab	0.46
8th	AI&ROBOT LAB	0.45
9th	UOS-Robotics	0.39
10th	AIS Shinshu	0.37
11th	3C1S	0.35

Open Source Code from Participants

The following teams have already generously shared their code:

Team Name	Code Repository
EWHA GLab	https://github.com/minseo10/cloth_unfolding
Team Ljubljana	https://github.com/vicoslab/CeDiRNet

The Challenge

Regrasping garments in the air is a popular cloth unfolding strategy. To get started, it’s common to grasp the lowest point. However, better grasp points are needed to unfold the garment completely, such as the shoulders of a shirt or the waist of shorts. The objective of the competition is to localise good grasp poses in colour and depth images of garments held in the air by one gripper. We will evaluate your predicted grasp by executing it on a dual UR5e setup live at ICRA. The cloth items that must be grasped will be towels, shirts and shorts. We will provide an image dataset of representative cloth items for preparation, but the evaluation will also include unseen garments.

Practicalities

Participants do not have to worry about the initial lifting of the cloth, grasp execution, camera calibration, etc. We will take care of all that with our open-source codebase airo-mono! At the competition, we will send you an image and you just send us a grasp back in a to-be-specified JSON format. Together with the dataset, we will also provide a Github repository with notebooks and tools to load, visualise and explore the data. Don’t hesitate to contact us if you have any questions!

Teams

Lab/Team Name	Members
UOS-Robotics	Uihun Sagong, JungHyun Choi, JeongHyun Park, Dongwoo Lee, Yeongmin Kim
Ewha Glab	Hyojeong Yu, Minseo Kwon
AI&ROBOT LAB	Chongkun Xia, Kai Mo, Yanzhao Yu, Qihao Lin, Binqiang Ma
3C1S	Carlos Mateo-Agullo
Shibata Lab	Kakeru Yamasaki, Tomohiro Shibata, Krati Saxena, Takumi Kajiwara, Yuki Nakadera
AIS Shinshu	Solvi Arnold, Kimitoshi Yamazaki, Daisuke Tanaka, Keisuke Onda, Akihisa Ishikawa, Yusuke Kuribayashi, Naoki Hiratsuka
Team Greater Bay	Fang Wan, Linhan Yang, Haoran Sun, Ning Guo, Chaoyang Song, Jia Pan, Lei Yang, Zeqing Zhang
Team Ljubljana	Domen Tabernik, Andrej Gams, Peter Nimac, Matej Urbas, Jon Muhovič, Danijel Skočaj, Matija Mavsar
SCUT-ROBOT	Supeng Diao, Yang Cong, Yu Ren, Ronghan Chen, Jiawei Weng, Jiayue Liu
Air-jnu	Giwan Lee, Jiyoung Choi, Jeongil Choi, Geon Kim, Phayuth Yonrith
Samsung Research China – Beijing	Yixiang Jin, Dingzhe Li, Yong A, Jun Shi, Yong Yang