CEDex
收藏资源简介:
CEDex是一个大规模的跨具身灵巧抓取数据集,通过将机器人运动学模型与生成的人类接触表示对齐,连接人类抓取运动学和机器人运动学。该数据集包含500K个对象,涵盖四种夹爪类型,总计2000万次抓取动作。目前已发布真实世界对象集的抓取数据,包括来自ContactDB和YCB的对象,并将很快发布更大规模的合成对象抓取数据。
CEDex is a large-scale cross-embodied dexterous grasping dataset that bridges human grasping kinematics and robotic kinematics by aligning robotic kinematic models with generated human contact representations. This dataset encompasses 500,000 objects across four types of grippers, with a total of 20 million grasping trials. Currently, grasping data for real-world object sets has been released, including objects sourced from ContactDB and YCB, and larger-scale synthetic object grasping data will be released soon.
CEDex数据集概述
数据集基本信息
- 数据集名称: CEDex
- 研究论文: CEDex: Cross-Embodiment Dexterous Grasp Generation at Scale from Human-like Contact Representations
- 论文链接: https://arxiv.org/abs/2509.24661
- 项目页面: https://georgewuzy.github.io/cedex-website/
数据集规模
- 对象数量: 500K个对象
- 抓取总数: 20M个抓取
- 夹爪类型: 四种夹爪类型
数据集内容
已发布数据
- 真实世界对象抓取数据
- 来源对象: ContactDB (https://contactdb.cc.gatech.edu/)
- 来源对象: YCB (https://www.ycbbenchmarks.com/)
即将发布数据
- 合成对象抓取数据
- 来源对象: Objaverse (https://objaverse.allenai.org/)
数据格式
- 文件格式:
.pt文件 - 存储路径:
cedex/{robot_name}.pt - 数据字段:
q: 抓取配置object_name: 对象名称robot_name: 机器人名称
数据获取
- 下载地址: https://drive.google.com/file/d/1xmBV66SO-TjkREYTujh08QkucCWHYLxx/view?usp=sharing
- 存储位置: 解压至
data/目录
使用说明
数据可视化
bash python vis_cedex.py --input_file cedex/allegro.pt
抓取验证
bash python eval_grasp.py --input_file cedex/allegro.pt --object_name ycb+055_baseball
相关依赖
- 基础环境: Python 3.8, Conda环境
- 深度学习框架: PyTorch 2.4.0
- 验证环境: Isaac Gym
引用信息
bibtex @article{wu2025cedex, title={CEDex: Cross-Embodiment Dexterous Grasp Generation at Scale from Human-like Contact Representations}, author={Wu, Zhiyuan and Potamias, Rolandos Alexandros and Zhang, Xuyang and Zhang, Zhongqun and Deng, Jiankang and Luo, Shan}, journal={arXiv preprint arXiv:2509.24661}, year={2025} }
联系方式
- 联系邮箱: zhiyuan.1.wu@kcl.ac.uk




