Dexterity-BEV
收藏资源简介:
Dexterity-BEV数据集是一个真实世界灵巧双手操作数据集,专为通用机器人策略学习设计,旨在对齐3D世界与动作表示。该数据集在Agilex双手机器人平台上采集,包含两个具体操作任务:1)折叠邮件盒任务,约有1623个演示轨迹,数据量约159.8 GB;2)折叠布料任务,约有377个演示轨迹,数据量约44.4 GB。数据以文件目录形式提供,并附带记录演示列表的文本文件(如demos_list-1623.txt)。该数据集服务于机器人学习研究,特别是双手灵巧操作和策略泛化等任务,采用Apache-2.0许可证发布。
The Dexterity-BEV dataset is a real-world dexterous bimanual manipulation dataset designed for general robot policy learning, aiming to align 3D world with action representations. It is collected on the Agilex bimanual robot platform and includes two specific manipulation tasks: 1) the Fold Mailer Box task, with approximately 1623 demonstration trajectories and a data volume of about 159.8 GB; and 2) the Fold Cloth task, with approximately 377 demonstration trajectories and a data volume of about 44.4 GB. The data is provided in file directory format, accompanied by text files recording demonstration lists (e.g., demos_list-1623.txt). This dataset serves robot learning research, particularly for tasks such as bimanual dexterous manipulation and policy generalization, and is released under the Apache-2.0 license.
数据集概述:Dexterity-BEV
基本信息
- 许可证:Apache-2.0
- 数据集用途:用于真实世界的灵巧双臂机器人操作实验,旨在对齐3D世界与动作,支持通用机器人策略学习。
数据集内容
该数据集包含两大任务的数据,均基于Agilex双臂机器人平台采集:
1. 任务:折叠邮件盒(Fold Mailer Box)
- 数据集名称:Agilex-FoldMailerBox
- 数据大小:约159.8 GB
- 样本列表文件:demos_list-1623.txt
2. 任务:折叠布料(Fold Cloth)
- 数据集名称:Agilex-FoldCloth
- 数据大小:约44.4 GB
- 样本列表文件:demos_list-377.txt
引用信息
若在研究中使用该数据集,请引用以下论文:
- 论文标题:Dexterity-BEV: Aligning 3D World and Actions for Generalizable Robot Policies Learning
- arXiv编号:2606.02274
- 论文链接:https://arxiv.org/abs/2606.02274
- 项目页面:https://hnuzhy.github.io/projects/Dex-BEV/
- 论文PDF:https://hnuzhy.github.io/projects/Dex-BEV/Dex-BEV.pdf
- 技术报告:https://dexforce.com/technical-report/#/Dexterity-BEV
- 博客文章:https://mp.weixin.qq.com/s/uBE0DO2VZzBjaHnMeQKw0A
相关链接
- 数据集页面:https://huggingface.co/datasets/HoyerChou/Dexterity-BEV
- 数据集子文件夹:





