五指灵巧手抓取数据集

五指灵巧手是机器人走向类人化的重要形态和标志之一，赋予灵巧手机器人抓取多样物体的能力可有效促进其走进人类生产生活，并在实际场景中服务人类。我们构建了一个五指灵巧手大规模抓取数据集，并预定义了五种抓取模态从而帮助机器人生成多样化的抓取配置。该数据集包含179个物体，5000个单视角场景，超过100万个单视角点云的抓取标注，可被应用于多种抓取网络的训练。Five-finger dexterous hands are one of the key forms and hallmarks of humanoid robotics. Endowing such robotic hands with the ability to grasp diverse objects can effectively facilitate their integration into human production and daily life, as well as enable them to serve humans in real-world scenarios. We have constructed a large-scale grasping dataset for five-finger dexterous hands, where five grasping modalities are pre-defined to help robots generate diverse grasping configurations. This dataset includes 179 objects, 5000 single-view scenes, and more than 1 million grasping annotations for single-view point clouds, which can be used for training various grasping networks.

搜集汇总

背景与挑战

背景概述

该数据集是一个大规模仿人手机器人抓取数据集，包含179个物体、5,000个单视角场景和超过100万个抓取标注，基于物体点云数据，旨在帮助机器人生成多样化的抓取配置，适用于抓取网络的训练。数据集由中国科学院脑科学与智能技术卓越创新中心发布，采用CC BY 4.0许可证，存储容量为1.91 GB，专注于仿人手、灵巧抓取和多分类抓取任务。