Evolved Grasping Analysis Dataset (EGAD)

Evolved Grasping Analysis Dataset (EGAD) 是由澳大利亚机器人视觉中心创建的一个包含2331个生成对象的数据集，旨在训练和评估机器人视觉抓取检测算法。该数据集的对象在几何上具有多样性，涵盖了从简单到复杂形状、从易到难抓取的范围。EGAD不仅提供了大量的数据，还包括一个由49个多样化的3D可打印评估对象组成的子集，以促进机器人抓取系统在不同复杂度和难度下的可重复测试。数据集的创建过程采用了进化算法，确保了对象在形状复杂度和抓取难度上的多样性。EGAD的应用领域主要集中在机器人抓取技术，旨在通过提供一个大型且多样化的数据集来解决现有抓取算法在面对未知条件时的泛化能力问题。

Evolved Grasping Analysis Dataset (EGAD) is a dataset consisting of 2331 generated objects, developed by the Australian Centre for Robotic Vision. It is designed for training and evaluating robotic vision grasp detection algorithms. The objects included in EGAD exhibit geometric diversity, spanning a range from simple to complex shapes and varying levels of grasping difficulty from easy to challenging. In addition to the large-scale dataset, EGAD also provides a subset of 49 diverse 3D-printable evaluation objects, which facilitates reproducible testing of robotic grasping systems across different complexity and difficulty levels. The dataset was constructed using evolutionary algorithms, which guarantee the diversity of object shape complexity and grasping difficulty. The primary application domain of EGAD focuses on robotic grasping technology, with the goal of addressing the generalization capability issue of existing grasping algorithms when faced with unknown conditions by providing a large-scale and highly diverse dataset.