自动生成的抓取数据集

本数据集由索邦大学、CNRS和ISIR的研究人员创建，包含超过7000条通过质量多样性（QD）方法生成的抓取轨迹，涵盖三种不同的机械臂和夹具。数据集旨在研究自动生成抓取的模拟到现实转移能力，通过分析域随机化（DR）质量标准与转移可性之间的相关性，解决抓取任务中的现实差距问题。该数据集适用于机器人抓取领域的研究，特别是针对模拟与现实环境之间的差异，以及如何通过优化抓取策略提高机器人在现实世界中的抓取成功率。

This dataset was developed by researchers from Sorbonne University, CNRS, and ISIR. It contains over 7,000 grasp trajectories generated via the Quality Diversity (QD) method, covering three distinct robotic arms and grippers. This dataset is designed to study the sim-to-real transfer capability of automatically generated grasps, addressing the reality gap in grasping tasks by analyzing the correlation between Domain Randomization (DR) quality metrics and transferability. It is applicable to research in the field of robotic grasping, particularly regarding the discrepancies between simulation and real-world environments, and how to improve robots' grasp success rates in the physical world by optimizing grasping strategies.