REPLAB Grasping Dataset

REPLAB Grasping Dataset是由加州大学伯克利分校创建的一个用于评估机器人抓取学习算法的数据集。该数据集包含超过92,000次随机抓取尝试的记录，这些尝试是在两个REPLAB平台上并行收集的，每天每个平台约收集2,500次尝试。数据集的创建过程涉及在不同光照条件下使用DBSCAN算法对3D点云进行聚类，以识别对象并随机选择集群进行抓取。该数据集主要应用于机器人抓取任务的评估，旨在通过标准化硬件平台和数据集，推动机器人学习领域的研究进展，降低研究门槛，并促进算法的可重复性和可比较性。

The REPLAB Grasping Dataset is a dataset developed by the University of California, Berkeley for evaluating robotic grasping learning algorithms. This dataset contains records of over 92,000 random grasping attempts, which were collected in parallel on two REPLAB platforms, with approximately 2,500 attempts collected per platform daily. The dataset creation process involves clustering 3D point clouds using the DBSCAN algorithm under varying lighting conditions to identify objects and randomly select clusters for grasping. This dataset is primarily applied to the evaluation of robotic grasping tasks, aiming to advance research in the field of robotic learning via standardized hardware platforms and datasets, lower the barrier to research, and promote the reproducibility and comparability of algorithms.