Fanqi-Lin/Processed-Task-Dataset

该存储库包含使用UMI收集的野外机器人操作数据集，并通过SLAM管道进行处理，如论文《机器人操作模仿学习中的数据缩放定律》所述。数据集涵盖四个任务：倒水、整理鼠标、折叠毛巾和拔下充电器。每个任务的数据集文件夹包含不同数量的环境-对象对和演示次数。例如，arrange_mouse和pour_water文件夹各包含32个独特的环境-对象对，每个对有120次演示；而fold_towel和unplug_charger文件夹各包含32个独特的环境-对象对，每个对有60次演示。此外，还提供了关于数据集合并和使用的额外信息。

This dataset contains robotic manipulation data for four tasks: pouring water, arranging mouse, folding towel, and unplugging charger. The data was collected using the UMI robot and processed through a SLAM pipeline. Each tasks data comes from 32 unique environment-object pairs, with 120 demonstrations for pouring water and arranging mouse tasks, and 60 demonstrations for folding towel and unplugging charger tasks. Additionally, there are two subsets containing data from 16 environments with 4 different manipulation objects each, with 120 demonstrations per object. The dataset can be used to train policies that generalize effectively to novel environments and objects.