jellyho/droid_subsets_wiping

这是一个基于DROID 1.0.1数据集的技能过滤子集，专门针对擦拭（wiping）技能。数据集采用LeRobot格式，通过droid_filter.py脚本从原始数据中筛选生成，包含300个唯一任务和118845帧数据，平均每集长度为396.15帧。数据保留了原始动作字段，如夹爪位置、速度等，适用于机器人技能条件微调实验。

This is a skill-filtered subset of the DROID 1.0.1 dataset, focusing specifically on the wiping skill. It is formatted for LeRobot and generated using droid_filter.py from the original data. The subset includes 300 unique tasks and 118845 frames, with an average episode length of 396.15 frames. It preserves original action fields such as gripper position and velocity, intended for lightweight skill-conditioned fine-tuning experiments in robotics.