jellyho/droid_merged_skills_wiping

这是一个合并的DROID技能数据集，专注于擦拭技能。它通过合并一个经过语言过滤的DROID技能子集（来自jellyho/droid_subsets_wiping）和针对同一技能新收集的环境数据（来自jellyho/droid_wipe_table）而构建。数据集用于机器人学任务，特别是擦拭操作，如使用毛巾擦拭桌子等。数据经过处理，标准化为LeRobot训练字段，使用8D向量表示观测状态（包括关节位置和夹爪位置）和动作（包括关节速度和夹爪位置）。数据集包含319个episodes、80582帧、300个独特任务，状态和动作形状均为[8]，并提供了视频数据（如手腕左侧和侧视图像）。数据集文件包括parquet文件、元数据JSON文件、CSV文件和视频文件。

This is a merged DROID skill dataset focused on the wiping skill. It is constructed by merging a language-filtered DROID skill subset (from jellyho/droid_subsets_wiping) with newly collected environment data for the same skill (from jellyho/droid_wipe_table). The dataset is designed for robotics tasks, specifically wiping operations such as using a towel to wipe a table. The data is processed and standardized into LeRobot training fields, using 8D vectors for observation state (including joint positions and gripper position) and action (including joint velocities and gripper position). The dataset includes 319 episodes, 80582 frames, 300 unique tasks, with state and action shapes of [8], and provides video data (e.g., wrist left and side view images). Dataset files consist of parquet files, metadata JSON files, CSV files, and video files.