jellyho/droid_merged_skills_picking

Merged DROID Skill Dataset: picking 是一个合并的机器人技能数据集，专门针对拾取（picking）任务。它整合了经过语言筛选的DROID技能子集和针对同一技能新收集的环境数据。数据集包含来自两个来源的301个episodes（其中300个来自droid_subsets_picking，1个来自droid_pick_carrot），总计58869帧和302个唯一任务。数据格式标准化为8维向量：观测状态包括7个关节位置和夹爪位置，动作包括7个关节速度和夹爪位置。数据集提供视频数据（如手腕左侧和侧视图视频），并包含多种拾取任务示例，例如拾取胡萝卜、从盒子中取出紫色标记并放在柜台上等。文件结构包括元数据JSON文件、Parquet数据文件、MP4视频文件以及汇总文件，适用于机器人学习和技能训练。

The Merged DROID Skill Dataset: picking is a merged robotics skill dataset focused on the picking task. It combines a language-filtered DROID skill subset with newly collected environment data for the same skill. The dataset includes 301 episodes (300 from droid_subsets_picking and 1 from droid_pick_carrot), totaling 58,869 frames and 302 unique tasks. The data format is standardized as 8-dimensional vectors: observation state includes 7 joint positions and gripper position, while action includes 7 joint velocities and gripper position. The dataset provides video data (e.g., wrist left and side view videos) and contains examples of various picking tasks, such as Pick up the carrot, Remove the purple marker from the box and put it on the counter, etc. The file structure includes metadata JSON files, Parquet data files, MP4 video files, and summary files, suitable for robotics learning and skill training.