Raffael-Kultyshev/mini_tug_tape_to_bowl

Mini-TUG手部姿势数据集（LeRobot格式）包含经过整理的RGB视频和6自由度关节轨迹数据，用于将胶带卷放入碗中的任务。数据集采用LeRobot数据集规范，每个片段来源于`Recordings_3D/vidXX`，使用MediaPipe测量的手部姿势存储在`processed/hand_pose.parquet`中，并包含修剪后的RGB MP4视频（`vidXX_rgb.mp4`）。数据集的状态和动作包含每帧24个浮点数（`{wrist, thumb_tip, index_mcp, index_tip} × {x,y,z,yaw,pitch,roll}`），时间戳与原始ARKit时间戳一致，语言指令固定为拿起胶带卷并放入碗中。数据集布局包括数据文件、视频文件和元数据文件。

The Mini-TUG Hand Pose Dataset (LeRobot Format) contains curated RGB videos and 6-DoF joint trajectories for the tape roll to bowl task, converted to the LeRobot dataset specification. Each episode was sourced from `Recordings_3D/vidXX`, using the MediaPipe-measured hand poses stored in `processed/hand_pose.parquet` and the trimmed RGB MP4 (`vidXX_rgb.mp4`). The `observation.state`/`action` contain 24 floats per frame (`{wrist, thumb_tip, index_mcp, index_tip} × {x,y,z,yaw,pitch,roll}`), the `timestamp` mirrors the original ARKit timestamps, and the `language_instruction` is always Pick up the roll of tape and put it in the bowl. The dataset layout includes data files, video files, and metadata files.