遇见数据集

grey_towel_folding_day_two

收藏
Hugging Face2026-06-04 更新2026-06-04 收录
官方服务:

资源简介:

该数据集由LeRobot工具创建,专注于机器人学任务,具体涉及双手机器人(bi_so_follower)执行折叠灰色毛巾的动作。数据集包含23个episodes,总计65594帧,以30fps的帧率采集。数据特征包括动作和观察:动作涵盖左右肩、肘、腕和夹爪的12个位置参数;观察包括机器人的状态(同样为12个位置参数)以及来自三个摄像头的图像:左左颚视角(left_left_jaw)、右右颚视角(right_right_jaw)和右俯视视角(right_topdown),每个图像分辨率为480x640x3(RGB)。数据集还包含时间戳、帧索引、episode索引、任务索引等元数据。数据以Parquet格式存储,视频以MP4格式存储,总数据文件大小为100MB,视频文件大小为200MB。该数据集用于训练或评估机器人控制算法,特别是在物体操作任务中。

This dataset is created by the LeRobot tool and focuses on robotics tasks, specifically involving a bimanual robot (bi_so_follower) performing the action of folding a gray towel. It contains 23 episodes, totaling 65,594 frames collected at 30fps. The data features include actions and observations: actions consist of 12 position parameters for the left and right shoulders, elbows, wrists, and grippers; observations include the robots state (also 12 position parameters) and images from three cameras: left-left jaw view (left_left_jaw), right-right jaw view (right_right_jaw), and right top-down view (right_topdown), each with a resolution of 480x640x3 (RGB). The dataset also includes metadata such as timestamps, frame indices, episode indices, and task indices. Data is stored in Parquet format, and videos are stored in MP4 format, with a total data file size of 100MB and video file size of 200MB. It is used for training or evaluating robot control algorithms, particularly in object manipulation tasks.

提供机构:
Janmeier820
创建时间:
2026-06-04
原始信息汇总
  • 许可证:Apache-2.0
  • 任务类别:机器人学(robotics)
  • 标签:LeRobot
  • 数据集创建工具:LeRobot(https://github.com/huggingface/lerobot)
  • 可视化:可通过 LeRobot 可视化空间查看数据集(https://huggingface.co/spaces/lerobot/visualize_dataset?path=Janmeier820/grey_towel_folding_day_two)

数据集结构

  • 机器人类型:bi_so_follower
  • 总片段数:23
  • 总帧数:65594
  • 总任务数:1
  • 数据文件大小:100 MB
  • 视频文件大小:200 MB
  • FPS:30
  • 数据格式:Parquet 文件(位于 data/*/*.parquet
  • 视频格式:MP4 文件,使用 AV1 编码,分辨率 640x480,帧率 30 FPS
  • 数据路径模板data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频路径模板videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
  • 分割:训练集包含所有 23 个片段(train: 0:23

特征

特征名 数据类型 形状 说明
action float32 (12,) 左右臂各6个自由度位置(肩部、肘部、腕部、夹爪等)
observation.state float32 (12,) 当前状态,与动作维度相同
observation.images.left_left_jaw 视频 (480, 640, 3) 左侧夹爪视角,每秒30帧,AV1编码
observation.images.right_right_jaw 视频 (480, 640, 3) 右侧夹爪视角,每秒30帧,AV1编码
observation.images.right_topdown 视频 (480, 640, 3) 顶部俯视视角,每秒30帧,AV1编码
timestamp float32 (1,) 时间戳
frame_index int64 (1,) 帧序号
episode_index int64 (1,) 片段序号
index int64 (1,) 全局索引
task_index int64 (1,) 任务索引
  • 编码器基础版本:v3.0
  • 块大小:1000
二维码
社区交流群
二维码
科研交流群
商业服务