five

put_bottle_into_basket

收藏
Hugging Face2026-06-01 更新2026-06-01 收录
下载链接:
https://huggingface.co/datasets/ClOBOT/put_bottle_into_basket
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集由LeRobot创建,是一个机器人控制数据集,专注于Unitree G1 Inspire Head机器人。数据集包含201个episodes,总计106,318帧,帧率为30fps。数据以parquet格式存储,总数据文件大小为100MB,视频文件大小为500MB。特征包括:观察状态(26维浮点数组,表示左右肩、肘、腕和手部关节的角度,如kLeftShoulderPitch、kRightHandThumbRotation等)、动作(26维浮点数组,与观察状态结构相同,用于控制关节)、观察图像(来自cam_ego_view的视频数据,分辨率为480x640,3通道,编码为av1,无音频),以及时间戳、帧索引、episode索引、索引和任务索引等元数据。数据集仅包含训练分割(0:201),适用于机器人策略学习和模仿任务。

This dataset was created by LeRobot, and is a robotic control dataset focused on the Unitree G1 Inspire Head robot. It contains 201 episodes, totaling 106,318 frames at a frame rate of 30 fps. The data is stored in Parquet format, with the total size of data files being 100 MB and the video files totaling 500 MB. The features of the dataset include: observation states which are 26-dimensional floating-point arrays representing the joint angles of the left and right shoulders, elbows, wrists and hands (e.g., kLeftShoulderPitch, kRightHandThumbRotation, etc.); actions which are 26-dimensional floating-point arrays with the same structure as the observation states and used for joint control; observation images which are video data from cam_ego_view with a resolution of 480x640, 3 channels, encoded in AV1 format and without audio; as well as metadata including timestamps, frame indices, episode indices, general indices and task indices. The dataset only includes the training split (0:201), and is suitable for robotic policy learning and imitation tasks.
提供机构:
ClOBOT
创建时间:
2026-06-01
原始信息汇总

数据集名称

put_bottle_into_basket

数据集简介

一个用于机器人操作任务的数据集,使用 LeRobot 框架创建,专注于将瓶子放入篮子这一单一任务。

数据集结构与规模

  • 机器人类型:Unitree_G1_Inspire_Head
  • 总片段数 (episodes):201
  • 总帧数 (frames):106,318
  • 总任务数:1
  • 数据块大小 (chunks_size):1000
  • 数据文件大小:约 100 MB
  • 视频文件大小:约 500 MB
  • 帧率 (fps):30
  • 数据集划分:全部 (0-200) 用作训练集

数据特征 (Features)

特征名称 数据类型 形状 说明
observation.state float32 (26,) 机器人关节状态,包含左肩、右肩、左肘、右肘、左右手腕及手指关节角度等26个维度
action float32 (26,) 动作指令,维度与状态相同,对应26个关节的动作目标
observation.images.cam_ego_view 视频 (480, 640, 3) 第一人称视角摄像头图像,分辨率 480x640,AV1 编码,30fps
timestamp float32 (1,) 时间戳
frame_index int64 (1,) 帧索引
episode_index int64 (1,) 片段索引
index int64 (1,) 全局索引
task_index int64 (1,) 任务索引

许可协议

Apache-2.0

二维码
社区交流群
二维码
科研交流群
商业服务