five

so101_glue_dataset-trajectory

收藏
Hugging Face2026-05-26 更新2026-05-26 收录
下载链接:
https://huggingface.co/datasets/edmos7/so101_glue_dataset-trajectory
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集是使用LeRobot创建的,专注于机器人技术领域,具体针对so101机器人类型。数据集包含51个episodes,总计22488帧,涉及1个任务。数据以parquet文件格式存储,总数据文件大小为100MB,视频文件大小为200MB,帧率为30fps。特征包括动作(6个浮点型关节控制值,如主肩部平移、主肩部升降等)、观察状态(与动作相同的6个浮点型值)、时间戳、帧索引、episode索引、任务索引等。此外,数据集提供来自两个摄像头的视频观察:MAIN_CAM和SEC_CAM,分辨率均为480x640像素,3通道,视频编码为av1,无音频;以及轨迹图像,同样为480x640像素的3通道图像。数据集结构支持训练分割,适用于机器人控制和视觉任务的研究。

This dataset was created using LeRobot, focusing on the field of robotics, specifically targeting the so101 robot type. The dataset contains 51 episodes, totaling 22,488 frames, and covers 1 task. The data is stored in Parquet file format, with a total data file size of 100 MB, video file size of 200 MB, and a frame rate of 30 fps. The features include actions (6 floating-point joint control values, such as primary shoulder translation, primary shoulder elevation, etc.), observation states (6 identical floating-point values as the actions), timestamps, frame indices, episode indices, task indices, and more. In addition, the dataset provides video observations from two cameras: MAIN_CAM and SEC_CAM, both with a resolution of 480x640 pixels, 3 channels, encoded in AV1, and without audio; it also includes trajectory images, which are also 3-channel images with a resolution of 480x640 pixels. The dataset structure supports training splits, making it suitable for research on robot control and visual tasks.
提供机构:
edmos7
创建时间:
2026-05-26
原始信息汇总

数据集概述

  • 名称: so101_glue_dataset-trajectory
  • 许可: Apache-2.0
  • 任务类型: 机器人学 (robotics)
  • 标签: LeRobot

数据集结构

基本信息

  • 机器人型号: so101
  • 总片段数: 51
  • 总帧数: 22488
  • 总任务数: 1
  • 分块大小: 1000
  • 数据文件大小: 100 MB
  • 视频文件大小: 200 MB
  • 帧率 (fps): 30
  • 数据路径: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频路径: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4

数据划分

  • 训练集 (train): 片段索引 0 至 50 (共51个片段)

特征详情

特征名称 数据类型 形状 描述
action float32 [6] 包含6个动作维度(肩部旋转、肩部升降、肘部弯曲、腕部弯曲、腕部旋转、夹爪)
observation.state float32 [6] 与动作维度相同,为观测到的机器人状态
timestamp float32 [1] 时间戳
frame_index int64 [1] 帧索引
episode_index int64 [1] 片段索引
index int64 [1] 索引
task_index int64 [1] 任务索引
observation.images.MAIN_CAM video [480, 640, 3] 主相机视频,分辨率480x640,3通道,编码AV1,帧率30fps
observation.images.SEC_CAM video [480, 640, 3] 副相机视频,分辨率480x640,3通道,编码AV1,帧率30fps
observation.images.trajectory image [480, 640, 3] 轨迹图像,分辨率480x640,3通道

引用信息

  • 主页: [更多信息待补充]
  • 论文: [更多信息待补充]
  • BibTeX: [更多信息待补充]
二维码
社区交流群
二维码
科研交流群
商业服务