遇见数据集

so101_whiteboard_wipe

收藏
Hugging Face2026-07-17 更新2026-07-17 收录
官方服务:

资源简介:

该数据集是一个机器人技术数据集,专门用于机器人跟随任务。它包含10个episodes,总计4907帧数据,帧率为10fps。数据集由动作、状态和图像观察组成:动作包括6个浮点数值,对应机器人的肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕旋转和夹爪位置;状态同样包含这6个浮点数值。图像观察包括前视图像和手腕图像,均为480x640分辨率的三通道视频,采用AV1编码,无音频。此外,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文件格式存储,总数据文件大小为100MB,视频文件大小为200MB。该数据集适用于机器人学习、控制算法开发和模拟任务。

This is a robotics dataset specifically designed for robot following tasks. It contains 10 episodes, with a total of 4907 frames at a frame rate of 10 fps. The dataset comprises three core components: actions, states, and image observations. The actions consist of 6 floating-point values, corresponding to the robot's shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position respectively; the states also include these 6 floating-point values. The image observations cover front-view and wrist-view images, both of which are 3-channel videos with a resolution of 480×640, encoded using the AV1 format without audio tracks. Additionally, the dataset provides metadata including timestamps, frame indices, episode indices, and task indices. The data is stored in Parquet file format, with a total data file size of 100 MB and a total video file size of 200 MB. This dataset is applicable to robotics learning, control algorithm development, and simulation tasks.

提供机构:
RajatDandekar
创建时间:
2026-07-17
原始信息汇总

数据集概述

  • 数据集名称:so101_whiteboard_wipe
  • 任务类别:机器人学(robotics)
  • 许可证:Apache-2.0
  • 创建工具:LeRobot

数据集规模

  • 总片段数:10
  • 总帧数:4907
  • 总任务数:1
  • 帧率:10 FPS
  • 数据文件大小:100 MB
  • 视频文件大小:200 MB

数据集结构

  • 机器人类型:so_follower
  • 数据格式:Parquet 文件(数据路径:data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频格式:MP4 文件(视频路径:videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
  • 划分:训练集(10个片段,索引0:10)

特征描述

动作(action)

  • 数据类型:float32
  • 形状:[6]
  • 含义:6维关节位置控制指令,包括:
    • shoulder_pan.pos
    • shoulder_lift.pos
    • elbow_flex.pos
    • wrist_flex.pos
    • wrist_roll.pos
    • gripper.pos

状态观测(observation.state)

  • 数据类型:float32
  • 形状:[6]
  • 含义:6维关节位置状态,与动作特征同名。

图像观测(observation.images)

  • 前视相机(front):分辨率480x640,3通道,AV1编码,10 FPS,无音频
  • 腕部相机(wrist):分辨率480x640,3通道,AV1编码,10 FPS,无音频

其他特征

  • timestamp:float32,形状[1]
  • frame_index:int64,形状[1]
  • episode_index:int64,形状[1]
  • index:int64,形状[1]
  • task_index:int64,形状[1]
二维码
社区交流群
二维码
科研交流群
商业服务