遇见数据集

rollout_pi0_lerobot_pick_cube_into_the_bowl_20260703_193951

收藏
Hugging Face2026-07-03 更新2026-07-03 收录
官方服务:

资源简介:

该数据集是一个机器人操作数据集,专门用于将立方体拾取到碗中的任务。它由LeRobot工具创建,包含机器人执行该任务时的动作和观测数据。数据集结构包括动作(action)、观测状态(observation.state)、来自基础摄像头和手腕摄像头的图像观测(observation.images.base_camera和observation.images.wrist_camera),以及时间戳、帧索引、回合索引等元数据。动作和观测状态都包含6个浮点数值,分别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。图像数据以视频格式存储,分辨率为480x640,帧率为30fps,使用AV1编码。数据集总共有1个回合、187帧和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。该数据集适用于机器人学习、模仿学习或强化学习研究,旨在帮助训练和评估机器人执行精细操作任务的能力。

This dataset is a robotic manipulation dataset specifically designed for the task of picking up a cube into a bowl. It is created using the LeRobot tool and includes action and observation data from a robot performing this task. The dataset structure comprises actions (action), observation states (observation.state), image observations from a base camera and a wrist camera (observation.images.base_camera and observation.images.wrist_camera), as well as metadata such as timestamps, frame indices, and episode indices. Both actions and observation states contain 6 floating-point values corresponding to the robots shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and gripper position. Image data is stored in video format with a resolution of 480x640, a frame rate of 30fps, and AV1 encoding. The dataset consists of 1 episode, 187 frames, and 1 task. Data is stored in Parquet file format, with a total data file size of 100MB and a video file size of 200MB. This dataset is suitable for research in robotic learning, imitation learning, or reinforcement learning, aiming to assist in training and evaluating robots ability to perform fine manipulation tasks.

提供机构:
dongdong123123
创建时间:
2026-07-03
原始信息汇总

数据集概述

该数据集是一个机器人操作相关的数据集,使用 LeRobot 框架创建,许可协议为 Apache-2.0。

数据集结构

基本信息

  • 总片段数 (episodes):1
  • 总帧数 (frames):187
  • 总任务数 (tasks):1
  • 帧率 (FPS):30
  • 机器人类型:so_follower

数据特征

特征名称 数据类型 形状 说明
action float32 [6] 6维机器人动作指令(关节位置)
observation.state float32 [6] 6维机器人状态观测(关节位置)
observation.images.base_camera video [480, 640, 3] 基座摄像头视频,分辨率 480×640,3通道,AV1编码,30帧
observation.images.wrist_camera video [480, 640, 3] 腕部摄像头视频,分辨率 480×640,3通道,AV1编码,30帧
timestamp float32 [1] 时间戳
frame_index int64 [1] 帧索引
episode_index int64 [1] 片段索引
index int64 [1] 数据索引
task_index int64 [1] 任务索引

动作与状态命名

动作和状态均包含6个关节名称:

  • shoulder_pan.pos
  • shoulder_lift.pos
  • elbow_flex.pos
  • wrist_flex.pos
  • wrist_roll.pos
  • gripper.pos

数据文件

  • 数据路径data/*/*.parquet(Parquet格式,分块存储)
  • 视频路径videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
  • 数据文件大小:约100 MB
  • 视频文件大小:约200 MB
  • 数据划分:全部数据用于训练(train 划分)
二维码
社区交流群
二维码
科研交流群
商业服务