遇见数据集

cucumber-subtask-grab-DAgger-iter2-task-unified

收藏
Hugging Face2026-06-27 更新2026-06-27 收录
官方服务:

资源简介:

该数据集是使用LeRobot创建的机器人任务数据集,专注于抓取黄瓜子任务(cucumber-subtask-grab),采用DAgger算法迭代2次,任务类型为统一任务(task-unified)。数据集包含150个总剧集和13484个总帧,数据以Parquet格式存储,视频文件以MP4格式存储。数据集特征包括动作(action)和观察(observation),其中观察包括状态(state)和图像(images),图像来自顶部(top)、手腕(wrist)和基座(base)三个视角,分辨率均为480x640像素,帧率为20 fps。动作和状态均包含7个关节位置(如肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕滚动、夹爪位置和倾斜位置)。数据集还包含时间戳、帧索引、剧集索引、索引和任务索引等元数据。机器人类型为vibeboard_follower_tilt,数据分割为训练集(0:150)。数据集总数据文件大小为100 MB,视频文件大小为200 MB。

This dataset is a robot task dataset created using LeRobot, focusing on the cucumber-subtask-grab task, with 2 iterations of the DAgger algorithm, and categorized as a unified task (task-unified). It contains a total of 150 episodes and 13484 frames, stored in Parquet format for data and MP4 format for video files. The dataset features include action and observation, where observation comprises state and images from three perspectives: top, wrist, and base, all with a resolution of 480x640 pixels and a frame rate of 20 fps. Both action and state include 7 joint positions (such as shoulder pan, shoulder lift, elbow bend, wrist bend, wrist roll, gripper position, and tilt position). The dataset also includes metadata such as timestamps, frame index, episode index, index, and task index. The robot type is vibeboard_follower_tilt, and the data is split into a training set (0:150). The total data file size is 100 MB, and the video file size is 200 MB.

提供机构:
VibeCuisine
创建时间:
2026-06-27
原始信息汇总

数据集概述

  • 数据集名称: cucumber-subtask-grab-DAgger-iter2-task-unified
  • 许可证: Apache-2.0
  • 任务类别: 机器人学(Robotics)
  • 标签: LeRobot
  • 创建工具: LeRobot

数据集结构

  • 配置: 仅包含 default 配置,数据文件路径为 data/*/*.parquet
  • 总帧数: 13,484 帧
  • 总集数: 150 个片段(episodes)
  • 总任务数: 1 个任务
  • 帧率 (FPS): 20
  • 机器人类型: vibeboard_follower_tilt
  • 数据文件大小: 约 100 MB
  • 视频文件大小: 约 200 MB
  • 数据路径格式: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频路径格式: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
  • 划分: 所有数据用于训练(train: 0:150

特征字段

字段名称 数据类型 形状 说明
action float32 [7] 机器人动作,包括7个关节位置(shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper, tilt)
observation.state float32 [7] 机器人状态观测,与动作维度一致
observation.images.top video [480, 640, 3] 顶部摄像头视频,编码为 AV1,帧率20 FPS
observation.images.wrist video [480, 640, 3] 腕部摄像头视频,编码为 AV1,帧率20 FPS
observation.images.base video [480, 640, 3] 底部摄像头视频,编码为 AV1,帧率20 FPS
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

视频信息

  • 分辨率: 480x640 像素
  • 色彩通道: 3 通道(RGB)
  • 编码: AV1
  • 像素格式: yuv420p
  • 是否为深度图: 否
  • 音频: 无
  • 编码参数: GOP=2, CRF=30, preset=12
二维码
社区交流群
二维码
科研交流群
商业服务