遇见数据集

taccap-g1-sort-defective-parts-0710

收藏
Hugging Face2026-07-10 更新2026-07-10 收录
官方服务:

资源简介:

该数据集是一个机器人学数据集,由LeRobot工具创建,专注于双TacCap抓手机器人(bi_taccap_gripper)的操作任务。数据集包含20个完整的情节(episodes),总计97060帧数据,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200MB。数据集特征包括:动作数据(action),包含左右机器人末端执行器(TCP)的6维位姿(位置和旋转)以及抓手的开合位置;观测状态(observation.state),与动作数据类似,记录机器人的实时状态;图像观测(observation.images),包括左右触觉传感器(tactile)和手腕摄像头(wrist)的视频数据,触觉传感器视频分辨率为400x700像素,手腕摄像头视频分辨率为480x640像素,均为彩色视频(3通道)。此外,数据集还包含时间戳、帧索引、情节索引和任务索引等元数据。该数据集适用于机器人控制、强化学习或触觉感知研究。

This is a robotics dataset created by the LeRobot toolkit, focusing on manipulation tasks with the bi_taccap_gripper dual gripper. The dataset contains 20 complete episodes, totaling 97060 frames at a frame rate of 30 fps. It is stored in Parquet format, with a total data file size of 100 MB and a total video file size of 200 MB. The dataset features include: 1. Action data (action): Contains 6-dimensional poses (position and rotation) of the left and right robot Tool Center Points (TCP), as well as the opening and closing positions of the gripper; 2. Observation state (observation.state): Similar to the action data, it records the real-time state of the robot; 3. Image observations (observation.images): Includes video data from the left and right tactile sensors and the wrist camera. The tactile sensor videos have a resolution of 400×700 pixels, while the wrist camera video has a resolution of 480×640 pixels, both of which are 3-channel color videos. Additionally, the dataset includes metadata such as timestamps, frame indices, episode indices and task indices. This dataset is suitable for research in robot control, reinforcement learning and tactile perception.

提供机构:
TacVerse
创建时间:
2026-07-10
原始信息汇总

数据集概要

  • 数据集名称:TacVerse/taccap-g1-sort-defective-parts-0710
  • 许可证:Apache-2.0
  • 任务类别:机器人学(Robotics)
  • 机器人类型:bi_taccap_gripper
  • 创建工具:使用 LeRobot 创建

数据规模与结构

  • 总片段数:20 个
  • 总帧数:97,060 帧
  • 总任务数:1 个
  • 分块大小:1000
  • 帧率:30 FPS
  • 数据文件大小:约 100 MB
  • 视频文件大小:约 200 MB
  • 数据分割:全部 20 个片段用于训练(train: 0:20)
  • 数据格式
    • 数据文件:data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
    • 视频文件:videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4(H.264 编码,YUV420P 像素格式)

特征与观测空间

  • 动作(action)
    • 类型:float32,形状 [20]
    • 包含左右 TCP 位置(x, y, z)和旋转(r1-r6),以及左右夹爪位置(pos)
  • 机器人状态(observation.state)
    • 类型:float32,形状 [20]
    • 与动作特征相同,包含左右 TCP 和夹爪状态
  • 视觉观测(observation.images):共 6 个摄像头视角
    • 左触觉左(left_tactile_left):400×700 像素,30 FPS
    • 左触觉右(left_tactile_right):400×700 像素,30 FPS
    • 左腕部(left_wrist):480×640 像素,30 FPS
    • 右触觉左(right_tactile_left):400×700 像素,30 FPS
    • 右触觉右(right_tactile_right):400×700 像素,30 FPS
    • 右腕部(right_wrist):480×640 像素,30 FPS
  • 元数据
    • timestampfloat32,形状 [1]
    • frame_indexint64,形状 [1]
    • episode_indexint64,形状 [1]
    • indexint64,形状 [1]
    • task_indexint64,形状 [1]

引用信息

bibtex @misc{xense-taccap-lerobot, author = {XenseRobotics Team}, title = {LeRobot-Xense: LeRobot with Xense Tactile Robotics Support}, howpublished = {url{https://github.com/Vertax42/xense-taccap-lerobot}}, year = {2026} }

二维码
社区交流群
二维码
科研交流群
商业服务