taccap-g1-cucumber-0702
收藏Hugging Face2026-07-02 更新2026-07-02 收录
下载链接:
https://huggingface.co/datasets/TacVerse/taccap-g1-cucumber-0702
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是一个机器人操作数据集,使用双TacCap夹爪(bi_taccap_gripper)收集,包含10个完整的情节(episodes),总计6017帧数据,覆盖1个任务。数据以30帧每秒的速率采集,包括动作(action)和观测状态(observation.state),其中动作和状态均包含左右夹爪的末端执行器位置(x, y, z)和旋转(r1到r6)以及夹爪位置信息,共20个浮点数值。观测部分还包括触觉图像和手腕图像:左右触觉传感器各提供两个图像(left_tactile_left, left_tactile_right, right_tactile_left, right_tactile_right),分辨率为400x700像素,3通道(彩色);左右手腕摄像头各提供一个图像(left_wrist, right_wrist),分辨率为480x640像素,3通道。此外,数据集还包含时间戳、帧索引、情节索引、索引和任务索引等元数据。数据以Parquet格式存储,视频以MP4格式存储,适用于机器人学习、触觉感知和视觉控制等研究。
This dataset is a robotic manipulation dataset collected using a dual TacCap gripper (bi_taccap_gripper), containing 10 complete episodes, totaling 6017 frames of data, covering 1 task. Data is captured at a rate of 30 frames per second, including actions and observation states. Both actions and states contain end-effector positions (x, y, z) and rotations (r1 to r6) for the left and right grippers, as well as gripper position information, totaling 20 floating-point values. The observation part also includes tactile and wrist images: the left and right tactile sensors each provide two images (left_tactile_left, left_tactile_right, right_tactile_left, right_tactile_right) with a resolution of 400x700 pixels and 3 channels (color); the left and right wrist cameras each provide one image (left_wrist, right_wrist) with a resolution of 480x640 pixels and 3 channels. Additionally, the dataset includes metadata such as timestamps, frame indices, episode indices, indices, and task indices. Data is stored in Parquet format, and videos are stored in MP4 format, suitable for research in robotic learning, tactile perception, and visual control.
提供机构:
TacVerse创建时间:
2026-07-02
原始信息汇总
数据集概述
- 名称: taccap-g1-cucumber-0702
- 创建工具: 使用 LeRobot 创建
- 许可证: Apache-2.0
- 任务类别: 机器人学 (robotics)
- 标签: LeRobot
数据集结构
基础信息
- 代码库版本: v3.0
- 机器人类型: bi_taccap_gripper(双TacCap夹爪)
- 总片段数: 10
- 总帧数: 6017
- 总任务数: 1
- 数据块大小: 1000
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 帧率: 30 FPS
- 分割: 训练集包含全部10个片段(索引0至9)
特征字段
数据集包含以下特征,所有数据以Parquet文件和MP4视频文件形式存储。
| 特征名称 | 数据类型 | 形状 | 描述 |
|---|---|---|---|
action |
float32 | [20] | 20维动作指令,包含左右TCP(工具中心点)的6维位姿、夹爪位置 |
observation.state |
float32 | [20] | 20维观测状态,与动作维度一致(左右6维TCP位姿 + 夹爪位置) |
observation.images.left_tactile_left |
video | [400, 700, 3] | 左侧触觉左摄像头视频(H264编码,30FPS) |
observation.images.left_tactile_right |
video | [400, 700, 3] | 左侧触觉右摄像头视频 |
observation.images.left_wrist |
video | [480, 640, 3] | 左手腕摄像头视频 |
observation.images.right_tactile_left |
video | [400, 700, 3] | 右侧触觉左摄像头视频 |
observation.images.right_tactile_right |
video | [400, 700, 3] | 右侧触觉右摄像头视频 |
observation.images.right_wrist |
video | [480, 640, 3] | 右手腕摄像头视频 |
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 |
动作与观测状态维度详解
action 和 observation.state 均为20维向量,按顺序包含:
- 左夹爪TCP位姿 (前10维):
left_tcp.x,left_tcp.y,left_tcp.z(位置)left_tcp.r1至left_tcp.r6(姿态/旋转表示)left_gripper.pos(夹爪开合位置)
- 右夹爪TCP位姿 (后10维):
right_tcp.x,right_tcp.y,right_tcp.z(位置)right_tcp.r1至right_tcp.r6(姿态/旋转表示)right_gripper.pos(夹爪开合位置)
引用
若使用本数据集,请引用以下 BibTeX: 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} }



