SWITCH-3-bottom-left
收藏Hugging Face2026-07-21 更新2026-07-21 收录
下载链接:
https://huggingface.co/datasets/lenawngr/SWITCH-3-bottom-left
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是一个使用LeRobot平台创建的机器人操作数据集,包含20个训练片段,总计5516帧数据,专为单一任务设计。数据特征包括6维动作空间(涵盖肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置),观测状态包含相同的6个关节位置,以及来自顶部摄像头和夹爪摄像头的视频观测(视频分辨率为480x640,帧率为30fps)。数据以parquet格式存储,视频文件采用AV1编码。机器人类型为so_follower,适用于机器人控制与模仿学习研究领域。
This dataset is a robot manipulation dataset created using the LeRobot platform, containing 20 training segments with a total of 5516 frames, designed for a single task. Data features include a 6-dimensional action space (covering shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and gripper position), observation states include the same 6 joint positions, as well as video observations from a top camera and a gripper camera (with a resolution of 480x640 and a frame rate of 30fps). The data is stored in parquet format, and video files use AV1 encoding. The robot type is so_follower, suitable for research in robot control and imitation learning.
提供机构:
lenawngr创建时间:
2026-07-21
原始信息汇总
数据集概述:SWITCH-3-bottom-left
- 许可证:Apache-2.0
- 任务类别:机器人学(robotics)
- 标签:LeRobot
- 创建工具:使用 LeRobot 创建
数据集结构
- 总片段数(episodes):20
- 总帧数(frames):5516
- 总任务数:1
- 帧率(FPS):30
- 机器人类型:so_follower
特征(Features)
| 特征名 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
action |
float32 | [6] | 包含6个关节动作(shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper) |
observation.state |
float32 | [6] | 与action相同的6个关节状态 |
observation.images.top |
video | [480, 640, 3] | 顶部摄像头视频,分辨率480×640,3通道,AV1编码,30 FPS |
observation.images.gripper |
video | [480, 640, 3] | 夹爪摄像头视频,分辨率480×640,3通道,AV1编码,30 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 |
数据文件
- 数据路径模式:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频路径模式:
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
- 训练集(train):所有20个片段(索引0至19)
可视化
可通过 可视化空间 直接浏览数据集内容。




