camMono
收藏资源简介:
该数据集是一个用于机器人技术的数据集,由LeRobot创建。它包含50个episodes,总计37918帧,数据以30帧每秒的速率采集。数据集的结构包括动作特征(如肩部、肘部、腕部和抓手的关节位置)、观测状态(与动作特征相同的关节位置)、来自前部和侧部摄像头的图像观测(视频格式,分辨率为480x640,3通道),以及时间戳、帧索引、episode索引等元数据。数据以parquet文件格式存储,视频文件以MP4格式存储,总数据大小约为100MB,视频文件大小约为200MB。数据集适用于机器人控制、模仿学习等任务。
This dataset is designed for robotics applications, created by LeRobot. It contains 50 episodes, totaling 37,918 frames, with data collected at a rate of 30 frames per second. The dataset structure includes action features (such as joint positions for shoulders, elbows, wrists, and grippers), observation states (joint positions identical to action features), image observations from front and side cameras (in video format, resolution 480x640, 3 channels), and metadata like timestamps, frame indices, and episode indices. Data is stored in parquet file format, with video files in MP4 format; total data size is approximately 100MB, and video file size is about 200MB. The dataset is suitable for tasks such as robot control and imitation learning.
数据集概要
数据集结构
基本信息
- 代码版本:v3.0
- 总帧数:37,918 帧
- 总片段数:50 个片段
- 总任务数:1 个任务
- 帧率:30 FPS
- 数据分块大小:1000
- 数据文件大小:100 MB
- 视频文件大小:200 MB
- 机器人类型:so_follower
- 数据拆分:训练集(0:50,即全部50个片段用于训练)
特征详情
| 特征 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
action |
float32 | [6] | 6维动作指令,包含肩关节、肘关节、腕关节和夹爪位置 |
observation.state |
float32 | [6] | 6维观测状态,与动作维度对应 |
observation.images.front |
video | [480, 640, 3] | 前置摄像头视频,分辨率480×640,3通道,AV1编码,YUV420P色彩格式,30 FPS |
observation.images.side |
video | [480, 640, 3] | 侧置摄像头视频,分辨率480×640,3通道,AV1编码,YUV420P色彩格式,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] | 索引 |
task_index |
int64 | [1] | 任务索引 |
动作/状态名称
动作和观测状态包含以下6个维度:
shoulder_pan.pos(肩关节水平旋转位置)shoulder_lift.pos(肩关节垂直抬升位置)elbow_flex.pos(肘关节弯曲位置)wrist_flex.pos(腕关节弯曲位置)wrist_roll.pos(腕关节旋转位置)gripper.pos(夹爪位置)
数据存储路径
- 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



