workspace_cleaning_v5
收藏资源简介:
该数据集是一个机器人控制数据集,基于LeRobot框架构建。它包含101个episodes,总计106,367帧数据,采样频率为30fps。数据以parquet格式存储,总大小约300MB(其中数据文件100MB,视频文件200MB)。数据集核心内容包括机器人动作指令和观测状态:动作指令(action)是一个6维浮点向量,控制机器人的六个关节位置(肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部旋转、夹爪);观测状态(observation.state)也是一个6维浮点向量,反映这些关节的实时位置。此外,提供两种视觉观测:顶置摄像头(observation.images.overhead)和腕部摄像头(observation.images.wrist)的视频流,分辨率均为640x480 RGB。每个数据点还包含时间戳、帧索引、episode索引、全局索引和任务索引。数据集仅包含一个任务,使用的机器人类型为“so_follower”,所有数据均被划分为训练集。适用于机器人模仿学习、视觉伺服控制、行为克隆等任务的研究与开发。
This robotic control dataset is built upon the LeRobot framework. It comprises 101 episodes and a total of 106,367 frames, with a sampling rate of 30 fps. The data is stored in Parquet format, with an overall size of approximately 300 MB (100 MB for data files and 200 MB for video files). The core contents of the dataset include robotic action commands and observation states: the action command (action) is a 6-dimensional floating-point vector that controls the six joint positions of the robot (shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and gripper); the observation state (observation.state) is also a 6-dimensional floating-point vector that reflects the real-time positions of these joints. In addition, two types of visual observations are provided: video streams from the overhead camera (observation.images.overhead) and the wrist camera (observation.images.wrist), both with a resolution of 640×480 RGB. Each data point also includes a timestamp, frame index, episode index, global index, and task index. The dataset includes exactly one task, utilizes the robot model "so_follower", and all data is partitioned into the training set. It is suitable for research and development of tasks such as robotic imitation learning, visual servo control, and behavior cloning.
数据集总览
- 数据集名称: workspace_cleaning_v5
- 数据集地址: https://huggingface.co/datasets/rishgang22/workspace_cleaning_v5
- 许可证: Apache-2.0
- 任务类别: 机器人学 (robotics)
- 标签: LeRobot
数据集描述
该数据集使用 LeRobot 框架创建,专为机器人相关任务设计。
数据集结构
根据 meta/info.json 文件,数据集结构如下:
- 代码库版本: v3.0
- 帧率 (FPS): 30
- 机器人类型: so_follower
- 总片段数 (Episodes): 101
- 总帧数 (Frames): 106,367
- 总任务数 (Tasks): 1
- 数据块大小 (Chunks Size): 1000
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
特征 (Features)
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
action |
float32 | [6] | 包含6个关节的动作指令:shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos |
observation.state |
float32 | [6] | 机器人状态观测,与action维度相同,对应6个关节的位置 |
observation.images.overhead |
video | [480, 640, 3] | 俯视摄像头视频,分辨率480x640,3通道,AV1编码,30fps |
observation.images.wrist |
video | [480, 640, 3] | 腕部摄像头视频,分辨率480x640,3通道,AV1编码,30fps |
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
数据集分割
- 训练集 (train): 片段索引范围 0 到 100 (共101个片段),所有片段均用于训练。
引用信息
当前无可用引用信息。




