Task-one-twobag
收藏资源简介:
该数据集使用LeRobot框架创建,是一个专门针对so100_wowskin型机器人的控制数据集。它包含62个episodes,总计42960帧数据,覆盖1个任务,数据组织为训练集(包含所有episodes)。数据以Parquet文件格式存储,总数据量约100MB,关联的视频文件总大小约200MB,视频帧率为30 FPS。数据集的核心是多模态时间序列数据,每个数据点(帧)包含:1) 机器人动作指令:一个6维浮点向量,对应肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转和夹爪的位置控制;2) 机器人状态观测:一个6维浮点向量,记录与动作指令相同的6个关节的当前位置;3) WoW皮肤传感器观测:一个15维浮点向量,可能代表触觉或皮肤传感器数据;4) 视觉观测:来自基座摄像头和腕部摄像头的两路RGB视频流,分辨率均为640x480,3通道,以AV1编码的MP4视频文件存储;5) 元数据:包括时间戳、帧索引、episode索引、全局索引和任务索引。该数据集适用于机器人模仿学习、强化学习、多模态感知与控制策略训练等任务。
This dataset is developed using the LeRobot framework, and is a specialized control dataset tailored for the so100_wowskin robot. It comprises 62 episodes, totaling 42,960 frames of data, covering 1 task, with the data organized as a training set containing all episodes. The data is stored in Parquet file format, with a total data volume of approximately 100 MB. The associated video files have a total size of around 200 MB, and the video frame rate is 30 FPS. The core of this dataset is multimodal time-series data, where each data point (frame) contains: 1) Robot action command: a 6-dimensional floating-point vector corresponding to the position controls of shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and gripper; 2) Robot state observation: a 6-dimensional floating-point vector recording the current positions of the same 6 joints as those in the action command; 3) WoW skin sensor observation: a 15-dimensional floating-point vector that may represent tactile or skin sensor data; 4) Visual observation: two RGB video streams from the base camera and wrist camera, both with a resolution of 640×480 and 3 channels, stored as AV1-encoded MP4 video files; 5) Metadata: including timestamp, frame index, episode index, global index, and task index. This dataset is suitable for tasks such as robot imitation learning, reinforcement learning, multimodal perception and control strategy training.
数据集概述
- 数据集名称:Task-one-twobag
- 许可协议:Apache-2.0
- 任务类别:机器人学(Robotics)
- 创建工具:使用 LeRobot 创建
数据集规模与结构
| 指标 | 数值 |
|---|---|
| 总片段数(Episodes) | 63 |
| 总帧数(Frames) | 43816 |
| 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 |
| 帧率(FPS) | 30 |
| 机器人类型 | so100_wowskin |
数据划分
- 训练集:全部63个片段(索引0至62)
数据特征(Features)
- action:6维浮点数组,对应机器人关节位置(shoulder_pan、shoulder_lift、elbow_flex、wrist_flex、wrist_roll、gripper)
- observation.state:6维浮点数组,与 action 维度含义相同
- observation.wowskin:15维浮点数组,代表触觉传感器数据
- observation.images.base:视频数据,分辨率480×640,RGB三通道,AV1编码,帧率30fps
- observation.images.wrist:视频数据,分辨率480×640,RGB三通道,AV1编码,帧率30fps
- timestamp:时间戳(1维浮点)
- frame_index:帧索引(1维整数)
- episode_index:片段索引(1维整数)
- index:全局索引(1维整数)
- task_index:任务索引(1维整数)
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
其他信息
- 代码库版本:v3.0
- 引用信息:待补充




