Task-one-combine
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot工具创建,旨在为机器人学习研究提供数据支持。数据集采集自型号为so100_wowskin的机器人平台,包含机器人在执行单一任务过程中的多模态记录。数据集规模庞大,总计包含504条完整操作序列(episodes),对应394,472个数据帧。数据以分块(chunk)形式存储于Parquet文件中,总数据量约为100 MB,并附带有约200 MB的视频文件。数据集结构清晰,主要包含三大类特征:1) 动作指令:一个6维浮点向量,分别控制肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转和夹爪的位置。2) 观测数据:包括机器人本体状态(一个与动作指令维度相同的6维浮点向量)、一个未详细说明的15维wowskin传感器数据,以及来自两个摄像头的视觉信息。视觉观测包含基座摄像头和腕部摄像头拍摄的彩色视频流,分辨率均为640x480,帧率为30 FPS。3) 丰富的元数据:如时间戳、帧索引、episode索引和任务索引,便于数据对齐与分析。该数据集适用于机器人模仿学习、行为克隆、强化学习以及多模态(状态+视觉)机器人控制策略的训练与评估。数据已预先划分为训练集。
This dataset is a robot operation dataset created using the LeRobot tool, designed to provide data support for robot learning research. It is collected from a robot platform model so100_wowskin and includes multimodal recordings of the robot performing a single task. The dataset is large-scale, containing a total of 504 complete operation sequences (episodes) corresponding to 394,472 data frames. The data is stored in chunked Parquet files with a total data volume of approximately 100 MB, accompanied by about 200 MB of video files. The dataset structure is clear, mainly comprising three categories of features: 1) Action commands: a 6-dimensional floating-point vector controlling shoulder translation, shoulder elevation, elbow bending, wrist bending, wrist rotation, and gripper position. 2) Observation data: including robot state (a 6-dimensional floating-point vector with the same dimensions as the action commands), an unspecified 15-dimensional wowskin sensor data, and visual information from two cameras. Visual observations include color video streams from a base camera and a wrist camera, both with a resolution of 640x480 and a frame rate of 30 FPS. 3) Rich metadata: such as timestamps, frame indices, episode indices, and task indices, facilitating data alignment and analysis. This dataset is suitable for robot imitation learning, behavior cloning, reinforcement learning, and training and evaluation of multimodal (state+vision) robot control strategies. The data has been pre-divided into a training set.
数据集概述
- 数据集名称:Task-one-combine
- 许可证:Apache-2.0
- 任务类别:机器人技术(robotics)
- 标签:LeRobot
- 创建工具:使用 LeRobot (https://github.com/huggingface/lerobot) 创建
数据集规模
- 总片段数:504
- 总帧数:394,472
- 总任务数:1
- 块大小:1,000
- 数据文件大小:100 MB
- 视频文件大小:200 MB
- 帧率(FPS):30
数据划分
- 训练集:0 到 504(全部数据用于训练)
机器人类型
- 机器人型号:so100_wowskin
数据特征
动作(Action)
- 数据类型:float32
- 维度:6
- 名称:shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
观测状态(Observation State)
- 数据类型:float32
- 维度:6
- 名称:shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
观测 wowskin
- 数据类型:float32
- 形状:[15]
观测图像 - base
- 数据类型:视频
- 分辨率:480×640(高×宽),3通道(RGB)
- 视频编码:AV1
- 色彩格式:yuv420p
- 帧率:30 FPS
- 是否为深度图:否
观测图像 - wrist
- 数据类型:视频
- 分辨率:480×640(高×宽),3通道(RGB)
- 视频编码:AV1
- 色彩格式:yuv420p
- 帧率:30 FPS
- 是否为深度图:否
时间戳(Timestamp)
- 数据类型:float32
- 形状:[1]
帧索引(Frame Index)
- 数据类型:int64
- 形状:[1]
片段索引(Episode Index)
- 数据类型:int64
- 形状:[1]
索引(Index)
- 数据类型:int64
- 形状:[1]
任务索引(Task Index)
- 数据类型:int64
- 形状:[1]
数据文件结构
- 数据路径格式:data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
- 视频路径格式:videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4




