pick_and_place
收藏资源简介:
该数据集是一个机器人任务数据集,使用LeRobot工具创建,专为机器人学任务设计,可能涉及拾取与放置任务。它包含50个episodes,总计86751帧数据,帧率为30 fps。数据结构包括:动作(action)为6维浮点向量,表示机器人关节(肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转、夹爪)的位置;观测状态(observation.state)同样为6维浮点向量,表示机器人关节的当前位置;视觉观测包括两个视角的图像:顶部摄像头图像(observation.images.top)和腕部摄像头图像(observation.images.wrist),均为分辨率480x640的RGB视频流(3通道)。此外,数据集还包含时间戳、帧索引、episode索引、全局索引和任务索引等元数据。数据以Parquet文件格式存储,并附带相应的MP4格式视频文件,适用于机器人离线强化学习、模仿学习或行为克隆等研究与应用场景。
This dataset is a robotic task dataset created using the LeRobot tool, specifically designed for robotics tasks, which may involve pick-and-place operations. It contains 50 episodes, totaling 86,751 frames of data with a frame rate of 30 fps. The dataset structure includes key feature fields: action is a 6-dimensional floating-point vector representing the positions of robot joints (shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, gripper); observation.state is also a 6-dimensional floating-point vector indicating the current positions of the robot joints; visual observations consist of images from two perspectives: a top camera image (observation.images.top) and a wrist camera image (observation.images.wrist), both as RGB video streams (3 channels) with a resolution of 480x640. Additionally, the dataset includes metadata such as timestamp, frame index, episode index, global index, and task index. The data is stored in Parquet file format and comes with corresponding MP4 format video files. It is suitable for research and application scenarios in robotic offline reinforcement learning, imitation learning, or behavior cloning.
数据集概述
该数据集名为 SJun99/pick_and_place,专注于机器人领域的拾放任务,基于 LeRobot 平台创建。
基本信息
- 许可证: Apache-2.0
- 任务类别: 机器人学 (robotics)
- 标签: LeRobot
- 首页: 暂无 (More Information Needed)
- 论文: 暂无 (More Information Needed)
数据集规模与结构
- 机器人类型: so_follower
- 总剧集数 (episodes): 50
- 总帧数 (frames): 86,751
- 总任务数: 1
- 帧率 (FPS): 30
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 分块大小 (chunks_size): 1000
- 数据划分: 训练集 (train) 包含所有 50 个 episode (索引 0 至 49)
数据特征 (Features)
数据集中每个时间步包含以下特征:
- action (动作):
- 数据类型: float32
- 维度: 6
- 包含: 6 个关节位置 (shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper)
- observation.state (状态观测):
- 数据类型: float32
- 维度: 6
- 包含: 与
action相同的 6 个关节位置
- observation.images.top (顶部摄像头图像):
- 数据类型: 视频 (video)
- 分辨率: 480x640 像素
- 颜色通道: 3 (RGB)
- 视频编码: av1
- 帧率: 30 FPS
- 非深度图
- observation.images.wrist (腕部摄像头图像):
- 数据类型: 视频 (video)
- 分辨率: 480x640 像素
- 颜色通道: 3 (RGB)
- 视频编码: av1
- 帧率: 30 FPS
- 非深度图
- timestamp: 时间戳 (float32)
- frame_index: 帧索引 (int64)
- episode_index: 剧集索引 (int64)
- index: 全局索引 (int64)
- task_index: 任务索引 (int64)
数据文件与存储
- 数据路径:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4 - 配置文件: 使用
default配置,数据文件匹配模式为data/*/*.parquet
引用信息
目前暂无 BibTeX 引用信息 (More Information Needed)。




