pick_bomb
收藏资源简介:
该数据集是一个用于机器人学研究的开源数据集,使用LeRobot代码库(v3.0版本)创建,专门针对双机械臂跟随机器人(bi_so_follower)平台。数据集旨在支持机器人模仿学习、强化学习或行为克隆等任务,特别是涉及双臂协调操作的研究。数据集规模为168个完整任务执行序列(episodes),共计136,483个时间步(帧)。数据以分块形式组织(每块约1000帧),原始数据文件总大小约为100MB,对应的视频文件总大小约为200MB。数据内容包含机器人的动作指令、观测状态以及多视角视觉信息。具体特征包括:动作是一个12维浮点数向量,控制左、右机械臂的6个关节位置;观测包含状态(12维浮点数向量)和图像(来自左主摄像头和右腕部摄像头的RGB视频流,分辨率240x320,帧率30 FPS,使用H.264编码);元数据包括时间戳、帧索引、episode索引、全局数据索引和任务索引。数据集文件以Parquet格式存储,视频文件为MP4格式,已划分为训练集(包含所有168个episodes),采用Apache 2.0许可证发布。
This dataset is an open-source dataset for robotics research, created using the LeRobot codebase (version 3.0), specifically designed for the bi_so_follower (dual-arm following robot) platform. It aims to support tasks such as robot imitation learning, reinforcement learning, or behavior cloning, particularly research involving dual-arm coordinated operations. The dataset consists of 168 complete task execution episodes, totaling 136,483 timesteps (frames). Data is organized in chunks (approximately 1000 frames per chunk), with the raw data files totaling about 100MB and corresponding video files totaling about 200MB. The content includes robot action commands, observation states, and multi-view visual information. Specific features are: action is a 12-dimensional floating-point vector controlling the 6 joint positions of the left and right robotic arms; observation includes state (a 12-dimensional floating-point vector) and images (RGB video streams from two fixed perspectives: left_main camera and right_right_wrist camera, with a resolution of 240x320, frame rate of 30 FPS, encoded using H.264); metadata includes timestamp, frame_index, episode_index, index, and task_index for data alignment and sequence analysis. The dataset files are stored in Parquet format, with video files in MP4 format, and it is divided into a training set (containing all 168 episodes). The dataset is released under the Apache 2.0 license.
数据集概述
- 数据集名称: pick_bomb
- 许可证: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 标签: LeRobot, TeLeRobot
- 创建工具: 使用 LeRobot 创建
数据集结构
基本信息
- 代码库版本: v3.0
- 机器人类型: bi_so_follower
- 总片段数 (Episodes): 168
- 总帧数: 136,483
- 总任务数: 1
- 数据块大小 (Chunks Size): 1000
- 数据文件大小: 约100 MB
- 视频文件大小: 约200 MB
- 帧率 (FPS): 30
数据划分
- 训练集 (Train): 片段 0 至 167 (共168个片段)
文件路径
- 数据文件路径:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频文件路径:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
特征 (Features)
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
action |
float32 | [12] | 12维动作值,包含左右臂各6个关节位置(肩部、肘部、腕部、夹爪) |
observation.state |
float32 | [12] | 12维观测状态值,与action维度对应 |
observation.images.left_main |
video | [240, 320, 3] | 左侧主摄像头视频,分辨率240x320,H.264编码,30 FPS |
observation.images.right_right_wrist |
video | [240, 320, 3] | 右侧腕部摄像头视频,分辨率240x320,H.264编码,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 |
可视化
- 可通过 可视化空间 预览数据集内容。
引用
- BibTeX: 待补充




