stackblocks_red_yellow_so100_blue
收藏资源简介:
该数据集是使用LeRobot机器人学习平台创建的机器人操作数据集,专注于机器人堆叠积木的任务(从文件名‘stackblocks_red_yellow_so100_blue’推断)。数据集包含100个完整的操作回合,总计79,781帧数据,帧率为30fps。数据以Parquet格式存储,视频文件为MP4格式,总数据量约为300MB(其中数据部分100MB,视频部分200MB)。使用的机器人类型为‘so_follower’。每个数据样本包括以下关键特征:1) 动作:一个6维浮点向量,控制机器人的肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置;2) 观测状态:与动作相同的6维关节位置状态向量;3) 观测图像:包含两个视角的RGB视频流,分别是顶部摄像头和腕部摄像头,分辨率均为640x480,3通道;4) 元数据:包括时间戳、帧索引、回合索引、样本索引和任务索引。所有数据被划分为训练集(覆盖0-100回合)。该数据集适用于机器人模仿学习、强化学习、视觉运动策略学习等研究任务,特别适合涉及多视角视觉观测和关节位置控制的操作任务研究。
This dataset is a robot manipulation dataset created using the LeRobot robot learning platform, focusing on a robot stacking blocks task (inferred from the filename stackblocks_red_yellow_so100_blue). It contains 100 complete operation episodes, totaling 79,781 frames with a frame rate of 30fps. The data is stored in Parquet format, with video files in MP4 format, and the total data size is approximately 300MB (100MB for data and 200MB for videos). The robot type used is so_follower. Each data sample includes the following key features: 1) Action: a 6-dimensional floating-point vector controlling the robots shoulder translation, shoulder lift, elbow bend, wrist bend, wrist rotation, and gripper position; 2) Observation state: a 6-dimensional joint position state vector identical to the action; 3) Observation images: RGB video streams from two perspectives, namely a top camera and a wrist camera, both with a resolution of 640x480 and 3 channels; 4) Metadata: including timestamp, frame index, episode index, sample index, and task index. All data is partitioned into a training set (covering episodes 0-100). This dataset is suitable for research tasks such as robot imitation learning, reinforcement learning, and visual-motor policy learning, particularly for operation tasks involving multi-view visual observations and joint position control.
数据集概述
- 数据集名称:stackblocks_red_yellow_so100_blue
- 许可证:Apache-2.0
- 任务类别:机器人学(robotics)
- 标签:LeRobot
- 创建工具:使用 LeRobot 创建
数据结构
- 机器人类型:so_follower
- 帧率:30 FPS
- 总帧数:79,781 帧
- 总片段数:100 个片段
- 总任务数:1 个任务
- 数据文件大小:100 MB
- 视频文件大小:200 MB
- 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 - 数据分块大小:1000
- 数据集划分:训练集(train)包含全部 100 个片段(索引 0:100)
特征字段
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
action |
float32 | (6,) | 机器人动作(肩部、肘部、腕部、夹爪等位置) |
observation.state |
float32 | (6,) | 机器人观测状态(与动作维度对应) |
observation.images.top |
video | (480, 640, 3) | 顶部摄像头图像(高度480,宽度640,3通道) |
observation.images.wrist |
video | (480, 640, 3) | 腕部摄像头图像(高度480,宽度640,3通道) |
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 |
图像与视频配置
- 编解码器:AV1
- 像素格式:yuv420p
- 帧率:30 FPS
- 无音频
- 视频高度:480,视频宽度:640,通道数:3
- 非深度图
可用资源
- 可视化数据集:Visualize this dataset




