dataset_merged_env_WALL_NO_10_15_task_pickcube_numepi_20_ctrl_cartesian
收藏资源简介:
该数据集是一个机器人学习数据集,使用LeRobot工具创建,专门用于机器人操作任务。数据集基于Franka Emika Panda机器人平台,包含60个训练片段,总计10052帧数据,帧率为15fps。数据以parquet格式存储,总数据文件大小为100MB,视频文件大小为200MB。数据集特征包括:观察图像(主摄像头和腕部摄像头,分辨率256x256,3通道,视频编码为av1,帧率30fps)、状态信息(笛卡尔坐标6维、夹爪状态1维、关节角度7维、目标位置6维,合并为20维状态向量)、动作信息(7维,包括位置和夹爪控制)。数据集仅包含一个任务类型,从特征名称推断为拾取立方体任务(pickcube),适用于机器人控制算法的训练和评估。
This is a robotics learning dataset developed using the LeRobot toolkit, dedicated to robotic manipulation tasks. It is built on the Franka Emika Panda robotic platform, consisting of 60 training episodes with a total of 10052 frames at a frame rate of 15 fps. The data is stored in Parquet format, with a total data file size of 100 MB and a video file size of 200 MB. The dataset includes the following features: observation images (captured by a main camera and a wrist camera, with a resolution of 256×256, 3 color channels, encoded in AV1 video format, at a frame rate of 30 fps), state information (6-dimensional Cartesian coordinates, 1-dimensional gripper status, 7-dimensional joint angles, and 6-dimensional target positions, merged into a 20-dimensional state vector), and action information (7-dimensional, including position and gripper control). The dataset only covers one single task type, which is inferred as the pickcube task based on the feature names, and is applicable for training and evaluating robotic control algorithms.
数据集概述
该数据集是一个机器人操作数据集,基于 LeRobot 框架创建,适用于机器人操控任务的训练与评估。
- 任务类型:机器人操控(robotics)
- 机器人平台:Franka Emika Panda(panda)
- 任务数量:1 个(目标为拾取立方体,
task_pickcube) - 控制方式:笛卡尔空间控制(
ctrl_cartesian) - 数据集规模:
- 总片段数(episodes):60
- 总帧数:10,052
- 总任务数:1
- 数据文件大小:100 MB
- 视频文件大小:200 MB
- 帧率(fps):15
数据划分
- 训练集:全部 60 个片段(
train: 0:60)
数据特征(Features)
数据包含多种观测值、动作以及元数据,具体如下:
| 特征名称 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
observation.images.primary |
video (uint8) | (256, 256, 3) | 主摄像头图像,AV1 编码,30 fps |
observation.images.wrist |
video (uint8) | (256, 256, 3) | 腕部摄像头图像,AV1 编码,30 fps |
observation.state.cartesian |
float32 | (6,) | 末端执行器笛卡尔位姿:x, y, z, roll, pitch, yaw |
observation.state.gripper |
float32 | (1,) | 夹爪状态 |
observation.state.joints |
float32 | (7,) | 7 个关节角度 |
observation.state.target |
float32 | (6,) | 目标笛卡尔位姿:target_x, target_y, target_z, target_roll, target_pitch, target_yaw |
observation.state |
float32 | (20,) | 综合状态向量(包含笛卡尔位姿、夹爪、关节角、目标位姿) |
action |
float32 | (7,) | 动作:笛卡尔位移 + 夹爪控制(x, y, z, roll, pitch, yaw, gripper) |
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 |
数据文件结构
- 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,视频编码为 AV1 - chunk 大小:1000 帧
许可证
- Apache-2.0
数据集可视化
可通过以下空间在线可视化该数据集: https://huggingface.co/spaces/lerobot/visualize_dataset?path=cbrian/dataset_merged_env_WALL_NO_10_15_task_pickcube_numepi_20_ctrl_cartesian



