rollout_pick_n_place_dagger_r11
收藏资源简介:
该数据集是一个用于机器人任务的开源数据集,使用LeRobot工具创建。数据集包含机器人在执行任务过程中的交互数据,适用于机器人学习与行为克隆等研究场景。数据集中包含1个完整的训练情节(episode),总计3351个数据帧,采样频率为30Hz。数据采用多模态形式组织,主要包含以下字段:1) 动作(action):一个6维浮点数组,表示机器人六个关节(肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转、夹爪)的目标位置。2) 观测状态(observation.state):一个与动作维度相同的6维浮点数组,表示机器人关节的当前实际位置。3) 图像观测(observation.images):包含两个视角的视频流(front和annotated),均为RGB彩色视频,分辨率为640x480,帧率30fps,采用AV1编码。4) 元数据:包括时间戳(timestamp)、帧索引(frame_index)、情节索引(episode_index)、全局数据索引(index)和任务索引(task_index)。数据以分块的Parquet文件格式存储,对应的视频文件以MP4格式单独存储。数据集对应的机器人平台类型为so_follower。
This dataset is an open-source dataset for robotic tasks, created using the LeRobot tool. The core content of the dataset is interaction data from robots during task execution, suitable for research scenarios such as robot learning and behavior cloning. The dataset contains 1 complete training episode, totaling 3351 data frames with a sampling frequency of 30 Hz. The data is organized in a multimodal format, primarily including the following fields: 1) Action: a 6-dimensional floating-point array representing the target positions of six robot joints (shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, gripper). 2) Observation.state: a 6-dimensional floating-point array with the same dimensions as the action, representing the current actual positions of the robot joints. 3) Observation.images: includes two perspective video streams (front and annotated), both RGB color videos with a resolution of 640x480, a frame rate of 30 fps, and encoded in AV1. 4) Metadata: includes timestamp, frame_index, episode_index, index, and task_index. The data is stored in chunked Parquet file format, with corresponding video files stored separately in MP4 format. The robot platform type for this dataset is so_follower.
数据集概述
- 数据集名称:
anikitakis/rollout_pick_n_place_dagger_r11 - 许可证: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 标签: LeRobot
数据集创建
该数据集使用 LeRobot 工具创建。
数据集结构
数据集包含一个配置(default),数据文件格式为 data/*/*.parquet。
元数据(meta/info.json)
- 代码库版本: v3.0
- 帧率 (fps): 30
- 机器人类型:
so_follower
特征 (Features)
| 特征名 | 数据类型 | 形状 | 描述 |
|---|---|---|---|
action |
float32 | [6] | 6个关节的动作指令:shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos |
observation.state |
float32 | [6] | 6个关节的状态观测,名称与动作维度一致 |
observation.images.front |
video | [480, 640, 3] | 前置摄像头图像,30fps,AV1编码,分辨率480x640,RGB三通道 |
observation.images.annotated |
video | [480, 640, 3] | 标注过的图像,规格与前摄像头相同 |
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 |
统计信息
- 总片段数 (total_episodes): 1
- 总帧数 (total_frames): 3351
- 总任务数 (total_tasks): 1
- 数据块大小 (chunks_size): 1000
- 数据文件大小: 100 MB
- 视频文件大小: 200 MB
- 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
数据划分 (Splits)
- 训练集 (train): 索引范围
0:1(共1个片段)
可视化
可通过 LeRobot 可视化空间 查看该数据集。




