so101_pen_wooden_box_20260519_103919
收藏资源简介:
该数据集是一个用于机器人操作任务的演示数据集,具体任务为“将笔放入木盒”。它由LeRobot项目创建,旨在为机器人模仿学习或强化学习提供训练数据。数据集包含200个完整的任务执行片段,总计75592个时间步(帧)。每个数据样本包含多模态信息:1)机器人动作:一个6维浮点向量,表示肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转和夹爪的位置指令。2)机器人观测状态:一个与动作维度相同的6维浮点向量,表示机器人关节的实时位置。3)视觉观测:包含两个固定视角的RGB视频流(顶部视角和腕部视角),分辨率均为640x480,帧率为30fps。4)元数据:包括时间戳、帧索引、片段索引等。数据以分块Parquet文件格式组织,并配有对应的MP4视频文件。该数据集适用于训练机器人执行拾取放置类任务,特别是基于视觉的机械臂控制策略学习。
This dataset is a demonstration dataset for robotic manipulation tasks, specifically the task of "placing a pen into a wooden box". It was created by the LeRobot project, aiming to provide training data for robot imitation learning or reinforcement learning. The dataset contains 200 complete task execution episodes, totaling 75592 time steps (frames). Each data sample includes multimodal information: 1) Robot action: a 6-dimensional floating-point vector representing the position commands for shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and the gripper. 2) Robot observation state: a 6-dimensional floating-point vector with the same dimension as the action, representing the real-time positions of the robot's joints. 3) Visual observations: two RGB video streams from fixed viewpoints (top-down viewpoint and wrist-mounted viewpoint), both with a resolution of 640×480 and a frame rate of 30 fps. 4) Metadata: including timestamp, frame index, episode index, etc. The data is organized in chunked Parquet file format, with corresponding MP4 video files. This dataset is suitable for training robots to perform pick-and-place tasks, especially for learning vision-based robotic arm control policies.
数据集概述
该数据集用于机器人抓取与放置任务,具体任务为将笔放入木盒中。
基本信息
- 数据集名称:Task: put pen in wooden box
- 许可证:Apache-2.0
- 创建工具:LeRobot
- 帧率:30 FPS
- 机器人类型:so_follower
数据集规模
- 总片段数:200
- 总帧数:75,592
- 总任务数:1
- 数据文件大小:100 MB
- 视频文件大小:200 MB
数据集划分
- 训练集:索引 0 至 199(共200个片段)
特征结构
每个数据样本包含以下特征:
| 特征名 | 数据类型 | 形状 | 说明 |
|---|---|---|---|
| action | float32 | (6,) | 6维动作:肩关节(pan/lift)、肘关节、腕关节(flex/roll)、夹爪位置 |
| observation.state | float32 | (6,) | 6维观察状态,与动作维度相同 |
| observation.images.top | video | (480, 640, 3) | 顶部摄像头视频,AV1编码,30 FPS,RGB |
| observation.images.wrist | video | (480, 640, 3) | 腕部摄像头视频,AV1编码,30 FPS,RGB |
| 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 - 数据格式:Parquet(动作/状态) + MP4(视频)




