so101_pickplace_plus
收藏资源简介:
该数据集是一个机器人技术数据集,专为机器人学习和控制任务设计,使用LeRobot框架创建。数据集包含50个完整的情节(episodes),总计51812帧数据,针对单一任务(可能为拾取放置操作)。机器人类型为so_follower,数据以分块形式存储,每个数据文件大小为100MB,视频文件大小为200MB,帧率为30fps。数据集结构包括动作数据(action)、状态观测(observation.state)、图像观测(observation.images.handeye和observation.images.front)、时间戳(timestamp)、帧索引(frame_index)、情节索引(episode_index)、索引(index)和任务索引(task_index)。动作和状态观测均包含6个关节位置:shoulder_pan.pos、shoulder_lift.pos、elbow_flex.pos、wrist_flex.pos、wrist_roll.pos和gripper.pos,数据类型为float32。图像观测来自两个视角:handeye和front,均为视频格式,分辨率为480x640,3通道(RGB),使用av1编解码器,无音频。数据以parquet文件格式存储,视频以mp4格式存储,适用于机器人模拟、强化学习或计算机视觉研究。
This is a robotics dataset designed for robotic learning and control tasks, developed using the LeRobot framework. The dataset contains 50 complete episodes, totaling 51,812 frames, targeting a single task (presumably pick-and-place manipulation). The robot type is so_follower, and the data is stored in chunks, with each standalone data file sized at 100 MB, while the video files are 200 MB each, with a frame rate of 30 fps. The dataset structure includes action data, state observations (observation.state), image observations (observation.images.handeye and observation.images.front), timestamps (timestamp), frame indices (frame_index), episode indices (episode_index), global index (index), and task indices (task_index). Both action data and state observations include 6 joint positions: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, and gripper.pos, with a data type of float32. The image observations are captured from two viewpoints: handeye and front, both in video format with a resolution of 480×640, 3 RGB channels, using the AV1 codec, and without audio. The data is stored in Parquet file format, while videos are stored in MP4 format. This dataset is suitable for robotic simulation, reinforcement learning, or computer vision research.
数据集概要
- 名称: so101_pickplace_plus
- 许可协议: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 创建工具: 基于 LeRobot 框架创建
数据集规模
| 指标 | 数值 |
|---|---|
| 总回合数 (Episodes) | 50 |
| 总帧数 (Frames) | 51,812 |
| 总任务数 (Tasks) | 1 |
| 块大小 (Chunks Size) | 1000 |
| 数据文件大小 | 100 MB |
| 视频文件大小 | 200 MB |
| 帧率 (FPS) | 30 |
数据划分
- 训练集 (Train): 回合 0 至 49(共 50 回合)
机器人类型
- 机器人: so_follower
数据结构
特征 (Features)
动作 (Action)
- 数据类型: float32
- 维度: 6
- 变量名称: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
观测状态 (Observation.State)
- 数据类型: float32
- 维度: 6
- 变量名称: 同动作变量(shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos)
观测图像 (Observation.Images)
手眼相机 (handeye)
- 分辨率: 480x640
- 通道数: 3 (RGB)
- 帧率: 30 FPS
- 编码格式: AV1
- 像素格式: yuv420p
前方相机 (front)
- 分辨率: 480x640
- 通道数: 3 (RGB)
- 帧率: 30 FPS
- 编码格式: AV1
- 像素格式: yuv420p
其他字段
- timestamp: float32,形状 [1]
- frame_index: int64,形状 [1]
- episode_index: int64,形状 [1]
- index: int64,形状 [1]
- task_index: int64,形状 [1]
文件组织结构
- 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



