record-sponge-1-merged
收藏资源简介:
该数据集是一个机器人任务数据集,专门用于so101_follower机器人。它包含55个episodes,总计29524帧数据,帧率为30fps。数据集主要用于训练和评估机器人控制算法。特征包括动作空间(6个关节位置:肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置)、观测状态(与动作相同的6个关节位置)、两个摄像头视频观测(前视和顶视,分辨率均为480x640,3通道彩色视频,编码为av1),以及时间戳、帧索引、episode索引等元数据。数据以parquet格式存储,视频以mp4格式存储。数据集适用于机器人学习任务,如模仿学习或强化学习。
This dataset is a robotic task dataset specifically designed for the so101_follower robot. It contains 55 episodes, totaling 29524 frames with a frame rate of 30 fps. It is primarily used for training and evaluating robotic control algorithms. The features include the action space (6 joint positions: shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position), the observation state (the same 6 joint positions as the action space), two camera video observations (front-facing and top-facing, both with a resolution of 480x640, 3-channel color video encoded with AV1), as well as metadata such as timestamps, frame indices, and episode indices. The dataset is stored in Parquet format, while the videos are stored in MP4 format. This dataset is applicable to robotic learning tasks such as imitation learning or reinforcement learning.
数据集概述:record-sponge-1-merged
该数据集是一个用于机器人技术研究的机器人操作数据集,基于 LeRobot 框架创建,采用 Apache-2.0 许可证。
数据集基本信息
- 数据集名称:record-sponge-1-merged
- 许可证:Apache-2.0
- 任务类别:机器人 (robotics)
- 机器人类型:so101_follower
- 总片段数:55
- 总帧数:29,524
- 总任务数:1
- 帧率:30 FPS
- 数据文件大小:100 MB
- 视频文件大小:200 MB
- 分片大小:1,000
数据划分
- 训练集:包含所有 55 个片段(索引 0 至 54)
特征结构
数据集包含以下特征:
| 特征名称 | 数据类型 | 形状 | 描述 |
|---|---|---|---|
action |
float32 | (6,) | 机器人6维动作指令(关节位置/夹爪) |
observation.state |
float32 | (6,) | 机器人6维状态观察(关节位置/夹爪) |
observation.images.front |
video | (480, 640, 3) | 前置摄像头视频(AV1编码,30 FPS) |
observation.images.top |
video | (480, 640, 3) | 顶部摄像头视频(AV1编码,30 FPS) |
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(仅1个任务) |
- 动作/状态维度含义:肩关节旋转、肩关节抬升、肘关节弯曲、腕关节弯曲、腕关节旋转、夹爪位置
- 摄像头图像:分辨率 480×640,3通道彩色视频,AV1编码,无深度信息
数据存储结构
- 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
可视化与工具
- 可通过 LeRobot 可视化空间在线交互查看数据集 (https://huggingface.co/spaces/lerobot/visualize_dataset?path=satoshi36/record-sponge-1-merged)



