遇见数据集

so101-fold-towel-2camera-13_20260718_133712

收藏
Hugging Face2026-07-19 更新2026-07-19 收录
官方服务:

资源简介:

该数据集是一个机器人学习数据集,专门用于折叠毛巾任务,由LeRobot平台创建。它包含机器人执行动作时的多模态数据:动作数据(包括肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕旋转和夹爪位置等6个浮点值)、观察状态数据(与动作相同的关节位置)、以及来自手腕和顶部两个摄像头的图像观察(视频格式,分辨率为480x640,30fps,编码为av1)。此外,数据集还包括时间戳、帧索引、episode索引、任务索引等元数据。数据集总共有15个episodes,8488帧,覆盖1个任务,数据以parquet文件格式存储,总数据文件大小为100MB,视频文件大小为200MB。该数据集适用于机器人控制、模仿学习或强化学习的研究和开发,使用Apache 2.0许可证。

This robotics learning dataset is dedicated to the towel folding task and was created by the LeRobot platform. It includes multimodal data collected during robot task execution: action data consisting of 6 floating-point values including shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position, observation state data with identical joint positions to the action data, and image observations from two cameras (wrist camera and overhead camera) in video format with a resolution of 480x640, 30 frames per second (fps), encoded with av1. Additionally, the dataset contains metadata such as timestamps, frame indices, episode indices, and task indices. The dataset has a total of 15 episodes and 8488 frames, covering one single task. All data is stored in Parquet file format, with a total data file size of 100 MB and video file size of 200 MB. This dataset is suitable for research and development in robot control, imitation learning, and reinforcement learning, and is licensed under the Apache License 2.0.

提供机构:
aHiroakiIshikawa
创建时间:
2026-07-18
原始信息汇总

数据集概述

  • 数据集名称: so101-fold-towel-2camera-13_20260718_133712
  • 许可证: Apache-2.0
  • 任务类型: 机器人(Robotics)
  • 来源: 基于 LeRobot 框架创建
  • 机器人类型: so_follower

数据集结构

  • 数据集版本: v3.0
  • 总片段数(Episodes): 15
  • 总帧数: 8488
  • 总任务数: 1
  • 帧率(FPS): 30
  • 数据文件大小: 约 100 MB
  • 视频文件大小: 约 200 MB
  • 数据分块大小: 每块 1000 帧
  • 数据路径格式: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
  • 视频路径格式: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
  • 数据集划分: 所有 15 个片段用于训练(train: "0:15"

特征(Features)

特征名称 数据类型 形状 说明
action float32 [6] 包含 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] 观测到的机器人状态,与动作维度相同
observation.images.wrist video [480, 640, 3] 腕部摄像头视频,分辨率 480x640,AV1 编码,30 fps
observation.images.top video [480, 640, 3] 顶部摄像头视频,分辨率 480x640,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

摄像头与视频参数

  • 摄像头数量: 2(wrist 和 top)
  • 视频分辨率: 480x640(高度 x 宽度)
  • 视频编码: AV1(codec: av1)
  • 像素格式: yuv420p
  • 颜色通道: 3
  • 帧率(视频): 30 fps
  • 视频压缩参数: g(关键帧间隔): 2, crf: 30, preset: 12
二维码
社区交流群
二维码
科研交流群
商业服务