five

so101-single-arm-sort-dice-delta-v2.1

收藏
Hugging Face2026-06-19 更新2026-06-19 收录
下载链接:
https://huggingface.co/datasets/SimKienTech/so101-single-arm-sort-dice-delta-v2.1
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集是一个专为机器人学设计的数据集,使用LeRobot工具创建。它包含100个剧集,总计120,681帧,帧率为30fps,所有数据均用于训练目的。数据集提供了丰富的多模态信息:机器人的动作数据为18维,涵盖肩部、肘部、腕部和夹持器的位置、速度和角度;观测状态也为18维,包括位置、速度和负载信息。图像观测来自多个视角:手腕摄像头、两个全局摄像头和两个低角度摄像头,所有图像均为480x640分辨率、3通道彩色视频,采用AV1编码。此外,数据集还包含时间戳、帧索引、剧集索引、任务索引等元数据。数据以Parquet格式存储,视频则以MP4格式存储。

This dataset is specifically designed for robotics and created using the LeRobot toolkit. It contains 100 episodes, totaling 120,681 frames at a frame rate of 30fps, with all data intended for training purposes. The dataset offers rich multimodal information: the robot's action data is 18-dimensional, covering the position, velocity and angle of the shoulder, elbow, wrist and gripper; the observation state is also 18-dimensional, including position, velocity and load information. Image observations are acquired from multiple perspectives: a wrist camera, two global cameras and two low-angle cameras. All images are 3-channel color videos with a resolution of 480×640, encoded using AV1. Additionally, the dataset includes metadata such as timestamps, frame indices, episode indices and task indices. The data is stored in Parquet format, while the videos are stored in MP4 format.
提供机构:
SimKienTech
创建时间:
2026-06-19
原始信息汇总

数据集概述:so101-single-arm-sort-dice-delta-v2.1

  • 许可证:Apache-2.0
  • 任务类别:机器人学(Robotics)
  • 标签:LeRobot
  • 创建工具:使用 LeRobot 创建

数据集规模与结构

  • 总片段数(Episodes):100
  • 总帧数(Frames):120,681
  • 总任务数:1
  • 总视频数:500
  • 总数据块数:1
  • 数据块大小:1000
  • 帧率(FPS):30
  • 数据分割
    • 训练集:0:100(全部数据用于训练)
  • 数据路径格式data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
  • 视频路径格式videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4

数据特征(Features)

数据集包含以下特征字段:

特征名称 数据类型 形状 说明
action float32 (18,) 包含18维动作指令,包括各关节位置、速度、角度(如肩部、肘部、腕部、夹爪)
observation.state float32 (18,) 包含18维观测状态,包括各关节位置、速度、负载(如肩部、肘部、腕部、夹爪)
observation.images.wrist video (480, 640, 3) 腕部摄像头视频,分辨率480×640,30fps,AV1编码,yuv420p格式
observation.images.global1 video (480, 640, 3) 全局摄像头1视频,规格同上
observation.images.global2 video (480, 640, 3) 全局摄像头2视频,规格同上
observation.images.low_camera1 video (480, 640, 3) 低位摄像头1视频,规格同上
observation.images.low_camera2 video (480, 640, 3) 低位摄像头2视频,规格同上
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

动作(Action)与观测状态(Observation.state)详细说明

  • 动作维度(18维)

    • 关节位置:shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
    • 关节速度:shoulder_pan.vel, shoulder_lift.vel, elbow_flex.vel, wrist_flex.vel, wrist_roll.vel, gripper.vel
    • 关节角度:shoulder_pan.angle, shoulder_lift.angle, elbow_flex.angle, wrist_flex.angle, wrist_roll.angle, gripper.angle
  • 观测状态维度(18维)

    • 关节位置:shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, gripper.pos
    • 关节速度:shoulder_pan.vel, shoulder_lift.vel, elbow_flex.vel, wrist_flex.vel, wrist_roll.vel, gripper.vel
    • 关节负载:shoulder_pan.load, shoulder_lift.load, elbow_flex.load, wrist_flex.load, wrist_roll.load, gripper.load

其他信息

  • 主页:[更多信息待补充]
  • 论文:[更多信息待补充]
  • 引用格式:[更多信息待补充]
二维码
社区交流群
二维码
科研交流群
商业服务