five

sobit_home-pickup_the_block-left-gazebo

收藏
Hugging Face2026-05-27 更新2026-05-27 收录
下载链接:
https://huggingface.co/datasets/team-sobits/sobit_home-pickup_the_block-left-gazebo
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集是一个机器人数据集,专门用于移动机械臂sobit_home在Gazebo模拟环境中执行pickup_the_block-left(拾取左侧方块)任务。它包含202个episodes,总计19464帧,以6fps的帧率收集。数据以parquet格式存储,视频以MP4格式存储。数据集的特征包括动作(action)、动作增量(action.delta)、观察状态(observation.state)、末端执行器姿态(observation.ee_pose.left)、来自头部相机和左手相机的图像(observation.images.head_camera和observation.images.hand_left_camera),以及时间戳、帧索引、episode索引等元数据。机器人信息详细描述了其形态结构,包括头部、身体、左右臂、左右末端执行器和移动基座,传感器信息包括RGB相机(头部相机、左右手相机)。数据集由Keith Valentin在Soka University, Japan创建,使用Apache 2.0许可证。

This dataset is a robot dataset specifically designed for the mobile manipulator sobit_home to perform the pickup_the_block-left (pick up the left block) task in the Gazebo simulation environment. It contains 202 episodes, totaling 19464 frames, collected at a frame rate of 6fps. The data is stored in parquet format, and videos are stored in MP4 format. The dataset features include action, action delta, observation state, end-effector pose (observation.ee_pose.left), images from the head camera and left-hand camera (observation.images.head_camera and observation.images.hand_left_camera), as well as metadata such as timestamps, frame indices, and episode indices. The robot information details its morphological structure, including the head, body, left and right arms, left and right end-effectors, and mobile base, with sensor information including RGB cameras (head camera, left and right hand cameras). The dataset was created by Keith Valentin at Soka University, Japan, and uses the Apache 2.0 license.
提供机构:
team-sobits
创建时间:
2026-05-27
原始信息汇总

数据集概述

  • 数据集名称: sobit_home-pickup_the_block-left-gazebo
  • 创建者: Keith Valentin(日本创价大学)
  • 许可证: Apache-2.0
  • 任务类别: 机器人学(Robotics)
  • 数据格式: LeRobot 格式(Parquet 文件 + 视频文件)
  • 机器人类型: 移动操作臂(Mobile Manipulator)
  • 机器人名称: sobit_home

数据集规模

  • 总片段数: 202 个片段(全部用于训练)
  • 总帧数: 19,464 帧
  • 总任务数: 1 个任务
  • 帧率 (FPS): 6
  • 数据文件大小: 100 MB
  • 视频文件大小: 200 MB

数据集结构

特征

特征名称 数据类型 形状 描述
action float32 [19] 包含 19 维动作指令(头、身体、左臂、左手手指、移动基座)
action.is_fresh bool [19] 指示每个动作维度是否为新值
action.delta float32 [19] 动作的差分值
action.delta.is_fresh bool [19] 指示每个动作差分维度是否为新值
observation.state float32 [19] 19 维机器人状态(关节角度 + 基座位置)
observation.state.is_fresh bool [19] 指示每个状态维度是否为新值
action.base float32 [3] 移动基座动作(base_x, base_y, base_theta)
action.base.is_fresh bool [3] 指示每个基座动作维度是否为新值
observation.ee_pose.left float32 [6] 左机械臂末端执行器位姿(x, y, z, roll, pitch, yaw)
observation.ee_pose.left.delta float32 [6] 左末端执行器位姿的差分
observation.images.head_camera video [3, 480, 640] 头部摄像头 RGB 视频(H.264, 6 FPS, 640x480)
observation.images.hand_left_camera video [3, 1200, 1920] 左手摄像头 RGB 视频(H.264, 6 FPS, 1920x1200)
timestamp float32 [1] 时间戳
frame_index int64 [1] 帧索引
episode_index int64 [1] 片段索引
index int64 [1] 总索引
task_index int64 [1] 任务索引

动作维度(19维)

  1. 头部关节: head_pan_joint, head_tilt_joint
  2. 身体关节: body_lift_joint
  3. 左臂关节: arm_left_shoulder_tilt_joint, arm_left_upper_roll_joint, arm_left_upper_flex_joint, arm_left_elbow_joint, arm_left_wrist_tilt_joint, arm_left_wrist_roll_joint
  4. 左手手指关节: hand_left_finger_l_mcp_joint, hand_left_finger_l_pip_joint, hand_left_finger_l_dip_joint, hand_left_finger_c_mcp_joint, hand_left_finger_c_ip_joint, hand_left_finger_r_pip_joint, hand_left_finger_r_dip_joint
  5. 移动基座: base_x, base_y, base_theta

传感器

  • 头部摄像头: Orbbec 336L, 640x480, 6 FPS
  • 左手摄像头: ELP USBGS1200P01 H120 J, 1920x1200, 6 FPS
  • 右手摄像头(未在数据特征中体现,但在机器人信息中列出): ELP USBGS1200P01 H120 J, 1920x1200

机器人形态部件

  • head(头部)
  • body(身体)
  • arm_left(左臂)
  • arm_right(右臂)
  • end_effector_left(左端执行器)
  • end_effector_right(右端执行器)
  • mobile_base(移动基座)
二维码
社区交流群
二维码
科研交流群
商业服务