five

real_world_force_image_1_2

收藏
Hugging Face2026-05-28 更新2026-05-28 收录
下载链接:
https://huggingface.co/datasets/yinongh/real_world_force_image_1_2
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任务数据集,使用LeRobot工具创建,专为机器人学习任务设计。数据集包含10个总集数、540个总帧数、40个总视频数,以及1个总任务数,数据以parquet格式存储,视频以mp4格式存储。数据特征包括动作数据(如位置和旋转变化,共9个浮点数维度)、观测数据(如末端执行器内力、姿态矩阵、深度图像、点云数据、相机外参矩阵,以及来自多个相机(如手腕相机和Azure Kinect相机)的RGB和深度图像)。数据集结构详细定义了每个特征的形状和类型,例如动作的dtype为float32,形状为[9];观测中的深度图像形状为[480,640]或[720,1280],RGB图像形状为[720,1280,3]。数据集以30帧每秒的速率采集,适用于机器人控制、视觉感知和强化学习等应用场景。

This dataset is a robotic task dataset created using the LeRobot tool, specifically designed for robotic learning tasks. It includes 10 total episodes, 540 total frames, 40 total videos, and 1 single task. The dataset’s data is stored in Parquet format, while the accompanying videos are saved in MP4 format. The dataset features two core categories of data: action data and observation data. The action data encompasses position and rotation changes, with a total of 9 float-valued dimensions. The observation data covers end-effector internal force, pose matrices, depth images, point cloud data, camera extrinsic matrices, as well as RGB and depth images captured by multiple cameras such as the wrist camera and Azure Kinect camera. The dataset structure explicitly defines the shape and data type of each feature. For example, the action data has a dtype of float32 and a shape of [9]; the depth images in the observations have shapes of [480, 640] or [720, 1280], while the RGB images have a shape of [720, 1280, 3]. This dataset was collected at a frame rate of 30 frames per second, and is suitable for application scenarios such as robotic control, visual perception, and reinforcement learning.
提供机构:
yinongh
创建时间:
2026-05-28
原始信息汇总

数据集概述

  • 数据集名称: yinongh/real_world_force_image_1_2
  • 许可证: Apache-2.0
  • 任务类别: 机器人学(robotics)
  • 标签: LeRobot
  • 创建工具: 基于 LeRobot 框架创建

数据集结构

基本统计信息

  • 代码库版本: v2.1
  • 机器人类型: script(脚本控制)
  • 总片段数(episodes): 10
  • 总帧数: 540
  • 总任务数: 1
  • 总视频数: 40
  • 总数据块数: 1
  • 数据块大小: 1000
  • 帧率(FPS): 30
  • 数据划分:
    • 训练集(train): 片段索引 0-9(全部10个片段)

数据文件结构

  • 数据文件路径: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
  • 视频文件路径: videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4

特征(Features)

特征名称 数据类型 形状 说明
action float32 [9] 动作向量,包含:delta_x, delta_y, delta_z, 以及6维旋转表示(delta_rot6d_0 至 delta_rot6d_5)
observation.eef_internal_forces float32 [3] 末端执行器内部力,映射到腕部相机坐标系(IsaacGym x/z轴约定),包含:isaacgym_cam_wrist_fx, isaacgym_cam_wrist_fy, isaacgym_cam_wrist_fz
observation.eef_pose float32 [4, 4] 末端执行器位姿矩阵(4x4),行列命名
observation.aligned_socket_depth_img float32 [480, 640] 通过旋转相反当前EEF偏航变化对齐的插座深度图像
observation.init_plug_depth_img float32 [480, 640] 从辅助点云渲染的初始插头深度图像
observation.points.initial_wrist_points_world pcd [-1, 3] 初始腕部在世界坐标系中的点云数据
observation.cam_wrist.extrinsics float32 [4, 4] 腕部相机外参矩阵(T_world_cam)
observation.images.cam_wrist.depth float32 [720, 1280] 来自腕部ZED Mini的FoundationStereo深度图,单位米
observation.images.cam_wrist video [720, 1280, 3] 腕部相机RGB彩色图像
observation.images.cam_azure_kinect_front.color video [720, 1280, 3] Azure Kinect前置相机的原始彩色图像
observation.images.cam_auxiliary.left video [720, 1280, 3] 脚本设置期间一次性捕获的辅助ZED左RGB图像
observation.images.cam_auxiliary.depth video [720, 1280, 1] 一次性辅助FoundationStereo深度图,单位为uint16毫米
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

注意事项

  • 数据集配置: 默认配置名称为 default,数据文件匹配模式为 data/*/*.parquet
  • 引用信息: 当前未提供详细的引用文献(BibTeX 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