huili_shoe_insole_retrieval_and_packing_0602
收藏资源简介:
本数据集使用LeRobot工具创建,是一个机器人操作数据集,采用Apache-2.0许可证。数据采集自bi_flexiv_rizon4_4型双臂机器人,包含68个episodes,总计949,788帧数据,对应1个任务,所有数据均划分为训练集。数据以Parquet文件格式存储,总数据量约100 MB,并包含总大小约500 MB的配套视频文件,采集帧率为30 FPS。数据集提供了丰富的多模态观测和动作数据:动作空间是一个20维浮点向量,包括左右机械臂末端执行器的6维位姿(各3维位置和6维旋转表示)以及左右夹爪的位置;状态观测与动作空间具有相同的20维结构。图像观测包括来自7个摄像头的视频流:一个头部摄像头(分辨率640x480)、左右手腕摄像头(分辨率各640x480)、以及左右机械臂上的各两个触觉摄像头(共四个,分辨率均为700x400),所有视频均为彩色(3通道),使用h264编码。此外,数据还包含时间戳、帧索引、episode索引等元数据。该数据集适用于机器人模仿学习、强化学习、多模态感知与控制策略研究等任务。
This dataset was constructed using the LeRobot toolkit, and it is a robotic manipulation dataset licensed under the Apache-2.0 license. Data was collected from a bi_flexiv_rizon4_4 dual-arm robot, comprising 68 episodes and a total of 949,788 frames, corresponding to one single task. All data is partitioned into the training set. The core data is stored in Parquet file format, with a total size of approximately 100 MB, alongside supporting video files with a combined size of around 500 MB. The data acquisition frame rate is 30 FPS. The dataset offers rich multimodal observation and action data: The action space is a 20-dimensional floating-point vector, which includes the 6-dimensional poses of the left and right robotic arm end-effectors (each consisting of 3-dimensional position and 6-dimensional rotation representation) as well as the positions of the left and right grippers. The state observation shares the identical 20-dimensional structure as the action space. The image observations consist of video streams from 7 cameras: one head camera with a resolution of 640×480, left and right wrist cameras each with a resolution of 640×480, and a total of four tactile cameras (two mounted on each of the left and right robotic arms) all with a resolution of 700×400. All videos are 3-channel color footage encoded using the h264 standard. Additionally, the dataset contains metadata including timestamps, frame indices, and episode indices. This dataset is applicable to research tasks such as robotic imitation learning, reinforcement learning, multimodal perception, and control strategy development.
数据集概述
- 任务类别:机器人学(robotics)
- 许可协议:Apache-2.0
- 创建工具:使用 LeRobot 创建
数据集结构
- 机器人类型:bi_flexiv_rizon4_rt(双臂 Flexiv Rizon 4 机器人)
- 总轮次(episodes):68
- 总帧数:949,788
- 总任务数:1
- 块大小(chunks_size):1000
- 数据文件大小:约 100 MB(parquet 格式)
- 视频文件大小:约 500 MB(mp4 格式)
- 帧率(fps):30
- 数据分割:全部 68 轮用于训练(train: 0:68)
- 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
特征(Features)
动作(action)与观测状态(observation.state)
- 数据类型:float32
- 形状:
[20] - 包含维度:左右 TCP 位姿(x, y, z, r1-r6)、左右夹爪位置(pos)
观测图像(observation.images)
- 头部相机(head):分辨率 480x640,3 通道,H.264 编码,30 fps
- 左腕相机(left_wrist):分辨率 480x640,3 通道,H.264 编码,30 fps
- 右腕相机(right_wrist):分辨率 480x640,3 通道,H.264 编码,30 fps
- 左触觉传感器 0(left_tactile_0):分辨率 400x700,3 通道,H.264 编码,30 fps
- 左触觉传感器 1(left_tactile_1):分辨率 400x700,3 通道,H.264 编码,30 fps
- 右触觉传感器 0(right_tactile_0):分辨率 400x700,3 通道,H.264 编码,30 fps
- 右触觉传感器 1(right_tactile_1):分辨率 400x700,3 通道,H.264 编码,30 fps
其他特征
- 时间戳(timestamp):float32,形状
[1] - 帧索引(frame_index):int64,形状
[1] - 轮次索引(episode_index):int64,形状
[1] - 索引(index):int64,形状
[1] - 任务索引(task_index):int64,形状
[1]




