indory_xlerobot_pick_delivery
收藏资源简介:
该数据集是一个用于机器人学习任务的演示数据集,使用LeRobot平台创建,包含30个完整的情节(episodes),共计30,287帧数据,涵盖3种不同的任务。数据采用分块存储,每块包含1000帧,总大小约为300MB(其中数据文件100MB,视频文件200MB),以15fps的帧率采集,仅提供训练集。数据内容包含多模态信息:动作(action)为17维浮点向量,控制双机械臂(左/右各6个关节+夹爪)和头部(2个电机)的运动及底盘速度(x, y, theta方向);状态观测(observation.state)为与动作相同的17维浮点向量,表示机器人的当前关节和速度状态;视觉观测(observation.images)包含三路RGB视频流,分别来自头部摄像头、左腕摄像头和右腕摄像头,分辨率均为640x480,采用H.264编码;索引信息包括时间戳、帧索引、情节索引、全局索引和任务索引。该数据集适用于机器人模仿学习、强化学习、多模态感知与控制等研究任务,特别针对双机械臂移动机器人的操作技能学习。
This is a demonstration dataset for robotic learning tasks, created using the LeRobot platform. It contains 30 full episodes, totaling 30,287 frames of data and covering 3 distinct tasks. The data is stored in chunks, with each chunk containing 1000 frames. The total size of the dataset is approximately 300 MB (100 MB for data files and 200 MB for video files), collected at a frame rate of 15 fps, and only the training split is provided. The dataset includes multimodal information: the action is a 17-dimensional floating-point vector, which controls the movements of the dual robotic arms (6 joints plus gripper for each left and right arm), the head unit (2 motors), and the base velocity in x, y, and theta directions; the state observation (observation.state) is a 17-dimensional floating-point vector matching the action space, representing the robot's current joint positions and velocity states; the visual observation (observation.images) consists of three RGB video streams sourced from the head camera, left wrist camera, and right wrist camera respectively, all with a resolution of 640×480 and encoded using H.264; the accompanying index information includes timestamps, frame indices, episode indices, global indices, and task indices. This dataset is applicable to research tasks including robotic imitation learning, reinforcement learning, multimodal perception and control, and is specifically designed for the learning of manipulation skills for mobile robots equipped with dual robotic arms.
数据集概述
- 数据集名称:
hanbin5/indory_xlerobot_pick_delivery - 许可协议: Apache-2.0
- 任务类别: 机器人学 (Robotics)
- 相关工具: 该数据集使用 LeRobot 创建,并支持通过 可视化空间 预览数据。
数据集结构
- 机器人类型:
xlerobot_client - 总集数: 30 集
- 总帧数: 30,287 帧
- 总任务数: 3 个任务
- 帧率: 15 FPS
- 数据文件大小: 约 100 MB
- 视频文件大小: 约 200 MB
- 数据分块: 每块 1000 帧
- 数据分割: 全部 30 集用于训练 (
train: "0:30")
数据文件路径
- 数据文件路径格式:
data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet - 视频文件路径格式:
videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4
数据特征
动作与状态特征 (action 与 observation.state)
两者均为 17 维的 float32 数组,包含以下关节状态:
- 左臂: 肩关节旋转、肩关节俯仰、肘关节弯曲、腕关节弯曲、腕关节旋转、夹爪
- 右臂: 肩关节旋转、肩关节俯仰、肘关节弯曲、腕关节弯曲、腕关节旋转、夹爪
- 头部: 电机 1、电机 2
- 移动: x 方向速度、y 方向速度、角度速度
图像观察特征 (observation.images)
包含三个摄像头视角,均为视频类型,分辨率为 480x640,3 通道,H.264 编码,15 FPS:
- 头部摄像头:
head - 左手腕摄像头:
left_wrist - 右手腕摄像头:
right_wrist
其他特征
timestamp: 时间戳,float32,形状[1]frame_index: 帧索引,int64,形状[1]episode_index: 集索引,int64,形状[1]index: 数据索引,int64,形状[1]task_index: 任务索引,int64,形状[1]




