burger-to-plate
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot工具创建,专门用于机器人学习任务。数据集包含名为burger-to-plate的任务,提供60个完整episodes的演示数据,共计91,719帧,所有数据均划分为训练集。数据采集自blueberry_ros机器人平台,帧率为15 fps。数据集包含多模态信息:1. 动作(action):26维连续控制向量,包括机器人左右手臂的线速度和角速度、左右手各手指的位置控制以及底盘的操纵杆输入。2. 状态观测(observation.state):55维向量,包括左右手臂7个关节和左右手各手指的位置与力矩反馈,以及用户视线坐标(x, y)和有效性标志。3. 图像观测(observation.images):包含四个固定视角的同步视频流,分别是机器人左眼、右眼、用户视角和用户视线视角,视频分辨率为640x480(宽x高),3通道彩色,采用AV1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、行为克隆等任务,特别是涉及双臂灵巧操作和视觉感知的研究。数据集以Parquet格式存储结构化数据,以MP4格式存储视频数据。
This dataset is a robotic manipulation dataset created using the LeRobot tool, specifically tailored for robotic learning tasks. It features the `burger-to-plate` task, providing demonstration data from 60 full episodes, totaling 91,719 frames, with all data allocated to the training set. The data was collected on the blueberry_ros robotic platform, with a frame rate of 15 fps. The dataset includes multi-modal information: 1. Action: a 26-dimensional continuous control vector, covering the linear and angular velocities of the robot's left and right arms, the position controls of the left and right hand fingers respectively, and the joystick input of the robot's chassis. 2. State observation (observation.state): a 55-dimensional vector, including the position and torque feedback of the 7 joints of the left and right arms, as well as the position and torque feedback of the left and right hand fingers, plus the user's gaze coordinates (x, y) and validity flag. 3. Image observation (observation.images): contains four synchronized video streams from fixed perspectives, namely the robot's left eye, right eye, first-person user perspective, and user gaze perspective. The videos have a resolution of 640×480 (width × height), 3-channel color, and are encoded in AV1. Additionally, the dataset contains metadata such as timestamps, frame indices, and episode indices. This dataset is applicable to tasks including robotic imitation learning, reinforcement learning, and behavioral cloning, particularly for research involving dual-arm dexterous manipulation and visual perception. Structured data of the dataset is stored in Parquet format, while video data is stored in MP4 format.




