base4-clean-table-02-BC
收藏资源简介:
该数据集使用LeRobot工具创建,遵循Apache 2.0许可证,专注于机器人学领域,针对u850机器人平台。数据集包含50个完整情节,总计108,534帧数据,对应150个视频文件,以1000帧为块存储,采样频率为50帧每秒。数据集仅提供训练集划分(0:50)。核心内容为机器人交互数据,包括多模态观测和对应动作:观测部分包含机器人状态(10维浮点向量,表示7个关节位置和基座的x、y、yaw坐标)以及来自三个摄像头的视觉信息(左眼、右眼和腕部摄像头,RGB格式,分辨率480x640,帧率50fps);动作数据为10维浮点向量,控制机器人的7个关节和夹爪。此外,数据集还包含每帧的索引、时间戳、情节是否结束的标志、是否为纠正动作的标志以及任务索引等元数据。该数据集适用于机器人模仿学习、强化学习、视觉运动控制等任务的研究与模型训练。
This dataset is created using the LeRobot tool, follows the Apache 2.0 license, and focuses on the field of robotics, specifically targeting the u850 robot platform. It contains 50 complete episodes, totaling 108,534 frames of data, corresponding to 150 video files. The data is stored in chunks of 1000 frames each, with a sampling frequency of 50 frames per second. Only training set splits (0:50) are provided. The core content consists of robot interaction data, including multimodal observations and corresponding actions: observations include robot state (a 10-dimensional floating-point vector representing 7 joint positions and the bases x, y, yaw coordinates) and visual information from three cameras (left eye, right eye, and wrist camera, in RGB format with 480x640 resolution and 50fps frame rate); action data is a 10-dimensional floating-point vector controlling the robots 7 joints and gripper. Additionally, the dataset includes metadata such as frame index, timestamp, episode termination flag, correction action flag, and task index. It is suitable for research and model training in tasks like robot imitation learning, reinforcement learning, and visual-motor control.




