HyeonseokE/SO101-cap_sort_RGBblock_to_matchingplate_10fps_40epi
收藏资源简介:
该数据集是一个机器人数据集,使用LeRobot创建,专门用于机器人任务。数据集包含40个episodes,总计34886帧,涉及1个任务,数据以parquet文件格式存储,视频以MP4格式存储,帧率为10fps。数据集基于so101_follower机器人类型,特征包括机器人的状态观察(如6个关节位置:肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置)、动作控制、图像观察(顶部和左腕摄像头,分辨率480x640,RGB通道)、末端执行器位置(XYZ坐标和欧拉角)、夹爪二进制状态、技能信息(如自然语言描述、验证问题、类型、进度和目标位置)以及子任务信息(如自然语言描述、对象名称和目标位置)。此外,还包括时间戳、帧索引、episode索引等元数据。数据集适用于机器人学习和控制任务,如模仿学习或强化学习。
This is a robotics dataset developed with LeRobot, specifically tailored for robotic tasks. It contains 40 episodes, totaling 34,886 frames, covering one single task. The dataset is stored in Parquet files, while the associated videos are saved in MP4 format at a frame rate of 10 fps. This dataset is based on the so101_follower robot type. Its features include robot state observations (e.g., 6 joint positions: shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position), action controls, image observations from the top and left-wrist cameras with a resolution of 480×640 and RGB channels, end-effector positions (XYZ coordinates and Euler angles), binary gripper state, skill information (such as natural language descriptions, validation questions, skill types, progress levels, and target positions), and subtask information (including natural language descriptions, object names, and target positions). Additionally, it includes metadata such as timestamps, frame indices, and episode indices. This dataset is applicable to robotics learning and control tasks, such as imitation learning or reinforcement learning.




