lyl472324464/2025-12-23-twist-one-bottle-without-rinse-merged-adjust-pickup

该数据集是使用LeRobot创建的机器人数据集，专为ALOHA机器人系统设计。它包含91个训练episodes，总计122,038帧，帧率为50fps。数据集的结构包括多个特征：机器人状态（14维浮点数组，表示左右机械臂的关节位置）、动作（14维浮点数组，控制机械臂关节）、速度（14维浮点数组，表示关节速度）、力矩（14维浮点数组，表示关节力矩），以及来自四个摄像头的图像数据（cam_high、cam_low、cam_left_wrist、cam_right_wrist），每个图像为3通道、480x640分辨率、AV1编码的视频。此外，还包括任务索引、子任务、训练标志、时间戳、帧索引、episode索引和全局索引等元数据。数据集用于机器人学习和控制任务，支持训练和评估模型。This dataset was created using LeRobot and is designed for the ALOHA robot system. It contains 91 training episodes with a total of 122,038 frames at 50fps. The dataset structure includes multiple features: robot state (14-dimensional float32 array representing joint positions for left and right arms), action (14-dimensional float32 array for controlling joints), velocity (14-dimensional float32 array for joint velocities), effort (14-dimensional float32 array for joint torques), and image data from four cameras (cam_high, cam_low, cam_left_wrist, cam_right_wrist), each as 3-channel, 480x640 resolution, AV1-encoded videos. Additionally, it includes metadata such as task index, subtask, training flag, timestamp, frame index, episode index, and global index. The dataset is intended for robot learning and control tasks, supporting model training and evaluation.