JunnDooChoi/robotwin_unified

该数据集使用LeRobot创建，是一个机器人学数据集，专门用于机器人控制任务。数据集包含27,500个episodes，总计6,075,103帧，涵盖23,559个任务。数据以parquet文件格式存储，视频文件为mp4格式，帧率为30fps。数据集特征包括：机器人状态观测（observation.state），为14维浮点数，对应左右机械臂的14个电机位置（如左腰、左肩、左肘、左前臂滚转、左手腕角度、左手腕旋转、左夹持器，以及右侧对应部件）；动作（action），同样为14维浮点数，控制这些电机；三个摄像头图像观测：高角度摄像头（observation.images.cam_high）、左手腕摄像头（observation.images.cam_left_wrist）和右手腕摄像头（observation.images.cam_right_wrist），均为480x640x3的RGB视频。此外，还包括时间戳、帧索引、episode索引、索引和任务索引等元数据。机器人类型为“aloha”。数据集适用于机器人学习、控制算法开发和仿真研究。This dataset was created using LeRobot and is a robotics dataset designed for robot control tasks. It contains 27,500 episodes, totaling 6,075,103 frames, and covers 23,559 tasks. The data is stored in parquet files, with video files in mp4 format at 30fps. The dataset features include: robot state observations (observation.state) as 14-dimensional float32 values corresponding to 14 motor positions of the left and right arms (e.g., left waist, left shoulder, left elbow, left forearm roll, left wrist angle, left wrist rotate, left gripper, and their right-side counterparts); actions (action) as 14-dimensional float32 values controlling these motors; three camera image observations: a high-angle camera (observation.images.cam_high), left wrist camera (observation.images.cam_left_wrist), and right wrist camera (observation.images.cam_right_wrist), all as 480x640x3 RGB videos. Additionally, metadata such as timestamp, frame index, episode index, index, and task index are included. The robot type is aloha. The dataset is suitable for robot learning, control algorithm development, and simulation research.