notstrober/desk-stationery-to-drawer-bimanual_v3

该数据集是一个机器人操作数据集，专门用于双臂机器人（robot_type: bi_so_follower）执行任务。数据集包含91个episodes，总计201,819帧，帧率为30fps。数据以parquet文件和视频文件形式存储，总数据文件大小为100MB，视频文件大小为200MB。数据集特征包括：动作数据（action），由12个浮点数组成，表示左右机械臂的关节位置（如左肩平移、左肩提升、左肘弯曲等）；观察状态（observation.state），同样为12个浮点数的关节位置；以及来自三个摄像头的图像观察（observation.images.left_front、right_front、right_top），每个视频分辨率为480x640，3通道彩色，使用h264编码。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任务，如模仿学习或强化学习，用于训练双臂机器人执行桌面文具到抽屉的双手操作任务。This dataset is a robotics manipulation dataset designed for bimanual robots (robot_type: bi_so_follower) to perform tasks. It contains 91 episodes with a total of 201,819 frames at 30fps. The data is stored in parquet and video files, with a total data file size of 100MB and video file size of 200MB. Dataset features include: action data, consisting of 12 float values representing joint positions for both left and right robot arms (e.g., left shoulder pan, left shoulder lift, left elbow flex, etc.); observation state, similarly with 12 float joint positions; and image observations from three cameras (observation.images.left_front, right_front, right_top), each video with a resolution of 480x640, 3 color channels, encoded in h264. Additionally, metadata such as timestamp, frame index, and episode index are included. The dataset is suitable for robotics learning tasks, such as imitation learning or reinforcement learning, for training bimanual robots to perform desk stationery-to-drawer manipulation tasks.