pick_two

该数据集由LeRobot创建，专用于机器人学任务，特别是机器人操作。它包含34个episodes，总计14497帧，数据文件大小为100MB，视频文件大小为200MB。数据以30fps的帧率采集，特征包括动作（用6个浮点数表示关节位置，如肩部、肘部、腕部等）、观察状态（同样用6个浮点数表示相同关节位置）、两个视频观察（来自头顶和手腕摄像头，分辨率为480x640，3通道，编码为av1），以及时间戳、帧索引、episode索引等元数据。数据集格式为parquet和mp4，机器人类型为so_follower，适用于训练和评估机器人控制模型。This dataset is created by LeRobot for robotics tasks, specifically robot manipulation. It includes 34 episodes with a total of 14497 frames, a data file size of 100MB, and a video file size of 200MB. Data is collected at 30fps, featuring actions (6 floating-point numbers representing joint positions such as shoulder, elbow, wrist), observation states (6 floating-point numbers for the same joint positions), two video observations (overhead and wrist cameras with 480x640 resolution, 3 channels, encoded as av1), and metadata like timestamps, frame indices, and episode indices. The dataset format is parquet and mp4, with robot type so_follower, suitable for training and evaluating robot control models.