tersoohaan/eval_pi0_pretrained

该数据集使用LeRobot创建，是一个机器人学数据集，专门针对so101_follower机器人类型。数据集包含15个完整的情节（episodes），总计9094帧，任务数为1。数据以分块形式存储，每个块大小为1000帧，数据文件总大小为100MB，视频文件总大小为500MB，帧率为30fps。数据集仅包含训练集，涵盖从0到15的情节。特征包括动作（action）和观察（observation），其中动作和状态观察均包含6个关节位置（如肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置）。观察部分还包括来自前置摄像头和顶部摄像头的图像数据，图像分辨率为480x640，3通道，视频编码为av1，无音频。此外，还提供前置摄像头的深度图像（uint16类型），以及时间戳、帧索引、情节索引、索引和任务索引等元数据。数据集结构设计用于支持机器人控制和学习任务，特别是基于视觉的机器人操作。

This dataset was created using LeRobot and is a robotics dataset specifically for the so101_follower robot type. It contains 15 total episodes with 9094 total frames and 1 total task. The data is stored in chunks of 1000 frames, with data files totaling 100MB and video files totaling 500MB, at a frame rate of 30fps. The dataset includes only a training set, covering episodes from 0 to 15. Features include action and observation, where both action and state observation consist of 6 joint positions (e.g., shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, gripper). Observations also include image data from front and top cameras with a resolution of 480x640, 3 channels, encoded in av1 video format without audio. Additionally, depth images from the front camera (uint16 type) are provided, along with metadata such as timestamp, frame index, episode index, index, and task index. The dataset structure is designed to support robot control and learning tasks, particularly for vision-based robot manipulation.