SO101-record

该数据集由LeRobot创建，主要用于机器人技术领域。数据集采用Apache-2.0许可证，包含parquet格式的数据文件。数据集结构包括1个episode、3571帧、1个任务，数据文件大小为100MB，视频文件大小为200MB，帧率为30fps。数据按1000帧的块进行组织，训练集包含所有数据。数据集包含多个特征字段，如动作（6个浮点数值，对应机器人关节位置）、观察状态（同样6个浮点数值）、来自主摄像头和腕部摄像头的视频观察（240x320分辨率，3通道，30fps，AV1编码）、时间戳、帧索引、episode索引等。视频数据包含详细的元信息，如编解码器、像素格式、是否深度图等。数据集适用于机器人控制、行为模仿等任务。

This dataset was created by LeRobot, primarily targeting the robotics domain. It is licensed under the Apache-2.0 open-source license and includes data files in Parquet format. The dataset structure consists of 1 episode, 3571 frames, and 1 task. The total size of the data files is 100 MB, while the video files are 200 MB with a frame rate of 30 fps. The data is organized into chunks of 1000 frames each, and the training set includes all available data. The dataset contains multiple feature fields, including actions (6 floating-point values corresponding to robot joint positions), observation states (also 6 floating-point values), video observations from the main camera and wrist camera (240×320 resolution, 3 channels, 30 fps, AV1 encoded), timestamps, frame indices, episode indices, and others. The video data includes detailed metadata such as codec, pixel format, whether depth maps are included, and other relevant information. This dataset is suitable for tasks such as robot control and behavior cloning.