lerobot/berkeley_gnm_sac_son

这是一个关于机器人学的数据集，包含2955个视频片段，每个片段1000帧，总共有241059帧。数据集由一个任务类别组成，所有数据都是未经分割的训练数据。数据集的特征包括视频观测、语言指令、机器人状态、动作、时间戳、剧集索引、帧索引、下一帧奖励和完成标志。每个视频都是MP4格式，编码为av1，像素格式为yuv420p，不包含深度图和音频。

This is a robotics dataset containing 2955 video clips, each with 1000 frames, totaling 241059 frames. The dataset consists of a single task category and all data is unsplit training data. The features of the dataset include video observations, language instructions, robot states, actions, timestamps, episode indices, frame indices, next frame rewards, and done flags. Each video is in MP4 format, encoded with av1, pixel format of yuv420p, without depth maps and audio.