grab-cube

该数据集是一个机器人相关的数据集，包含50个episodes，35753帧，30fps的视频数据。数据集提供了机器人的动作和状态观测数据，包括肩部、肘部、腕部和夹持器的位置信息。视频数据的分辨率为480x640，格式为av1编码的yuv420p。数据集结构包含多个特征，如动作、观测状态、时间戳、帧索引等。

This is a robotics-oriented dataset that includes 50 episodes and 35,753 frames of video data captured at 30 frames per second. The dataset provides robot motion and state observation data, covering position information of the shoulder, elbow, wrist, and gripper. The video data has a resolution of 480×640 and is encoded in AV1 with the yuv420p pixel format. The dataset structure contains multiple features such as actions, observation states, timestamps, frame indices, and so on.