nvidia_workspace

LeRobot v3.0数据集是一个机器人任务执行数据集，记录了SO-101 follower机器人执行拿起小瓶并放入架子任务的过程。数据集规模包括27个episodes和19,913帧，仅包含一个任务和一个训练分割，采样率为30 Hz。模态为时间序列，源数据集还提供视频。源摄像头包括ego视角和外部Intel RealSense D455流，分辨率为640 x 480，编码为AV1视频，帧率为30 fps。数据集以Apache TsFile格式存储，文件为data/nvidia_workspace_train.tsfile。数据模式包含时间戳（毫秒）、episode索引、任务索引、帧索引、样本索引，以及机器人六个轴的观察状态和动作命令（浮点数）。视频特征（observation.images.ego和observation.images.external_D455）未包含在此TsFile中，但可在原始视频目录中获取。数据集基于Apache License 2.0许可证发布。The LeRobot v3.0 dataset is a robotic task execution dataset that records the process of the SO-101 follower robot completing the task of picking up a small vial and placing it onto a shelf. It comprises 27 episodes and 19,913 frames, includes only one task and one training split, with a sampling rate of 30 Hz. The dataset modality is time-series, and the source dataset also provides accompanying video. The source cameras include an ego-centric view and external Intel RealSense D455 streams, with a resolution of 640 × 480, encoded in AV1 video format at 30 fps. The dataset is stored in Apache TsFile format, with the specific file located at data/nvidia_workspace_train.tsfile. The data schema includes timestamp (in milliseconds), episode index, task index, frame index, sample index, as well as the observed states and action commands (floating-point numbers) for the robot's six axes. The video features (observation.images.ego and observation.images.external_D455) are not included in this TsFile, but are available in the original video directory. The dataset is released under the Apache License 2.0.