leokswang/23122025-foldclo-04_lerobot_format
收藏资源简介:
该数据集是使用LeRobot创建的机器人学习数据集,专门用于机器人任务。数据集基于Yam机器人类型,包含10个episodes,总计26913帧数据,帧率为30fps。数据以parquet格式存储,分为一个chunk,训练集覆盖所有episodes(0到10)。特征包括来自顶部、左侧和右侧摄像头的图像观察(每个图像尺寸为360x640x3),机器人状态观察(14维浮点数组),动作(14维浮点数组),以及时间戳、帧索引、episode索引、索引和任务索引等元数据。数据集适用于机器人控制和模仿学习任务,支持多视角视觉输入和状态-动作对记录。
This dataset is a robotics learning dataset developed using LeRobot, specifically designed for robotic tasks. Built on the Yam robot platform, it contains 10 episodes with a total of 26913 frames at a frame rate of 30 fps. The data is stored in Parquet format as a single chunk, and the training set covers all episodes ranging from 0 to 10. The dataset features visual observations from top, left, and right cameras (each with a resolution of 360×640×3), robot state observations (14-dimensional floating-point arrays), actions (14-dimensional floating-point arrays), as well as metadata including timestamps, frame indices, episode indices, item indices, and task indices. This dataset is suitable for robotic control and imitation learning tasks, supporting multi-view visual inputs and state-action pair logging.



