leokswang/23122025-foldclo-03_lerobot_format
收藏资源简介:
该数据集使用LeRobot工具创建,专注于机器人学领域,具体针对yam机器人类型。数据集包含10个完整的情节(episodes),总计18972帧,覆盖1个任务,无视频数据,以30帧每秒的速率采集。数据被分割为1个数据块,每个块大小为1000,训练集包含所有10个情节。数据以parquet格式存储,路径模式为data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet。数据集的特征包括:来自顶部、左侧和右侧摄像头的观测图像(每张图像尺寸为360x640x3),观测状态(14维浮点数组),动作(14维浮点数组),以及时间戳、帧索引、情节索引、索引和任务索引等元数据。这些数据适用于机器人控制、行为克隆或强化学习等任务。
This dataset was created using the LeRobot tool, focusing on the robotics domain, specifically for the yam robot type. It contains 10 complete episodes, totaling 18972 frames, covering 1 task, with no video data, captured at 30 frames per second. The data is split into 1 chunk with a chunk size of 1000, and the training set includes all 10 episodes. Data is stored in parquet format with the path pattern data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet. The dataset features include: observation images from top, left, and right cameras (each image sized 360x640x3), observation state (14-dimensional float array), action (14-dimensional float array), and metadata such as timestamp, frame index, episode index, index, and task index. This data is suitable for tasks like robot control, behavior cloning, or reinforcement learning.



