HyeonseokE/SO101-DistributeChoco_Ours_20epi
收藏资源简介:
该数据集是一个用于机器人技能学习的机器人数据集,基于LeRobot平台创建。数据集包含20个完整的情节(episodes),总计9530帧数据,涉及1个任务。数据以parquet格式存储,总数据文件大小为100MB,视频文件大小为200MB,帧率为10fps。数据集的特征丰富,包括观测状态(如6个关节的位置、末端执行器的XYZRPY坐标、二进制夹爪状态)、动作(6个关节的位置控制)、技能信息(如自然语言描述、验证问题、技能类型、进度、目标位置等)、子任务信息(如自然语言描述、对象名称、目标位置)以及时间戳和索引信息。观测图像包括顶部摄像头和左腕摄像头的视频数据,分辨率为480x640,格式为h264编码的MP4视频。数据集适用于机器人控制、模仿学习、强化学习等研究,特别关注技能学习和任务分解。
This robotic dataset is developed for robot skill learning and built on the LeRobot platform. It contains 20 complete episodes, totaling 9530 frames of data, corresponding to one single task. The non-video data is stored in Parquet format, with a total size of 100 MB, while the video files occupy 200 MB and run at a frame rate of 10 fps. The dataset features rich modalities, including observation states (e.g., positions of 6 joints, XYZRPY coordinates of the end-effector, binary gripper state), actions (position control commands for 6 joints), skill-related metadata (e.g., natural language descriptions, validation questions, skill types, progress status, target positions), subtask information (e.g., natural language descriptions, object names, target positions), as well as timestamps and index information. The observation images include video data captured by the top-down camera and the left-wrist camera, with a resolution of 480×640 and encoded in H.264 format as MP4 videos. This dataset is applicable to research in robot control, imitation learning, reinforcement learning and other related domains, with a particular emphasis on skill learning and task decomposition.




