CoRL2026-CSI/SO101-Teleop-Open_drawer_30epi_8to2

该数据集是一个机器人数据集，专门用于模仿学习或机器人控制任务，具体任务为打开抽屉。数据集包含30个episodes，总共6846帧，帧率为30fps。数据集中记录了机器人的动作（包括肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕滚动和夹爪位置）和观察状态（关节位置），以及来自两个摄像头的视频观察：左手腕摄像头和顶部摄像头，分辨率均为480x640像素，3通道。数据集以parquet格式存储数据文件，以mp4格式存储视频文件，总数据文件大小为100MB，视频文件大小为200MB。数据集使用Apache 2.0许可证，并基于LeRobot代码库（版本v3.0）创建，机器人类型为so_follower。数据集仅包含训练分割，涵盖所有episodes。This dataset is a robotics dataset designed for imitation learning or robot control tasks, specifically for opening a drawer. It contains 30 episodes with a total of 6846 frames at 30 fps. The dataset records robot actions (including shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions) and observation states (joint positions), along with video observations from two cameras: a left wrist camera and a top camera, both with a resolution of 480x640 pixels and 3 channels. The dataset is stored in parquet format for data files and mp4 format for video files, with a total data file size of 100MB and video file size of 200MB. It is licensed under Apache 2.0 and created using the LeRobot codebase (version v3.0), with a robot type of so_follower. The dataset includes only a training split covering all episodes.