JonathanGiegold/cylinder-in-hole-70-annotated
收藏资源简介:
该数据集是一个用于机器人任务的数据集,专门针对圆柱体插入孔洞的操作。它包含70个训练片段,总帧数达45853,以每秒30帧的速率采集。数据集的结构包括动作特征(如肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置)、观察状态特征(与动作相同的关节位置)、来自正面和侧面摄像头的视频观察(分辨率为480x640,3通道RGB),以及时间戳、帧索引、片段索引、任务索引等元数据。机器人类型为so_follower,数据以Parquet文件格式存储,视频文件使用AV1编解码器。数据集主要用于机器人学习和控制任务,支持从视觉和状态信息中学习动作策略。
This dataset is designed for robotics tasks, specifically focusing on cylinder-in-hole operations. It includes 70 training episodes with a total of 45,853 frames, captured at 30 frames per second. The dataset structure comprises action features (such as shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions), observation state features (identical joint positions as actions), video observations from front and side cameras (resolution 480x640, 3-channel RGB), and metadata like timestamp, frame index, episode index, and task index. The robot type is so_follower, with data stored in Parquet files and video files using the AV1 codec. The dataset is primarily used for robot learning and control tasks, enabling the learning of action policies from visual and state information.




