HyeonseokE/teleop_pnp_80epi
收藏资源简介:
该数据集是一个用于机器人学领域的数据集,专门针对so_follower类型的机器人。它包含80个完整的情节(episodes),总计28,456帧数据,所有数据均用于训练(划分比例为0:80)。数据以Parquet文件格式存储,总数据文件大小为100MB,视频文件大小为200MB。数据集的特征包括:动作(action)和观察状态(observation.state),均为6维浮点数组,表示机器人的关节位置(如肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部滚动和夹爪位置)。观察部分还包括两个摄像头视角的图像数据:左腕摄像头(observation.images.left_wrist)和顶部摄像头(observation.images.top),均为480x640分辨率、3通道的RGB视频,帧率为30fps,使用AV1编解码器。此外,数据集还包含时间戳(timestamp)、帧索引(frame_index)、情节索引(episode_index)、索引(index)和任务索引(task_index)等元数据。数据集基于LeRobot平台创建,适用于机器人控制和学习任务。
This dataset is designed for the robotics domain, specifically for the so_follower robot type. It comprises 80 complete episodes, totaling 28,456 frames, all allocated for training (split ratio 0:80). The data is stored in Parquet format, with a total data file size of 100MB and video file size of 200MB. Key features include: action and observation.state, both as 6-dimensional float32 arrays representing robot joint positions (e.g., shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions). Observations also include image data from two camera perspectives: left wrist camera (observation.images.left_wrist) and top camera (observation.images.top), both as RGB videos with 480x640 resolution, 3 channels, 30fps frame rate, using the AV1 codec. Additionally, the dataset contains metadata such as timestamp, frame_index, episode_index, index, and task_index. Created using the LeRobot platform, it is suitable for robot control and learning tasks.




