edmos7/ball_in_the_box-trajectory

该数据集由LeRobot项目创建，用于机器人学任务，包含50个训练片段，总帧数为43930。数据集使用so100机器人类型，数据以parquet格式存储，视频以mp4格式存储。特征包括：动作（action）和观测状态（observation.state），均为6维浮点数组，对应机器人的关节控制（如主肩部平移、主肩部提升、主肘部弯曲、主腕部弯曲、主腕部滚动和主夹持器）；时间戳（timestamp）、帧索引（frame_index）、片段索引（episode_index）、索引（index）和任务索引（task_index）；以及多个图像观测，如MAIN_CAM和SEC_CAM摄像头视频（分辨率480x640，30fps，AV1编码）和轨迹图像（trajectory）。数据集适用于机器人控制、视觉导航和轨迹学习等任务。

This dataset was created using the LeRobot project for robotics tasks, containing 50 training episodes with a total of 43930 frames. It uses the so100 robot type, with data stored in parquet format and videos in mp4 format. Features include: action and observation.state, both as 6-dimensional float32 arrays corresponding to robot joint controls (e.g., main_shoulder_pan, main_shoulder_lift, main_elbow_flex, main_wrist_flex, main_wrist_roll, main_gripper); timestamp, frame_index, episode_index, index, and task_index; and multiple image observations, such as MAIN_CAM and SEC_CAM camera videos (resolution 480x640, 30fps, AV1 codec) and trajectory images. The dataset is suitable for tasks like robot control, visual navigation, and trajectory learning.