axiboai/piper-stack-scripted-pi0fast-v4

该数据集是使用LeRobot创建的机器人数据集，包含200个episodes，总计125458帧数据，帧率为20fps。数据以parquet格式存储，分为训练集（全部200个episodes）。数据集特征包括：三个摄像头（camera1、camera2、camera3）的视频观测，每个视频分辨率为128x128，3通道，编码为h264；观测状态（observation.state）为30维浮点数组，包含关节位置、速度及夹爪状态；末端执行器姿态（observation.ee_pose）为14维浮点数组，表示左右夹爪的位置和四元数；动作（action）为14维浮点数组，表示关节目标位置。此外，还包含元数据字段，如演示者ID、工作站ID、任务变体、毛巾ID、光照条件、垫子ID、成功标志、笔记等，以及时间戳、帧索引、episode索引等辅助信息。数据集用于机器人学习任务，但具体任务内容未明确说明。

This dataset is a robotics dataset created using LeRobot, containing 200 episodes with a total of 125458 frames at 20fps. The data is stored in parquet format, with a training split covering all 200 episodes. Dataset features include: video observations from three cameras (camera1, camera2, camera3), each with a resolution of 128x128, 3 channels, encoded as h264; observation state (observation.state) as a 30-dimensional float array, including joint positions, velocities, and gripper states; end-effector pose (observation.ee_pose) as a 14-dimensional float array, representing left and right gripper positions and quaternions; action (action) as a 14-dimensional float array, indicating joint target positions. Additionally, metadata fields are included, such as demonstrator ID, station ID, task variant, towel ID, lighting conditions, mat ID, success flag, notes, etc., along with auxiliary information like timestamp, frame index, episode index. The dataset is intended for robotics learning tasks, though the specific task content is not explicitly stated.