jae0311/serving_a

该数据集是LeRobot v3.0版本，专为OpenManipulator-X (OMX)机器人设计的咖啡厅服务任务（拾取和放置）数据集，属于GR00T N1.7微调数据集的Category A（长片段）子集。数据集包含265个episodes和144,788帧，帧率为30 FPS，提供两个摄像头视角（顶部和手腕，分辨率640×480）。状态和动作维度均为6（单臂5维+夹爪1维），动作horizon设置为16。任务包括三种：拾取盘子并放置到目标区域、拾取杯子并放置到目标区域、返回原位，任务分布分别为34.3%、46.2%和19.5%。场景由7个子数据集合并而成，涵盖不同物体组合（如一个杯子、两个盘子等），其中211个episodes成功完成，52个包含重试。数据集采用多任务分段方案，以支持语言-动作关联学习，数据以Parquet和MP4格式存储，并配有配套的Category B数据集用于两阶段微调。LeRobot v3.0 dataset for OpenManipulator-X (OMX) cafe serving task. This is the Category A (Long Episode) subset of the GR00T N1.7 fine-tuning dataset. It includes 265 episodes with 144,788 frames at 30 FPS, featuring two camera views (top and wrist, 640×480). State and action dimensions are 6 each (single_arm × 5, gripper × 1), with an action horizon of 16. Tasks involve picking up plates or cups and placing them in a target zone, or returning to home, with distributions of 34.3%, 46.2%, and 19.5% respectively. The dataset is composed of 7 merged sub-datasets representing various object scenarios, with 211 successful episodes and 52 retry episodes. It uses a label boundary scheme for multi-task segmentation to facilitate language-action grounding, stored in Parquet and MP4 formats, and is paired with a Category B dataset for two-stage fine-tuning.