CoRL2026-CSI/Isaaclab-so101_11task_baseCaP_3300epi_10fps

IsaacLab SO101 11-Task baseCaP数据集是一个机器人学习数据集，包含3,300个episodes和1,175,352帧数据，帧率为10 FPS。该数据集源自CoRL2026-CSI/Isaaclab-so101_11task_baseCaP_3300epi源数据集，通过每三帧取一帧（frame_index % 3 == 0）进行下采样，并重建了每个episode的frame_index、timestamp和全局index以支持10 FPS播放。数据集涵盖11个任务家族，每个任务家族有300个episodes，任务包括拾放立方体、按颜色分类立方体、堆叠两个立方体、推按钮、打开杠杆、关闭杠杆、提取立方体、打开盒子、关闭盒子、推立方体和拉立方体。摄像头数据包括顶部视角RGB摄像头（480x640x3）和左腕部RGB摄像头（480x640x3）。特征包括机器人状态/动作字段（以度和URDF弧度形式）、末端执行器姿态和夹爪状态、每帧的技能自然语言描述、技能类型和进度、每技能的目标关节、末端执行器姿态和夹爪字段、每帧的子任务自然语言描述、对象名称和目标位置，以及SCRAPE元数据。数据集已通过SCRAPE审计验证，确保数据完整性和一致性。The IsaacLab SO101 11-Task baseCaP Dataset is a robotics learning dataset containing 3,300 episodes and 1,175,352 frames at 10 FPS. It is derived from the source dataset CoRL2026-CSI/Isaaclab-so101_11task_baseCaP_3300epi by downsampling every third episode-local source frame (frame_index % 3 == 0) and rebuilding each episodes frame_index, timestamp, and global index for 10 FPS playback. The dataset includes 11 task families, each with 300 episodes, covering tasks such as pick and place cube, sort cubes by color, stack two cubes, push button, turn lever on, turn lever off, extract cube, open box, close box, push cube, and pull cube. Camera data includes a top-view RGB camera (480x640x3) and a left wrist RGB camera (480x640x3). Features encompass robot state/action fields in degree and URDF-radian forms, end-effector pose and gripper state, per-frame skill natural language, skill type, and skill progress, per-skill goal joint, EE pose, and gripper fields, per-frame subtask natural language, object name, and target position, along with SCRAPE sidecar metadata. The dataset has passed SCRAPE audit verification, ensuring data integrity and consistency.