ICRA-WBCD/IO-AI-SenseXperience-UMI

SenseXperience UMI人类示范数据集是一个开源数据集，展示了SenseXperience可穿戴具身AI数据收集平台在真实世界中的数据采集能力。该数据集包含1,051个事件、220,963帧图像和92个操作任务，这些数据均来自自然人类示范，并通过同步的多模态传感器收集。数据集强调在真实场景中的自然人类示范，无需主臂遥操作，降低了硬件和部署复杂性。它包括6个同步摄像头视图（如左/右自我中心头部视图、左/右手腕操作视图、头部深度摄像头和夹具摄像头），帧率为30 FPS，视频采用H.264编码。数据还包含头部、左夹具和右夹具的7自由度姿态信息，以及夹具状态。数据集支持具身AI预训练、视觉-语言-动作训练、模仿学习、人机转移研究和多视图操作理解等应用。数据集以Parquet和MP4格式存储，兼容LeRobot工作流程。SenseXperience UMI Human Demonstration Dataset is an open dataset that showcases the real-world data acquisition capability of SenseXperience, a wearable embodied AI data collection platform. It provides 1,051 episodes, 220,963 frames, and 92 manipulation tasks collected from natural human demonstrations with synchronized multimodal sensors. The dataset emphasizes natural human demonstrations in real scenarios without master-arm teleoperation, reducing hardware and deployment complexity. It includes 6 synchronized camera views (e.g., left/right egocentric head views, left/right wrist manipulation views, head depth camera, and gripper camera) at 30 FPS, with videos encoded in H.264. Data also features 7-DoF poses for the head, left gripper, and right gripper, along with gripper states. The dataset is suitable for embodied AI pretraining, vision-language-action training, imitation learning, human-to-robot transfer research, and multiview manipulation understanding. It is stored in Parquet and MP4 formats and compatible with LeRobot workflows.