USC-PSI-Lab/humanoid-everyday

Humanoid Everyday is a large-scale humanoid manipulation dataset designed for open-world robotic learning and embodied intelligence research. It contains more than 260 tasks across 7 major categories, covering dexterous manipulation, human–humanoid interaction, and locomotion-integrated activities in diverse real-world scenarios. Collected through a human-supervised teleoperation pipeline, the dataset records synchronized multimodal sensory streams at 30 Hz, including RGB images, depth, LiDAR, tactile, and IMU data. The dataset supports research on perception, decision-making, and manipulation for humanoid robots operating in open environments. It is compatible with the LeRobot v2.0 format and provides task metadata, robot type annotations, and downloadable task resources for applications in robot learning, imitation learning, embodied AI, and humanoid robotics research.

Humanoid Everyday 是一个面向开放世界类人机器人操作学习与具身智能研究的大规模综合数据集，涵盖超过 260 项任务与 7 大类别，包括灵巧操作、人机交互以及融合移动能力的复杂活动场景。数据通过人工监督的遥操作流程采集，以 30Hz 频率同步记录 RGB 图像、深度信息、LiDAR、触觉传感器与 IMU 等多模态感知数据，能够支持类人机器人在真实世界中的感知、决策与操作研究。该数据集兼容 LeRobot v2.0 格式，并提供任务分类、机器人类型及对应下载链接，可广泛应用于机器人学习、模仿学习、具身 AI 与开放世界操作任务研究。