EgoVerse

EgoVerse is a large-scale egocentric dataset designed for robot learning, embodied AI, and human-to-robot transfer research. The dataset contains diverse first-person human demonstration data collected from wearable cameras and embodied sensing platforms, including egocentric RGB videos, camera poses, hand tracking information, task annotations, and robot learning trajectories. EgoVerse captures real-world human activities across various tasks, environments, and embodiments, enabling research on imitation learning, behavior cloning, multimodal perception, human-object interaction understanding, and intelligent agent learning. By providing large-scale egocentric observations with rich spatial and semantic information, EgoVerse supports the development of embodied AI systems capable of learning from human experiences and performing complex real-world manipulation tasks.EgoVerse 是一个面向机器人学习、具身智能以及人类到机器人迁移研究的大规模第一视角数据集。该数据集包含通过可穿戴摄像设备和具身感知平台采集的大规模人类示范数据，包括第一视角 RGB 视频、相机位姿、手部跟踪信息、任务标注以及机器人学习轨迹等多模态数据。EgoVerse 覆盖多种真实世界任务、环境和具身形态，记录人类从第一视角执行任务过程中的视觉观察、动作行为以及环境交互信息，可支持模仿学习、行为克隆、多模态感知、人机交互理解以及智能体学习等研究方向。通过提供具有丰富空间信息和语义信息的大规模第一视角数据，EgoVerse 支持开发能够从人类经验中学习，并在真实世界环境中完成复杂操作任务的具身智能系统。