AoE: 用于具身 AI 的始终开启的体姿式人类视频集合
摘要
具身基础模型需要大规模高质量的真实世界交互数据进行预训练和扩张。然而，现有数据收集方法存在高基础设施成本、复杂硬件依赖和有限交互范围，使得可扩展扩张具有挑战。事实上，人类本身是理想的物理具身代理人。因此，从全球分布的"人类代理人"那里获取体姿式真实世界交互数据具有低成本和可持续性优势。为此，我们提出了 Always-on Egocentric (AoE) 数据收集系统，旨在通过利用人类及其智能手机来简化硬件依赖，实现低成本、高效且与场景无关的真实世界交互数据收集，以解决数据稀缺的挑战。具体而言，我们首先采用一种人体工学颈部挂载式智能手机支架， enables low-barrier, large-scale egocentric data collection through a cloud-edge collaborative architecture。 Second, we develop a cross-platform mobile APP that leverages on-device compute for real-time processing, while the cloud hosts automated labeling and filtering pipelines that transform raw videos into high-quality training data。 Finally, the AoE system supports distributed Ego video data collection by anyone, anytime, and anywhere. We evaluate AoE on data preprocessing quality and downstream tasks, demonstrating that high-quality egocentric data significantly boosts real-world generalization。
引用
@article{arxiv.2602.23893,
title = {AoE: Always-on Egocentric Human Video Collection for Embodied AI},
author = {Bowen Yang and Zishuo Li and Yang Sun and Changtao Miao and Yifan Yang and Man Luo and Xiaotong Yan and Feng Jiang and Jinchuan Shi and Yankai Fu and Ning Chen and Junkai Zhao and Pengwei Wang and Guocai Yao and Shanghang Zhang and Hao Chen and Zhe Li and Kai Zhu},
journal= {arXiv preprint arXiv:2602.23893},
year = {2026}
}