While datasets for video understanding have scaled to hour-long durations, they typically consist of densely concatenated clips that differ from natural, unscripted daily life. To bridge this gap, we introduce MM-Lifelong, a dataset designed for Multimodal Lifelong Understanding. Comprising 181.1 hours of footage, it is structured across Day, Week, and Month scales to capture varying temporal densities. Extensive evaluations reveal two critical failure modes in current paradigms: end-to-end MLLMs suffer from a Working Memory Bottleneck due to context saturation, while representative agentic baselines experience Global Localization Collapse when navigating sparse, month-long timelines. To address this, we propose the Recursive Multimodal Agent (ReMA), which employs dynamic memory management to iteratively update a recursive belief state, significantly outperforming existing methods. Finally, we establish dataset splits designed to isolate temporal and domain biases, providing a rigorous foundation for future research in supervised learning and out-of-distribution generalization.
@article{arxiv.2603.05484,
title = {Towards Multimodal Lifelong Understanding: A Dataset and Agentic Baseline},
author = {Guo Chen and Lidong Lu and Yicheng Liu and Liangrui Dong and Lidong Zou and Jixin Lv and Zhenquan Li and Xinyi Mao and Baoqi Pei and Shihao Wang and Zhiqi Li and Karan Sapra and Fuxiao Liu and Yin-Dong Zheng and Yifei Huang and Limin Wang and Zhiding Yu and Andrew Tao and Guilin Liu and Tong Lu},
journal= {arXiv preprint arXiv:2603.05484},
year = {2026}
}