“ProAgent: Harnessing On-Demand Sensory Contexts for Proactive Agent Systems in the Wild” was recently accepted to IMWUT and will be presented at UbiComp 2027.
In this paper, we propose ProAgent, an end-to-end proactive agent system that harnesses on-demand sensory contexts to provide proactive assistance in the wild. ProAgent continuously perceives users’ surroundings through on-demand tiered perception, extracts hierarchical contexts that integrate sensory information with user preferences, and proactively reasons about users’ needs to invoke external tools and deliver timely assistance. We evaluate ProAgent on both public and real-world datasets, along with a user study of 20 participants, demonstrating its effectiveness in enabling proactive assistance in everyday scenarios.
Please checkout our paper: https://arxiv.org/abs/2512.06721