Risk-aware robotic infection control: Integrating computer vision and physics-informed decision-making for resilient ventilation in offices

This study proposes a risk-aware robotic personalized ventilation framework that adaptively relocates a mobile air purifier to reduce airborne infection risk in offices. The system integrates computer vision-based occupancy sensing, a physics-informed surrogate model that predicts infection risk by learning single-source computational fluid dynamics (CFD) scalar transport fields and reconstructing multi-occupant conditions via linear superposition, and uncertainty-aware decision-making under infection-source uncertainty among detected occupants. Evaluated in a university meeting room, the surrogate achieves a mean absolute error of 0.77% in infection probability on blind test scenarios, significantly outperforming purely data-driven algorithms. Across 255 occupancy configurations, results reveal a Pareto conflict between aggregate cleaning efficiency and individual safety equity, motivating an adaptive planner that switches between risk-neutral and risk-averse policies. Compared with static baselines, the proposed strategy reduces mean risk by 35.5% and up to 74% in worst-case exposure. On-site prototype trials demonstrate a scalable, low-latency solution for upgrading indoor safety. • Proposed a risk-aware robotic framework for adaptive purifier relocation. • Seat-level occupancy sensing drives real-time ventilation decisions. • Physics-informed surrogate model predicts risk with high accuracy (MAE = 0.77%). • Risk-neutral and risk-averse planning handle unknown infectious occupants. • Adaptive control reduces mean infection risk by 35.5% versus static baselines.