Adaptive robotic sensing for exposure-oriented mapping of hazardous airborne particle dispersion in indoor environments.

Rapid hazardous airborne particle mapping enables localized exposure risk detection and timely intervention. Yet heterogeneous time-varying plumes under ventilation elude sparse fixed sensors or pre-planned scans. We propose an exposure-oriented adaptive robotic framework converting one-shot scanning into iterative sampling for targeted reconstruction. Kernel DM + V was used as the base field reconstruction model, and the framework was evaluated through CFD-based virtual sampling and a full-scale PM2.5 aerosol dispersion experiment under mechanically ventilated conditions. In the CFD case, adaptive sampling reduced the normalized reconstruction error in the high-concentration region of interest to 9.5% using 65 sampling points, compared with 18.3% for uniform scanning, whereas optimized uniform scanning achieved a slightly lower final global NRMSE than adaptive sampling (2.5% vs. 3.0%). In the physical experiment, adaptive sampling reduced the average reconstruction error at four fixed reference locations from 59.7% to 34.3% under the tested condition. With the same 72-point sensing budget, the adaptive workflow generated an initial concentration map after 18 exploratory samples and then updated the map after each six-point batch. These results demonstrate the proof-of-concept potential of adaptive robotic sensing for the timely identification of localized hazardous particle plumes and for providing spatially resolved information to support exposure-risk-oriented indoor contaminant control.