Vision-Based Autonomous Surface Sampling in Healthcare Robotics: An Integrated Perception–Control Framework with Experimental Validation

Abstract

Healthcare-Associated Infections (HAIs) remain a critical challenge, with surface contamination representing a key transmission vector. Current sampling procedures are performed manually, exposing operators to risks and introducing variability in data collection.The proposed pipeline combines YOLOv8-based target detection, surface localization, depth-derived geometric representation and fixed-base swabbing trajectory validation. Autonomous base navigation is treated as an upstream capability, while the proposed contribution focuses on perception-guided surface modeling and swabbing execution. The swabbing mission is divided into an arm reaching step and a trajectory execution step, with contact and tool-orientation constraints enforced only during trajectory execution.The framework is evaluated through a fluorescence-based protocol that compares robotic and manual sampling. Human sampling achieves a higher mean raw fluorescence signal (mean µ = 26080.5, standard deviation σ = 8371.8), whereas robotic sampling yields slightly lower values (µ = 24713.5, σ = 7794.1). The coefficients of variation are similar, with 32.1% for manual sampling and 31.5% for robotic sampling. These findings support the feasibility of perception-guided robotic surface sampling under controlled conditions, without establishing superiority over expert human performance.