Abstract
Cleaning is an essential factor in every aspect of our day-to-day life. For the past ten years, the leading-edge technologies in robotics have been successfully applied in the field of domestic and professional. However, the problem of quantifying the quality of cleaning delivered by any method remains unsolved so far. Presently, post-cleaning auditing is often left ignored or done by manual supervision. This work aims to address the fundamental research problem of quantifying the cleanliness using robotaided cleanliness audit. This founding effort toward auditing the level of cleanliness starts with analytical formulations for cleanliness estimation and tactile-vision-based sensing techniques. The robot-aided cleaning auditing framework uses a mobile robot to estimate the magnitude of cleanliness, using the derived dirt estimation models. The robot gathers the audit information from multiple locations in its area of operation and assesses the cleanliness in a fully autonomous fashion. The major contribution of this research work includes the formulation of a cleaning-auditing framework for assessing the cleaning-quality using audit robots. As part of the proposed framework, the development of a first of its kind sample audit sensor that analyzes the dirt density of a point location using the principle of adhesive dust lifting (tactile) and computer vision is reported. The research work also contributes a comprehensive domestic-dirt dataset for AI-based dirt analysis techniques and systematic analysis on the usability of the dataset is conducted. Considering the challenges on adopting the conventional coverage planning algorithms for robot-aided cleaning auditing, an novel optimal non-uniform sampling strategy using geometrical ques that contributes to a region’s dirt accumulation is devised. This study also covers the experience-driven dirt exploration strategy for audit robots using deep-reinforcement learning principles. The robot-aided cleaning auditing method further pushes the boundaries of automated cleaning by providing a scientific method for cleaning quality analysis. The proposed framework can bring a significant impact to the cleaning and maintenance domain by providing a scientific and viable approach for cleaning benchmarking in a fully automated fashion. Above all, the proposed work aims to solve the paradox of – how clean is clean?