Robot-based measurement of comfort through thermal infrared imaging and physiological signals

This master thesis shows the outcome of an innovative measurement methodology for the assessment of thermal comfort through a robotic structure. The thermal sensation vote (TSV) is measured by a social robot that acquires indoor environmental parameters (air temperature and relative humidity) and physiological parameters such as heart rate variability and skin temperature. Low-cost and minimally invasive sensors (i.e. smartwatch, infrared sensor) were integrated into the robot and the proposed solution has been tested on 10 participants during a three days measurement campaign. The development of this method aims at overcoming the lack of a personalized and robot-based technique for thermal comfort measurement. Indeed, the traditional methodology is based on the computation of Predicted Mean Vote (PMV) and Percentage of People Dissatisfied (PPD) which give an overall estimation of thermal comfort in a group of subjects not considering thermal differences among individuals. Further studies have introduced adaptive methods which assess human thermal sensation starting from both physiological signals and environmental parameters. However, the tools and devices used in this case are invasive, expensive and they do not guarantee an accurate measurement. Hence, wearable sensors and infrared technologies have been introduced. Moreover, the increasing relevance of the Internet of Things (IoT) and Artificial Intelligence (AI) led scientific research to thermal comfort assessment procedures with Machine Learning (ML) and robotic systems. Nevertheless, they are still not user-centered methodologies. Based on these considerations, the proposed robot-based method succeeds in measuring thermal comfort expressed as TSV via ML algorithms. The used dataset is composed of physiological and environmental parameters acquired by the integrated robotic structure. The employed ML algorithms have reached different predictive accuracies and among them, the Random Forest classifier and the Naïve Bayes provided the highest one. Hence, the present study demonstrates the feasibility of the robot-based methodology in the scenario of thermal comfort assessment. Nevertheless, this master thesis shows some limitations that mainly concern the quality of the collected data and the real-life context application. Future studies may enlarge the input dataset of ML algorithms by developing longer experimental protocols. Moreover, an automatization procedure can be developed to obtain a methodology tailored to people’s everyday lives.