PhD Alumni

Tognetti Simone

Present position: CEO at EmoticaLab. Developing business by applying the technology developed during the Ph.D at Politecnico di Milano

Thesis title:  A methodological framework for physiology based affective computing: definition and evaluation
Advisor:  Andrea Bonarini
Research area:  Artificial intelligence, robotics and computer vision
Thesis abstract:  
In this work, we have faced the problem of estimating emotions of people that interact with an artificial system in real contexts. This problem has been faced in the past by two different, but interlinked, disciplines: Psychophysiology and Affective Computing.

Psychophysiology has a strong grounding on methodological aspects that support the identification of valid, reliable and general descriptive models of the phenomena; however, it suffers from applicability limited mainly to lab settings, and from an attention to classical emotional processes such as stress or mental work load. On the other hand, Affective Computing has a strong grounding on techniques and methods to achieve the goal of facing the problem in an application scenario; however it suffers of poor attention to methodological aspects so that, often, it is difficult to replicate the obtained results.

By taking into account the interesting aspects and to overcome the limitations of both disciplines, the main contribution of our work is the proposal of a general methodological framework that provides: (1) a general model of the affective phenomena that takes place into real life Human-Computer Interaction (HCI) scenarios; (2) a methodology that guides the definition and realization of experiments to estimate valid and reliable models in these scenarios.

The second contribution of our work are the application our methodological framework to three different scenarios and the discussion about the results we have obtained. The first application is a Rehabilitation Robotics task, in which we involved impaired patients that have been asked to perform a rehabilitation protocol characterized by a set of sessions with a rehabilitation robot. A second application scenario is a multimedia context in which 15 subjects have been exposed to different multimedia contents in order to recognize their arousal. Finally, a third context is given by a video game scenario, in which we involved 75 subjects and we have modeled their preferences for a game setting or another.

The third contribution of our work is related to the problem of the suitability of devices to measure physiological signals. Most of the current research on emotion recognition is based on laboratory instruments that become less suitable the more the research outside laboratory contexts is brought. We designed and implemented a wireless headset that is easy-to-wear, that does not compromise the interaction and that measures the physiological signals within many application contexts in a reliable way.