Exploring Electrodermal activity as a means of detecting engagement in child robot interactions
Social robots such as Tega serve as learning aids in children’s education. A merit Tega provides is the ability to personalize curriculums to fit the current learning capability of the child which contributes to better learning outcomes. Tega uses emotional facial recognition platforms to detect engagement but this method provides only a one-dimensional view of engagement and does not give insight to the child’s cognitive state. Physiological signals such as electrodermal activity (EDA) could be leveraged as a complementary tool to detect engagement because it reflects changes in internal activity such as cognitive load.
To understand how EDA signals can be used as inputs in machine learning models to detect engagement, I analyzed data recorded while young children interacted with a Tega during an educational storytelling activity. I applied signal processing techniques such as noise reduction and filtering to clean the EDA data and created a series of visualizations of the data in different storytelling interaction states to enable effective data analysis.
To further explore the data, I performed a statistical test such as the student T- test on a set of conditions to quantify EDA mean values across two independent groups of kids. Then, I used the questions asked during the storytelling sessions and matched them to the recorded raw data to observe changes in the EDA signal. The ability to identify factors that provide a viable method of tracking when children are engaged and creating predictive models based on these identified factors can lead to better learning outcomes in the child to social robot learning scenarios.