Incremental Learning of Human Emotional Behavior for Social Robot Emotional Body Expression

Generating emotional body expressions for social robots has been gaining increased attention to enhance the engagement and empathy in human-robot interaction. In this paper, an enhanced model of robot emotional body expression is proposed which places emphasis on the individual user's cultural traits. Similar to our previous paper, this approach is inspired by social and emotional development of infants interacting with their parents who have a certain cultural background. Social referencing occurs when infants perceive their parents' facial expressions and vocal tones of emotional situations to form their own interpretation. On the other hand, this model replaces the batch learning self-organizing map with the dynamic cell structure, incrementally training a neural network model with a variety of emotional behaviors obtained from the users with whom the robot interacts. We demonstrate the validity of our incremental learning model through a public human action dataset, which will facilitate the acquisition of emotional body expression of socially assistive robots as a reflection of the individual user's culture.