JMIR Publications

ABSTRACT

Social robots are becoming increasingly important in our daily life as companion. Therefore, humans expect to interact with them using the same mental rules applied to human-human interaction, including the use of non-verbal channels. Over the last years, research efforts are devoted to understanding users’ needs and mental models and to develop behavioral models for robots that can perceive the user state and properly plan a reaction. In this context, the aim of this paper is dual. Firstly, it aims to assess the user’s perceived emotions, in terms of valence, arousal, and dominance, when interacting with a robot that shows/does not show non-verbal cues with respect to the same emotions elicited through a personal computer; secondly, it aims to evaluate the robot performance in the recognition of emotions using three supervised machine learning, namely Support Vector Machine, Random Forest, and K-Nearest Neighbor. Particularly, this paper presents a study where 60 healthy subjects were requested to interact with Pepper robot to elicit three “basic” emotions (i.e. positive, negative, and neutral) using a set of 60 images, retrieved from a standardized database thus to investigate i) the role of the robot embodiment and ii) its coherent movement respectively in emotion’s perception and recognition. The results show significant differences in the user perceived emotion with respect to the web interface underlying the importance of the robot non-verbal communication and embodiment. Additionally, the results underline a good robot’s emotion recognition rate (accuracy higher than 0.85 with the best classifiers), suggesting that the use of multimodal communication channels improves the recognition of the user's emotional state.