Affective Robots: Evaluation Of Automatic Emotion Recognition Approaches On A Humanoid Robot Towards Emotionally Intelligent Machines

Abstract

One of the main aims of current social robotic research<br> is to improve the robots’ abilities to interact with humans. In order<br> to achieve an interaction similar to that among humans, robots<br> should be able to communicate in an intuitive and natural way<br> and appropriately interpret human affects during social interactions.<br> Similarly to how humans are able to recognize emotions in other<br> humans, machines are capable of extracting information from the<br> various ways humans convey emotions—including facial expression,<br> speech, gesture or text—and using this information for improved<br> human computer interaction. This can be described as Affective<br> Computing, an interdisciplinary field that expands into otherwise<br> unrelated fields like psychology and cognitive science and involves<br> the research and development of systems that can recognize and<br> interpret human affects. To leverage these emotional capabilities<br> by embedding them in humanoid robots is the foundation of<br> the concept Affective Robots, which has the objective of making<br> robots capable of sensing the user’s current mood and personality<br> traits and adapt their behavior in the most appropriate manner<br> based on that. In this paper, the emotion recognition capabilities<br> of the humanoid robot Pepper are experimentally explored, based<br> on the facial expressions for the so-called basic emotions, as<br> well as how it performs in contrast to other state-of-the-art<br> approaches with both expression databases compiled in academic<br> environments and real subjects showing posed expressions as well<br> as spontaneous emotional reactions. The experiments’ results show<br> that the detection accuracy amongst the evaluated approaches differs<br> substantially. The introduced experiments offer a general structure<br> and approach for conducting such experimental evaluations. The<br> paper further suggests that the most meaningful results are obtained<br> by conducting experiments with real subjects expressing the emotions<br> as spontaneous reactions.