RNN with Russell's circumplex model for emotion estimation and emotional gesture generation

Abstract

Interactive Emotion Communication (IEC) has been proposed[1] and studied so far. IEC consists of three processes, recognition of human emotion, generation of robot emotion, and expression of robot emotion. Conventional studies designed those processes by hand one by one. This report proposes a comprehensive system that learns human emotion recognition and robot emotion expression both. The proposed system is a recurrent neural network introducing Russell's circumplex model explicitly and learns human emotion and corresponding motion pattern simultaneously. We show the validity of the proposed method through experiments.