Learning System for Emotion Estimation and Emotional Expression Motion Generation based on RNN with Russell's Circumplex Model

Abstract

Recently, various robots trying to communicate with and support for human beings, for example, pet-type and service robots, have been increasing. It is required to realize smooth communication skills with human beings for the robots. In this research, we aim to realize Interactive Emotion Communication(Interactive Emotion Communication: IEC) -which is a bidirectional communication based on emotional behaviors between a human and a robot. IEC consists of three processes- (1)inferring human emotion, (2)generating robot emotion, and (3)expressing robot emotion. The purpose of IEC is to raise the personal a.nity which the robot gives to the human by interactive emotional behaviors. In our previous research, the authors have proposed the "Fuzzy Emotion Inference System(FEIS)". The FEIS particularly focuses only on the process of "human emotion inference" by analyzing the human body motion values based on Laban's theory. It measures the basic psychological value by fuzzy reasoning and infers the emotion based on Russell's circumplex model. The "human emotion inference" should be tightly related to the "expressing robot emotion", however, the conventional methods do not take it into account. This paper proposes "Recurrent Neural Network with Russell's Circumplex Model(RNNRCM)" -which introduces Russell's circumplex model to a Recurrent Neural Network that learns human emotion inference through motion and robot emotional motion generation bidirectionally. The RNNRCM realizesthe process of "recognizing human emotion" and "expressing robot emotion" in the IEC. We confirm the efficacy of the proposed method with experiments.