A personalized model to estimate emotion of individual based on observed facial expression and generation of robot behavior

Abstract

We aim to construct a system which allows robot to learn/generate a behavior favorable to an individual through interaction. The core idea is estimating emotion based on facial expression observed during a robot behavior to self-evaluate how the behavior is favorable to the individual. Based on the self-evaluated data gathered in the past interaction experiences, when the robot does some behaviors, our method is expected to help the robot to predict emotion of the individual gradually and properly. In this paper, we show the user studies to evaluate the proposed method. The study resulted that a robot changed its behavior to get better reaction. As future works, we discuss how robot should use observed data or knowledge to personalize robots behavior.