Human Behavior Prediction and Induction by Gaze Information

Abstract

In recent years, demand of intelligent robots is increasing. When such an intelligent robot performs cooperative work with a human, it is necessary to predict human intention and behavior in order not to apply unnecessary load or stress to human beings. Currently, techniques for predicting such intention and behavior of human beings have been extensively studied, but it is difficult to produce results beyond human ability simply by matching actions to humans based on predictions, and limits on efficiency etc. However, giving instructions from the robot side to humans may be stressful, and it is necessary to conduct casual guidance. Therefore, in this research, we focus on human passing each other, and model human evasive behavior by using gaze and neural network. Based on this model, the aim is to not only perform actions tailored to human beings but also conduct induction by robot. As a result, we showed that it is possible to change the choice of behavior by giving subjects a slight change.