Estimation system of human behaviors using fuzzy neural network based object selection

Abstract

The estimation of human behaviors by robots is one of key technologies in an environment in which humans and robots coexist. In this paper, a method is proposed for promptly estimating the behavioral targets by applying a fuzzy neural network (FNN). Here, inputs to the FNN are the human velocity, the angle of the human relative to an object, and the distance between the human and an object, whereas outputs are confidences that each object among all candidates is selected to be an intended object. The resultant human behavior can be estimated as a combination of the human action estimation and the behavioral target estimation.