Rapid Behavior Adaptation for Human-centered Robots based on Integration of Primitive Confidences on Multi-sensor Elements

Abstract

This paper presents a method for teleoperated mobile robots to rapidly adapt to a behavior policy. Rapid policy adaptation cannot be achieved when significant data are not differentiated from insignificant data in every process cycle. Our method solves this problem by evaluating the significance of data for learning based on the change in degree of confidence. A small change in the degree of confidence can be regarded as reflecting insignificant data for learning (that data can be discarded). Accordingly, the system can avoid having to store too frequent experience data, and the robot can adapt more rapidly to changes in the user's policy. In this paper, we confirm that by taking advantage of a significance evaluation of each proposition of each sensor level data, a robot can rapidly adapt to a user policy. We discuss the results of an experiment on a mobile robot in which the user switched policies between 'avoid' and 'approach'.