Integration of Brain-like neural network and infancy behaviors for robotic pointing

Abstract

This paper introduces a new approach to learning pointing behavior in a developmental robot by using a type of constructive neural network and Q-learning algorithm, taking inspirations from human infant development. The pointing behavior is considered as the first movement that human infants use to communicate with other person during human development, it is also the foundation of the human social interaction abilities. We rebuilt this developmental course in our robot simulation system. The learning algorithm of the pointing is implemented by Q-Learning, and a radial based function neural network with resource allocating algorithm is applied to hold the learning result and to control robot movements. The experimental results show that the approach is able to lead our development robot to generate pointing behavior.