Imitation bootstrapping: experiments on a robotic hand

Abstract

Imitation is a vast topic for both human sciences and robotics. Recent advances in the understanding of the neural mechanisms of imitation offer methodologies that bring the two research domains closer. In this paper we analyze how an imitation system can be bootstrapped from a non-imitative system at an abstract level so that the ideas derived can be applicable to infant development as well as robotics implementation. The main idea put forward is that all the imitation learning systems -whether artificial or not-can be broadly seen as learning by self-observation or social learning. Human infants possibly make use of both during development. This study explores imitation learning on a robotic hand platform using connectionist architecture with minimal conventional engineering approach to imitation. In the paper we present the details of the implementation and discuss the implications of our results. This study, at a higher level, serves as an example how the interplay between brain sciences and robotics can not only guide us in building human-like behaving machines but also help us understand the mechanisms of human behavior