Future neuro mobile robots

Abstract

Neurocomputing is considered as an approach to robotic processing. Based on transformations, it autonomously develops operational capabilities in adaptive response to information training. The idea of training a mobile robot to carry out a function (instead of programming it) seems to have a great appeal, perhaps because of our familiarity with training in some robotic applications (e.g. teaching trajectory generation) as an easy and natural way to acquire new skill processing capabilities. Even though neural networks mobile robotic applications are still fully undeveloped, this paper tries to show that their promise and potential is evident. We show this by describing which neural networks have been used (or are suitable to be used) for various mobile robot capabilities, and what sort of problems are expected to surface out; if any. We then propose a system of interacting neural networks to be used as a model for future neuro mobile robots. Finally, experiments on Neurobot, a six legged articulated robot testbed, for testing various networks as well as for providing a perspective on the usefulness of integrating various mobile robot capabilities are presented.<>