Abstract

Robotics is one of the most challenging applications of Machine Learning (ML) techniques. It is characterized by direct interaction with a real world, sensory feedback and an enormous complexity of the control system. In recent years several approaches to apply ML to specific robotics tasks have been published. Nevertheless we are still far from a complete autonomous robot control system with learning components. This paper aims at pointing out the problems and possible applications for integrating learning capabilities into a robot control system and then describes a new integrated system architecture which shows up a set of components necessary for (partially) autonomous systems with learning facilities. 1. Introduction In recent years there has been an increasing interest in applying machine learning techniques to robotics. The applications are manipulator as well as mobile system tasks. The learning techniques used range from rote learning [25, 22, 1, 8, 27] and inductive learning...

Citations