Automatic configuration of autonomous robot systems

Making reinforcement learning industrially applicable

Abstract

This paper focuses on an approach for configuring autonomous robot systems based on a formalized task description. The aim is to enable the industrial use of reinforcement learning, a machine learning method that has made considerable progress in recent years in increasing the autonomy of robot systems through the use of deep learning. The use of presvious approaches to the automation of the engineering of robotic systems offers the possibility to simplify the configuration of these learning systems, for which deep expert knowledge is still necessary.