Robot learning to walk: An architectural problem for intelligent controllers

Abstract

The purpose of this paper is to analyze the design premises of a new control architecture based upon the principle of Nester Hierarchical Control which has the feature of learning from the already programmed skills of a teleoperated system. We will explore different architectural principles and compare them. We will show how high level languages can be built from low level languages by using cooperation of the already existing skills, new expert knowledge (teaching new heuristics), and self-experience (example based) learning.