Combining Knowledge Acquisition and Machine Learning to Control Dynamic Systems

Abstract

This paper presents an interactive method for building a controller for dynamic systems by using a combination of knowledge acquisition and machine learning techniques. The aim is to build the controller by acquiring the knowledge of an operator skilled at that task. This method has been demonstrated for the skill of learning to f i l l an aircraft in a flight simulator. The simulator has been augmented to interact with a knowledge acquisition program for creating rules and logging the pilot's actions along with fl ight information. We have developed a method called Dynamic Ripple Down Rules for knowledge acquisition and Learning Dynamic Ripple Down Rules for automatically generating rules from the logged data. The rules were tested by running the flight simulator in autopilot mode where the autopilot code is implemented by the rules.