Man, machine cooperation for learning to control dynamic systems

This paper describes experiments in building a controller to pilot an aircraft by using machine learning techniques and knowledge acquisition methods. The aim was to build the controller by acquiring the knowledge of a skilled operator at that task. A flight simulator program has been modified to interact with a knowledge acquisition program for creating rules and logging the pilot´s actions along with the flight information. Ripple down rules method is used as knowledge acquisition tool and Induct software as induction program to automatically create rules from the logged data. The created rules were tested by running the flight simulator in autopilot mode where the autopilot code has been replaced by the created rules. The autopilot must fly the plane according to a defined flight plan