Human-computer cooperation for building controller

This paper proposes an interactive method called Parvaz, in which the expert and a learning program cooperate with each other to create a controller for dynamic systems. This method has been demonstrated for the skill of flying an aircraft in a flight simulator. Dynamic ripple down rules (DRDR) have been developed for those parts of the system where it is possible for the pilot to verbalise his or her control strategy. For those parts of the flight where it is difficult or impossible for the pilot to formulate his or her strategy, learning dynamic ripple down rules (LDRDR) has been developed to automatically produce rules from the logged data of the expert´s behaviour. Parvaz has been tested using three volunteers. All the subjects were successful in creating a set of rules that can fly the aircraft through a given flight plan