Identifying tacit strategies in aircraft maneuvers

Skilled performance at complex control tasks is especially suited for models produced by machine learning because of its consistency. In the study, two machine learning methods were used to find descriptions of avoidance strategies employed by skilled pilots in simulated aircraft encounters. A general approach to describing strategic components of skilled behavior through qualitative representation of situations and responses is introduced. “Conceptually equivalent” descriptions of the pilots' maneuvers were discovered by a concept learning algorithm A^qand a classifier system employing a generic algorithm. Satisficing and “buggy” strategies not apparent in earlier analyses of these data were revealed. The agreement of different algorithms using different generalization criteria demonstrates the robustness of this machine learning approach to describing skilled behavior.