A hybrid approach for modelling intelligent agents in air combat simulations
Simulating air combat encounters has proven an efficient way of training fighter pilots. The Dutch National Aerospace Laboratory (NLR) aims to improve training simulations by developing intelligent humanlike opponents for military purposes. At present, several learning Computer Generated Forces (CGF) have been proposed that adapt to the environment. Yet, it remains hard to develop CGF’s that learn in an efficient humanlike manner from changing and complex environments.
This thesis will attempt to involve the Situation Awareness (SA) model Endsley into CGFs. Within this prominent information processing model, SA simply refers to the state of knowledge about a dynamic environment; a crucial capacity for eliciting intelligent behaviour. Combined with Dynamic scripting (a machine learning algorithm), a CGF will learn the most optimal behaviour given the circumstances. To be more specific, this study addresses the question whether a cognitive model can efficiently be combined with a machine learning algorithm to improve CGFs behaviour in changing air combat simulations.