Assessing behaviour of cognitive agents in a flight simulator with fighter pilots

Abstract

Intelligent, human-like, computer-controlled opponents could improve the training value of tactical training simulators for fighter pilots. To create such opponents, realistic computational cognitive models are needed. We present the evaluation of a cognitive model that has been developed to simulate Situation Awareness (SA) and the ability to be surprised. The crucial element in the model is a mechanism that matches beliefs about the situation from the SA sub-model with expectations about the world from the surprise sub-model. To evaluate its use and realism for the domain, the model was evaluated by operational fighter pilots in a fighter aircraft simulator. They performed in air combat scenarios against opponents whose actions were controlled by the cognitive model. Results indicate that computer-controlled opponents with the integrated SA/Surprise model contribute to a challenging and realistic training environment and could therefore be included in tactical training for operational pilots. The surprise effect in the virtual opponent was indeed recognizable by the human pilots. The human pilots found it less straightforward to judge the quality of the SA of their opponents.