An important challenge associated with driving simulation development is the computational representation of agent behaviours. This paper describes the development of a preliminary autonomous agent behaviour model (based on the Recognition-Primed Decision (RPD) model, and Hintzman's multiple-trace memory model) mimicking human decision making in approaching an intersection controlled by a traffic light. To populate the model, an initial Cognitive Task Analysis was conducted with six drivers to learn the important cues, expectancies, goals, and courses of action associated with traffic light approach. The agent model learns to associate environmental cues (such as traffic light colour) with expectancies of upcoming events (like light colour change) and appropriate courses of action (such as decelerating). At present, the model is currently being evaluated for its successful representation of the Recognition-Primed Decision Making process.
Abstract