Modeling the effect of driving experience on lane keeping performance using ACT-R cognitive architecture

Abstract

ACT-R (adaptive control of thought-rational) is a computational cognitive architecture that consists of multiple modules for perception and attention (e.g., a visual module), response making (e.g., a motor module), and mental information processing (e.g., a goal module, declarative memory module, and procedural memory module). The output of an ACT-R model is a time course that shows the coordination of the activation of these modules in solving a cognitive task. The research reported in this paper used a driver performance model developed in ACT-R to examine the effect of driving experience on driversperformance of vehicle lateral position control. First, we compared the lane keeping performance between novice and experienced drivers in a simulated driving experiment with different road, speed, and driving conditions. Behavioral data collected from both groups of drivers were then modeled in the ACT-R driving model to analyze the cognitive mechanisms of the behaviors. The modeling results were similar to the human results. There were two main findings. (1) Novice drivers had poor lateral control skills evidenced by significantly greater lateral deviation compared with experienced drivers. The lack of control was more evident on curved roads than on straight roads. On curved roads, the novice driversdriving paths significantly shifted to the outside of the curve, whereas there was no such shift for experienced drivers. (2) The cognitive-architecture-based modeling in ACT-R revealed the mental processes including parallel processing between modules and serial processing between sub-tasks. One of the factors responsible for the difference in curved-road driving paths between the two groups of drivers may be the location of visual attention focus. Our model showed that the behavioral difference in curved-road driving paths can be modeled by assigning a visual attention focus point to the novice-driver model that is closer than that assigned to the experienced-driver model. The method and results of this study can be used to provide cognitive-psychological bases for improving the effectiveness and efficiency of driver training courses and developing intelligent driving assistance systems. This research also provides an application example and a successful experience of computational cognitive modeling using cognitive architectures such as ACT-R.