Making Driver Modeling Attractive

I n t e l l i g e n t T r a n s p o r t a t i o n S y s t e m s intervention. Many DAS applications are safety oriented, such as lane departure warning systems. Some are comfort oriented, such as automated parking assistants. Still others are designed to alleviate the driver's attentional load—for example, traffic signal detectors or the already deployed intelligent cruise control systems. These systems must adapt to varying traffic situations to achieve the goal of enhanced safety and comfort. Moreover, DAS human-machine interfaces must support careful communication with potentially taxed drivers. Driver models support DAS design in several ways. First, they let researchers analyze system performance in a variety of driving situations. They also help optimize parameter values for certain performance measures, 1 such as the degree to which a desired speed is maintained. More sophisticated yet, driver models can form the basis for classifying driver types, such as aggressive versus defensive. DAS applications can use these classifications to activate different parameter value sets that adjust warning thresholds appropriately for different driver types. Driver models might also help classify traffic scenarios, such as fluid versus congested traffic. Analyzing the scenario and the driver's behavior, DASs could autonomously adapt parameter settings to best respond to the situation. Eventually , onboard operation of driver models could enhance scene analysis and interpretation by predicting the behavior of not only the car in which they are installed but also other cars in the vicinity. 2 As the driver model more fully accounts for the traffic scene, DASs increase their power to compare observed with predicted behaviors, 3,4 and thereby detect anomalies. We have applied attractor dynamics to modeling driving behaviors. Originally developed to generate behavior in autonomous robots, attractor dynamics encode behavioral policies for modeling a driver with meaningful parameters that support optimization by direct policy search. We used a powerful evolutionary algorithm to vary the parameters in order to generate three driver models that capture the behavioral patterns of different driver types. Our work focuses on modeling the tactical level of driving decisions, such as when to brake and whether to accelerate and pass another vehicle. Such decisions are based on local, instantaneously available environmental information about the road and other cars in the same or an adjacent lane. The tactical level poses considerable modeling difficulty. First, tactical driver …