Driver models to emulate human anomalous behaviors leading to vehicle lateral and longitudinal accidents

A new kind of driver model is developed to emulate anomalous driving behaviors. This new driver model will be developed based on the concept that a driver model that normally achieves driving tasks could be perturbed to emulate anomalous behaviors like human drivers by considering humans' inherent limitations or by incorporating error mechanisms. If driver limitations or error mechanisms are properly designed, the driver model can generate accident or near-accident behaviors that are of interest to engineers who are developing active safety technology.;Driver limitations can be physical and/or mental. Those limitations may cause driving accidents and need to be considered in the model. Another major contributor of driving accidents is driving error. Most existing models focus on describing driver behavior under certain tasks, and few of them include driving errors. The main contribution of this study is to fulfill the missing link between modeling normal driving tasks and modeling driving accidents. The development of an architecture and modeling process for driver models that emulates anomalous behaviors will be provided. Despite our best effort, no research on such driver models was found in literature.;The model architecture and modeling process will be demonstrated by two examples. Lateral disturbance rejection for a lane-keeping task will be used to illustrate driver behavior under lateral disturbance. Mother example studies the effect of driving errors during longitudinal car-following. The goal of the lateral driving example is to analyze crosswind induced vehicle stability problems and the driving accident induced by human driver limitations. Both numerical simulations and driving simulator experiments were conducted to collect lateral driving behaviors. Lateral normal driving behaviors and accident inducing behaviors were studied. A lateral driver model with accidents was developed and used to evaluate vehicle crosswind stability and active safety system design. In the second example, we focus on the longitudinal car-following behavior. An errable driver model was constructed and used to capture human/control interaction and thus accelerate the collision warning/collision avoidance system development process. Driver errors can be viewed as a recurring event. If proper human cognition/error mechanisms are included and proper probability distribution functions are used to introduce human errors, it is possible to reproduce accident/incident behavior that is statistically similar to field testing results.