Modeling Driver Behavior : A Control Theory Based Approach

Abstract: The aim of the project, which led to this thesis, was to suggest a micro-level driver behavior model, which would allow estimating the effects of safety measures on road traffic safety. The empirical research on driver behavior should be conducted in urban areas, because they are much less explored than rural areas, even though most non-fatal accidents occur in urban areas. The high complexity of urban areas and their general diversity make it necessary to adopt a systems perspective. In that way it is possible to investigate interactions of several variables and to become aware of complex relationships. The model should provide access to different so-called "safety-indicators", which allow estimating changes in road traffic safety. Safety indicators are for the most part easier to measure in a model than in reality, because in a model more parameters are known.Various aspects of driver behavior have been modeled in different disciplines and with different purposes. Their common denominator is that they only deal with subsets of driver behavior. So far no attempt has been made at modeling all aspects of driving behavior. In this thesis it is attempted to provide a framework for a general model of driver behavior. It is represented in the style of control theory, and it includes important aspects of existing models. It is assumed that the driver plans her actions based on a mental model of the situation, which helps her predicting what might happen in the future. Motivational aspects are included, and a distinction is made between what a driver wants to do and what a driver can do by considering physical and physiological limitations.The representational form of the model allows considering driver behavior on different levels of detail, which should make it usable for many different purposes. So far the model mainly demonstrates general relationships between different variables related to driving, and it is expected that the systematic incorporation of empirical results will make it possible to specify the functions in the model more and more.Within the frame of this thesis three empirical studies have been conducted and their results are presented in relation to the suggested model. One study examined an urban junction with a "view from the outside", that is, mainly macro data were gathered to obtain an understanding for the overall situation in the junction. It was established, for example, how much a vehicle is influenced by the speed of the vehicle ahead depending on time headway. Furthermore, it was determined that drivers decelerate when they approach a junction, even though they have right of way. Driving speed during daytime and nighttime was compared, and it was found that the average speed is higher at night in an urban location, but during the day in a rural location. Additionally, it was found that more safety critical interactions take place close to an intersection that further away from it. With help of the second study it was investigated how drivers experience the junction by letting a number of participants drive through the junction and report afterwards what disturbed them and how stressed they felt. Other road users were perceived as disturbing, but also driver-related factors like age influenced the perceived stress-level. In the third study it was examined whether the skill to predict the development of a situation and the perceived complexity of the situation were related to environmental or driver factors. It was found that situations, in which traffic rule violations occurred, were harder to predict and also judged to be more complex. Experience and gender did not have any influence on prediction skill, but more experienced drivers gave lower complexity ratings. The results of these studies show that it is important to consider the many interactions in road traffic when conducting research and when modeling driver behavior.