Modeling driver performance: The effects of rear-end collision warning algorithms

Rear-end collisions are the single largest type of collision and account for almost 30% of all collisions. Because rear-end crashes account for such a large percentage of automobile collisions, and because inattention is the most frequent cause of these crashes, there has been considerable research into the possibility of alerting inattentive drivers to potential collision situations. This dissertation develops and validates two models of driver performance and demonstrates them by evaluating collision avoidance systems.; The first model is a simplified representation of the driver that responds to a warning following a fixed delay with a constant deceleration. This model predicts 81% of the outcomes from a driver-in-the-loop simulation, and had a 0.85 correlation between braking profiles generated by the drivers and the model. However, this model does not account for the closed-loop nature of driver response in which the driver modulates braking based on the observed environment.; To capture this closed-loop behavior, an attentional model of the driver was developed. This model was based on the field theory of driving and assumes that a field of safe travel exists based on affordances associated with the driving environment. In this model, driver behavior depends on the driver's current perception of the field of safe travel. This allows for different responses based on the severity of the situation. More severe situations illicit ballistic responses whereas less severe situations elicit modulated responses. This model of the driver was also able to accurately predict the probability of a collision based on RECAS parameters and combinations of headway and lead vehicle deceleration.; The simple model of the driver showed that even very simple models can produce interesting and counterintuitive findings. Identifying complex interactions with a simple model helps avoid misattributing the source of complex behavior. The more complicated driver model showed that a field theory approach can accurately predict actual driver performance and provides a starting point to better understanding how people interact with the driving environment. This approach can be expanded to examine a range of issues beyond rear-end crashes.