Modeling lateral control in driving studies.

In driving studies based on simulators and instrumented vehicles, specific models are needed to capture key aspects of driving data such as lateral control. We propose a model that uses weighted polynomial projections to predict each data point from the previous three time points, and accommodates the attempts of the drivers to re-center the vehicle before crossing the borders of the traffic lane. Our model also allows the possibility that average position within the lane may vary from driver to driver. We demonstrate how to fit the model using standard statistical procedures available in software packages such as SAS. We used a fixed-base driving simulator to obtain data from 67 drivers with Alzheimer's disease and 128 elderly drivers without dementia. Using these data, we estimated the subject-specific parameters of our model, and we compared the two groups with respect to these parameters. We found that the parameters based on our model were able to distinguish between the groups in an interpretable manner. Hence, this model may be a useful tool to define outcome measures for observational and interventional driving studies.