Modeling human lane keeping control in highway driving with validation by naturalistic data

This paper considers two models of steering control relevant to lane-keeping during normal driving. The first is well-known from the literature and uses linear feedback. To test whether the model is capable of representing real-world driving under conditions of lane-keeping with low workload, parameter fitting is carried out using naturalistic driving data (NDD). It is found that the model can fit the NDD quite well, and that two of the three control parameters may be estimated in a consistent and repeatable way. The lack of fit in the third parameter is explained by the recognition that lane-center tracking does not occur in practice; hence response to lane boundaries is considered more relevant. However the instability of the resulting closed-loop controller calls into question the validity of the linear model. Further, it is found that the linear model does not adequately represent the intermittent pulse-like qualities of real-world steering control. Based on these considerations, a second model is formulated and initial comparisons with NDD are presented. It is proposed that the new model may inherently take account of workload demands, and is therefore relevant to issues of visual attention allocation during driving under reduced workload.