We have described a CA model for bi-directional traffic which exhibits qualitatively correct behavior. This model is a relatively concise description of a complicated phenomenon, despite the number of parameters it contains. Further work may show that some of these parameters can be combined while preserving the underlying dynamics. Though systematic exploration of the parameter space should surely be undertaken, we find that our model is qualitatively insensitive to small changes in parameter value. Throughout the experiments described in this paper, we have fixed the values of most parameters at reasonable values, and focused on the two most important variables: the density of vehicles on each lanes. We have localized the regions where lane interactions are strongest, and explored how these interactions occur. Passing interactions generally have a weak effect on the density-flow diagram. Still, due to the high relative speeds involved, these interactions have practical importance in traffic safety.
Various modifications of our model could be investigated. For instance, we do not use "look back" as in [2], though this certainly could be considered. Rather than conditioning the decision to pass on the local density, one could decide on the basis of the gap in front of the vehicle to be passed. This would be more in the spirit of [1], but, we suspect, less robust.
On a winding road, sections of roadway are such that vehicles on either lane can pass, or none can pass or both can pass. The distribution and length of these sections depends on the curvature of road. In our model, curvature can be approximated by changing the distribution of these various sections of road. Thus, traffic on roads with realistic topography can be modeled with cellular automata.