|M.Sc Student||Katz Ada Romina|
|Subject||State Dependence in Lane Changing Models|
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Tomer Toledo|
|Full Thesis text|
The purpose of this research is to enhance existing lane changing models to incorporate and capture the persistence of the driver’s behavior, through modeling of the underlying lane selection process. Persistent behavior is assumed and accepted for strategic travel choices (e.g. destination, path and schedule). Based on this, we can also assume that drivers may persist in trying to complete driving goals such as lane changing. Thus, the decisions drivers make over time are not independent, but are related by a logical and stable relation.
Hidden Markov Models (HMMs) are appropriate for taking into account the transitions between phases and find their use in categorizing sequences of data. HMMs are based on two hypotheses: there exists a latent selection process which evolves from state to state (in our case, the selection of the target lane) and that the study of an observable output (i.e. the observed lane changing action) could provide information on this process. The observable state depends on the previous choices, which are the underlying hidden states. For example, we observe that a driver stays in his current lane, but we can not observe the real reason that caused him to stay there. The driver may have chosen not to pursue a lane change and to stay in his current lane or he may have chosen to move to another lane but could not complete the lane change. In summary, we can assume that the lane changing decision process is latent and only the driver’s actions (lane changes) are observed.
A framework for modeling the lane changing behavior taking into account the state dependence between observations of a given driver over time, which utilizes the above mentioned concepts, is developed. Statistical tests show that the State Dependence Model does better fit the data compared to previous models and therefore should be selected for prediction.