Prediction of driving behavior based on sequence to sequence model with parametric bias

Prediction of driving behavior is important in safer advanced driving assistance systems to prevent potential risk of traffic accidents. To predict the driving behavior, driving style, i.e., the difference of tendency of driving behavior for each driver, is also important. We propose a prediction method for time-series driving behavior based on the driving style. The proposed method consists of a sequence-to-sequence (S2S) model and embeds driver information considering with the driving style into S2S model in order to improve prediction performance. We evaluated efficiency of the proposed method by two experiments using actual driving behavior data on a test track. We found that the proposed method could predict driving behavior better than comparative methods. We also found that the proposed method could change the driving style from one driver to another driver by changing driver information input to the proposed model.