Driving Style Modelling with Adaptive Neuro-Fuzzy Inference System and Real Driving Data

Abstract

With different cognitive abilities and driving style preferences, car-following behaviors can vary significantly among human drivers. To facilitate the replications of human driving behaviors on chassis dynamometer using a robot driver, this paper proposes a novel fuzzy logic driver model that attempts to perform humanized driving behaviors in the car-following regimes. An adaptive neuro-fuzzy inference system was developed to tune the fuzzy model using real driving data collected from an instrumented vehicle. Driver’s cognition parameters, such as headway distance, vehicle speed and pedal positions, were modelled as system inputs. Meanwhile, driver’s action parameters, such as pedal movements and gear selection, were selected as system outputs. Three models that possess different driving styles were calibrated using the system. Afterwards, in order to evaluate its performance of emulating human behaviors, the established fuzzy models were examined in a simulation scenario that is anchored to standard WLTC drive cycle tests.