Abstract
Keywords
Driver Modeling, Fuzzy Logic, Bayesian Networks, C-clustering, Emergency situation.
Abstract
Active safety systems, such as ABS, ESP and EBA, improve safety aboard vehicles and on roads. Their functioning is based essentially on driver's behavior. It is necessary to analyze the real behavior in both "normal" and "emergency" for extracting the most relevant parameters (particularly in braking maneuver). Based on this analysis, we can realize a driver's digital model, which associated with a digital model of vehicle, will allow to test and to optimize safety systems during their development phase.
Within the framework of the researches on the drivers' behavior in emergency situations, the LAB has created a knowledge database for two types of scenario: "rear-end" accidents and loss of control (in bend). These situations were experimented with lambda drivers on track and on driving simulator.
This paper is focused on driver's behavior analyzing and modeling during emergency braking situations in a straight line. Two methods were evaluated: a qualitative or probabilistic one based on Bayesian networks and a quantitative one based on fuzzy logic. The first approach is efficient for the qualitative description of the events and the driver's actions in a reflex and/or regulating mode during a limited time. Fuzzy logic proposes a quantitative method allowing to have numerical values directly used.