Online parameter estimation of driving behavior using probability-weighted ARX models

The dynamical characteristics of driving behavior may change due to various reasons, such as the increase of experience, fatigue, and change of driving condition. In the design of a driver-assisting system that exploits a mathematical model of the driving behavior, the online adaptation mechanism for the driving behavior model must be developed and implemented. This paper presents an online parameter estimation scheme for the Probability weighted ARX (PrARX) model, which is a class of a hybrid dynamical system model, and is known to capture the complex characteristics of the driving behavior together with an explicit understanding of the drivers´ motion control and decision making aspects. Since the parameter estimation for the PrARX model is originally based on a steepest descent manner, it is quite natural to extend it to the online version. The proposed method is first demonstrated using artificial data, and then applied to the online modeling of the driving behavior.