Prediction of driver’s handling assessment using a general regression neural network

Abstract

This paper describes a method for predicting the driver's subjective assessment of tire handling, based on vehicle dynamics measurements taken during driving a double lane change. Often used methods for regression have limitations that prevent their proper implementation for tire handling data. A General Regression Neural Network (GRNN) is developed in this research to model the driver's subjective assessment, as this modelling technique does not have these limitations and additionally, offers a clear interpretation of the method and results. The results show that the GRNN can predict the subjective handling aspects scores well, even for handling aspects not directly related to driving the double lane change. Analysis of the important measures used by the GRNN provides additional information on the vehicle dynamics behavior relevant for the driver. For general handling behavior, regarding all aspects, this showed that metrics based on lateral velocity, steering wheel moment and vehicle slip angle were most relevant.