SVM Classification and Kalman Filter Based Estimation of the Tire-Road Friction Curve

Abstract

An estimation method is proposed for the maximum tire-road friction coefficient µMAX and its corresponding wheel-slip λpeak. After a preliminary analysis of Burckhardt’s model analytic properties, a machine learning approach involving Support Vector Machines is used to classify the curves representing different road surfaces. A metric to evaluate the misclassification risk is used to determine boundaries for the curve coefficients. These boundaries are then used to form linear inequality constraints for two recursive parameter estimation algorithms based on the Kalman Filter, in its Extended and Unscented forms. Finally, the proposed method is evaluated via simulations in Matlab environment.