Combined observer design for road vehicles using LPV-based and learning-based methods

In this paper a novel observer design method is proposed, which combines Linear Parameter-Varying-based (LPV) and machine-learning-based design tools. As a first step, a parameter optimization technique is developed to achieve a polytopic LPV formulation of the system model. This modeling technique also involves a machine-learning-based solution to determine scheduling parameters for the LPV system. In the second step, a LPV-based observer design based on the achieved system representation is proposed. Finally, the operation and the effectiveness of the proposed observer algorithm are demonstrated through a vehicle-oriented estimation problem, i.e., estimation of the lateral velocity. In the paper two simulations illustrate the accuracy and the advantageous impact of the observer on the control performances of the closed-loop system.