Multiple-Model Robust Adaptive Vehicle Motion Control

Abstract

An improvement in active safety control systems has become necessary to assist drivers in unfavorable driving conditions. In these conditions, the dynamic of the vehicle shows rather different respond to driver command. Since available sensor technologies and estimation methods are insufficient, uncertain nonlinear tire characteristics and road condition may not be correctly figured out. Thus, the controller cannot provide the appropriate feedback input to vehicle, which may result in deterioration of controller performance and even in loss of vehicle control. These problems have led many researchers to new active vehicle stability controllers which make vehicle robust against critical driving conditions like harsh maneuvers in which tires show uncertain nonlinear behaviour and/or the tire-road friction coefficient is uncertain and low.

In this research, the studied vehicle has active front steering system for driver steer correction and in-wheel electric motors in all wheels to generate torque vector at vehicle center of gravity. To address robustness against uncertain nonlinear characteristics of tire and road condition, new blending based multiple-model adaptive schemes utilizing gradient and recursive least squares (RLS) methods are proposed for a faster system identification. To this end, the uncertain nonlinear dynamics of vehicle motion is addressed as a multiple-input multiple-output (MIMO) linear system with polytopic parameter uncertainties. These polytopic uncertainties denote uncertain variation in tire longitudinal and lateral force capacity due to nonlinear tire characteristics and road condition. In the proposed multiple-model approach, a set of fixed linear parametric identifi cation models are designed in advance, based on the known bounds of polytopic parameter set. The proposed adaptive schemes continuously generates a weighting vector for blending the identifi cation model to achieve the true model (operation condition) of the vehicle. Furthermore, the proposed adaptive schemes are generalized for MIMO systems with polytopic parameter uncertainties. The asymptotic stability of the proposed adaptive identifi cation schemes for linear MIMO systems is studied in detail.

Later, the proposed blending based adaptive identi fication schemes are used to develop Linear Quadratic (LQ) based multiple-model adaptive control (MMAC) scheme for MIMO systems with polytopic parameter uncertainties. To this end, for each identi fication model, an optimal LQ controller is computed on-line for the corresponding model in advance, which saves computation power during operation. The generated control inputs from the set of LQ controllers is being blended on-line using weighting vector continuously updated by the proposed adaptive identifi cation schemes. The stability analysis of the proposed LQ based optimal MMAC scheme is provided. The developed LQ based optimal MMAC scheme has been applied to motion control of the vehicle. The simulation application to uncertain lateral single-track vehicle dynamics is presented in Simulink environment. The performances of the proposed LQ based MMAC utilizing RLS and gradient based methods have been compared to each other and an LQ controller which is designed using the same performance matrices and fixed nominal values of the uncertain parameters. The results validated the stability and effectiveness of the proposed LQ based MMAC algorithm and demonstrate that the proposed adaptive LQ control schemes outperform over the LQ control scheme for tracking tasks.

In the next step, we addressed the constraints on actuation systems for a model predictive control (MPC) based MMAC design. To determine the constraints on torque vectoring at vehicle center of gravity (CG), we have used the min/max values of torque and torque rate at each corner, and the vehicle kinematic structure information. The MPC problem has been redefi ned as a constrained quadratic programming (QP) problem which is solved in real-time via interior-point algorithm by an embedded QP solver using MATLAB each time step. The solution of the designed MPC based MMAC provides total steering angle and desired torque vector at vehicle CG which is optimally distributed to each corner based on holistic corner control (HCC) principle. For validation of the designed MPC based MMAC scheme, several critical driving scenarios has been simulated using a high- fidelity vehicle simulation environment CarSim/Simulink. The performance of the proposed MPC based MMAC has been compared to an MPC controller which is designed for a wet road condition using the same tuning parameters in objective function design. The results validated the stability and effectiveness of the proposed MPC based MMAC algorithm and demonstrate that the proposed adaptive control scheme outperform over an MPC controller with fixed parameter values for tracking tasks.