Cornering Stiffness Adaptive, Stochastic Nonlinear Model Predictive Control for Vehicles

Abstract

The vehicle control behavior is highly dependent on the road surface. However, accurate and precise models for the tire-road interaction are typically unknown a priori. It is therefore important that the vehicle's control algorithm updates its tire-force model, to adapt to the changing conditions. In this paper, we propose a stochastic nonlinear model-predictive control (SNMPC) scheme that uses a linear tire-force model, where the mean and covariance of the cornering stiffness parameters are estimated and updated online. We formulate constraints based on the stiffness estimates to ensure that the vehicle maintains stability on low-friction surfaces. In extensive simulations, where the road surface transitions from asphalt to snow, we compare the proposed controller with various MPC implementations; for example, the proposed approach reduces average closed-loop cost over 30% on aggressive maneuvers, when compared to a non-stochastic controller.