Hybrid Parameter-varying Model Predictive Control for Autonomous Vehicle Steering

In this paper the concept of Hybrid Parameter-Varying Model Predictive Control (HPV-MPC) is applied for autonomous vehicle steering. Parameter-varying in the MPC context means that a prediction model with non-constant, parameter-varying system matrices is employed. In the investigated scenarios, the displacement of a car on an icy road due to a side wind gust shall be mitigated, and a double lane-change maneuver shall be performed autonomously. In order to explore a possible reduction of online computations and the inherent degradation of control performance, the nonlinear model of the lateral dynamics is approximated in various ways. A comparison between controllers using prediction models varying from the full nonlinear model, as an indication for the maximal achievable performance, to a linear model was performed. Particular emphasis was put on the hybrid parameter-varying prediction model, to investigate their potential in terms of computational effort and control performance.