Effective adoption of vehicle models for autonomous vehicle path tracking: a switched MPC approach

Abstract

Efficient path tracking plays a key role in the overall ride experience of Autonomous Vehicles (AVs). Model Predictive Control (MPC) is one of the most competent techniques which is capable of handling multiple variables and constraints. The performance of this controller heavily relies on the proper choice of the vehicle model used to predict future states. This work proposes a novel MPC framework for the effective adoption of vehicle models to achieve a compromise between MPC’s performance and computational cost. To this aim, a Switched MPC (SMPC) using vehicle models with different levels of complexity and fidelity is developed. The SMPC uses a novel supervision scheme to adopt the appropriate vehicle model based on the models’ prediction error and MPC’s solution time. Two different configurations of SMPC are implemented: (1) A Fuzzy Logic System (FLS), and (2) An adaptive switching supervisor. The outcomes show that both approaches can perform path tracking accurately on roads with varying curvatures for different AV speeds. Moreover, comparisons with the conventional MPCs show that the proposed controllers can perform comparably with selective use of a complex vehicle model. Notably, they benefit from a higher computation efficiency from sidestepping the most complex models during path tracking.