Bayesian Optimizer-Based MPC for Trajectory Tracking of Autonomous Vehicles Subject to Steering Lag

In this article, a Bayesian optimizer-based model predictive control (MPC) strategy is proposed for trajectory tracking of an autonomous vehicle subject to steering lag. The strategy integrates a nonlinear MPC algorithm with a Bayesian optimizer to compensate for the steering lag and to enable offline automatic tuning of parameters. The nonlinear MPC algorithm includes a first-order steering model that explicitly accounts for the lag inherent to a dual-model steering system of the autonomous vehicle. The Bayesian optimizer is designed to automatically tune both controller and model parameters offline using historically recorded vehicle data. The resulting optimal parameters are then incorporated into the nonlinear MPC algorithm to enhance the online trajectory tracking performance. Recursive feasibility and asymptotic stability of the closed-loop system are rigorously established under the MPC framework. Experimental results for continuous-curve and lane-change scenarios are provided to validate the effectiveness and the advantages of the proposed Bayesian optimizer-based MPC strategy.