Hierarchical control for autonomous lane changing: integrating Frenet planning with augmented LPV-MPC for passenger comfort

Safe and comfortable lane changing remains a critical challenge in autonomous vehicle control, requiring a balance between dynamic tracking accuracy and ride quality. This study presents a hierarchical control framework that integrates a high-level Frenet-frame trajectory planner with a low-level Augmented Linear Parameter Varying Model Predictive Controller (LPV-MPC). Unlike standard Linear Parameter Varying formulations that prioritize state tracking alone, the proposed augmented state-space model explicitly treats the rate of change of control inputs as the decision variable, allowing for the direct minimization of jerk, cross-track error, and steering rates. Comparative simulations in dynamic manoeuvring scenarios demonstrate the superiority of this approach against a baseline geometric Stanley controller and a Non-Augmented LPV-MPC. The proposed framework achieved a high tracking accuracy with a maximum cross-track error of 0.458 m, significantly outperforming the geometric baseline. Furthermore, it reduced peak longitudinal jerk by 42% compared to the Non-Augmented LPV-MPC (1.720 m/s³ vs. 2.955 m/s³), ensuring compliance with ISO 2631 passenger comfort standards. With an average computation time of 12 ms per iteration on the tested hardware, the framework demonstrates high computational efficiency, suggesting significant potential for real-time implementation on automotive- grade embedded controllers.