Hierarchical Adaptive MPC for Autonomous Vehicle Lane Changing via PSO and GWO

Abstract

This paper presents an intelligent hierarchical two-layer highway lane-change planning and control framework for autonomous vehicles, integrating Adaptive Model Predictive Control (AMPC) with metaheuristic-based parameter optimization algorithms, namely Particle Swarm Optimization (PSO) and Grey Wolf Optimizer (GWO). In the proposed architecture, the lower control layer employs AMPC to generate optimal control actions based on predicted vehicle dynamics, while the upper supervisory optimization layer utilizes PSO and GWO to iteratively refine the AMPC weighting parameters, selected control constraints, and prediction horizon for each lane-change scenario through a simulation-in-the-loop optimization framework. The hybrid AMPC–PSO/GWO framework is formulated to improve trajectory-tracking performance while reducing vehicle jerk, thereby enhancing control performance and passenger comfort. A comprehensive comparative analysis is conducted in MATLAB/Simulink under four representative highway-driving scenarios, including single and double lane-change maneuvers on both straight and curved roads. Simulation results demonstrate that the proposed two-layer hybrid approach significantly outperforms conventional AMPC by achieving more accurate trajectory tracking, reduced lateral deviation, lower relative yaw angle RMSE, and substantially decreased longitudinal and lateral jerks. These improvements indicate enhanced vehicle stability and smoother driving behavior during lane-change maneuvers. Robustness and stability assessments further demonstrate satisfactory closed-loop performance under sensor-noise disturbances, tire-model uncertainties, and sudden-braking scenarios while maintaining safety-constraint satisfaction. Overall, the proposed framework demonstrates promising performance for autonomous-vehicle lane-change planning and control in representative highway-driving scenarios.