Adaptive Curvature-Aware Model Predictive Control Using Hybrid BO–TPE for Accurate Autonomous Vehicle Path Tracking

Model predictive control (MPC) has emerged as one of the most effective control strategies for autonomous vehicle path tracking owing to its capability to explicitly handle dynamic constraints and optimize future control actions. However, its tracking performance strongly depends on the appropriate tuning of prediction, control, and weighting parameters, which remains a challenging and computationally demanding task under varying driving conditions. This paper proposes an Adaptive Curvature-Aware MPC (CAMPC) framework optimized through a hybrid Bayesian Optimization–Tree-structured Parzen Estimator (BO–TPE) approach to automatically identify optimal MPC parameters while accounting for upcoming road curvature. The proposed controller incorporates future curvature information to adapt the vehicle speed profile and steering behavior, thereby improving tracking accuracy and control smoothness in complex road geometries. The framework is evaluated in the CARLA autonomous driving simulator under three challenging driving scenarios, including a roundabout, an urban environment, and a highly curved road. Experimental results demonstrate that the proposed CAMPC consistently outperforms a conventional PID controller, achieving improvements of 47%, 58%, and 85% in trajectory-tracking performance across average and maximum lateral and angular errors while reducing steering, braking, and acceleration efforts by more than 80%. Furthermore, the controller satisfies real-time execution requirements with an average computation time of 31 ms and a 95th-percentile latency of 35 ms, confirming its suitability for practical autonomous driving applications. These results demonstrate that integrating curvature-aware prediction with adaptive BO–TPE parameter optimization significantly enhances the robustness, accuracy, computational efficiency, and real-time capability of MPC for autonomous vehicle path tracking in challenging driving environments.