Hierarchical Graph-MPC Trajectory Planning With Multi-Objective Genetic Algorithm-Based Driving Style Optimization

Abstract

Trajectory planning is a critical component in the structure of autonomous vehicles (AVs). However, generating trajectories that simultaneously satisfy multiple requirements for safety, comfort, and low computational cost in stochastic environments remains a challenge. This study proposes a hierarchical trajectory planning framework that integrates a cell-based graph search, utilizing Dijkstra’s algorithm with Model Predictive Control (MPC) to enable efficient spatio-temporal trajectory generation. Unlike conventional decoupled approaches that rely on empirically tuned parameters, this work formulates parameter selection as a multi-objective optimization problem and employs a Genetic Algorithm (GA) to systematically identify optimal parameter sets. The proposed formulation explicitly captures the trade-offs among safety, comfort, and computational efficiency, enabling reproducible and interpretable tuning across different driving styles. Furthermore, the proposed method embeds driving style parameters into the MPC weighting matrices, enabling seamless online adaptation. To support the reliability of the system under dynamic parameter variations, a recursive feasibility analysis is presented, establishing sufficient conditions for recursive feasibility. The proposed framework is validated in high-fidelity IPG CarMaker simulations, including complex multi-vehicle overtaking scenarios and real-world traffic flow conditions. The obtained results demonstrate that the proposed trajectory planning approach effectively adapts to different driving styles by adjusting overtaking initiation distances while achieving significant improvements in lateral comfort metrics. Furthermore, the framework reduces computational time compared with previous studies. These results highlight the effectiveness, robustness under the tested stochastic traffic scenario, and practical applicability of the proposed trajectory planning approach.