Quantifying Trade-Offs in Autonomous Driving with a DRL-Based Multi-Objective Control System via the SVC

Abstract

Advancing autonomous driving technology demands balancing multiple evaluation metrics, including ride comfort, energy efficiency, and vehicle performance. This study presents a novel steering control system that incorporates the 6-DoF-SVC model into the reward function of a DDPG framework, enabling the optimization of ride comfort while preserving energy efficiency and vehicle performance. The proposed system utilizes a hybrid architecture combining DRL-based decision-making with a PI controller, allowing flexible trade-offs among evaluation metrics. Numerical experiments conducted in MATLAB and Simulink under varying ride comfort thresholds (th_svc) demonstrate that appropriate th_svc settings can simultaneously enhance ride comfort and either energy efficiency or vehicle performance, depending on operational objectives. Additionally, the study identifies a significant trade-off between steering effort and ride comfort, indicating that specific th_svc settings permit the simultaneous optimization of both. However, challenges such as oscillatory behavior during lane changes were observed, highlighting potential areas for improvement. This research provides valuable insights for designing autonomous vehicle control systems that address competing objectives. Future work will focus on dynamic velocity planning, integrating more realistic driving scenarios, and validating system performance in dynamic environments involving other vehicles and pedestrians. By addressing these challenges, the proposed system aims to improve the safety, efficiency, and comfort of autonomous driving solutions.