Steering Control Considering Motion Sickness and Vehicle Performance via DDPG Algorithm and 6-DoF-SVC Model

Autonomous driving demands sophisticated control systems that optimize safety, performance, passenger comfort, and fuel efficiency. This study proposes a steering control system that integrates the Deep Deterministic Policy Gradient (DDPG) for speed planning with a novel feedback mechanism based on Subjective Vertical Conflict (SVC) in the reward function. Using simulations in MATLAB and Simulink, we evaluate the system's performance across various thresholds of SVC, examining its impact on ride comfort, fuel efficiency, and vehicle behavior during lane changes. Results reveal a trade-off relationship between ride comfort and fuel efficiency, with lower SVC thresholds generally improving comfort but potentially increasing steering input. Additionally, excessively low SVC thresholds degrade target-reaching performance and lengthen lane change distances, highlighting the need for careful parameter tuning. Overall, our findings demonstrate the potential of reinforcement learning-based steering control systems to optimize multiple evaluation criteria simultaneously while emphasizing the importance of balancing trade-offs in autonomous driving scenarios.