Application of intelligent search algorithm in LKS of autonomous vehicles

As autonomous driving technology becomes increasingly integrated into intelligent transportation systems, enhancing the resilience and swift responsiveness of Lane Keeping Systems (LKS) under challenging driving conditions has emerged as a pivotal area of investigation. In response, a sophisticated dual-layer LKS control approach has been devised, harmonizing input scene refinement with output control optimization through learning. This methodology introduces a novel modeling framework for continuous variable testing scenarios, leveraging intelligent search algorithms, alongside an advanced deep reinforcement learning control model. The latter incorporates priority experience replay and soft update mechanisms to precisely emulate decision-making processes within continuous action domains. In typical urban intersections, the trajectory deviation of the proposed method is controlled within ± 0.5 m, and the deviation of the circular ramp is about ± 0.4 m. During the disturbance phase, the lane acceleration remains within ± 0.02 g, the lane offset is 0.0031 m, and the steering wheel angle is -6.9°. The research results indicate that this method has good key variable mining ability and control strategy adaptability in typical extreme scenarios, and can effectively improve the response robustness and decision stability of LKS systems to sudden environments. The research aims to provide a reference path for intelligent algorithm driven integrated testing and control systems for autonomous driving.