Personalized Lane-Change Trajectory Planning for Autonomous Driving in Dynamic Environments

Proper lane changing operations are critical for autonomous driving, and consideration of personalized driving preferences during lane changing can promote riding experience, especially in dynamic driving environments. While existing methods seldom comprehensively integrate specific passenger preferences and dynamic environment uncertainties within a unified decision-planning framework. For this reason, a hierarchical lane change trajectory planning framework encompassing decision-making and planning layers is proposed to comply with personalized requirements and adapt to dynamic scenarios. In the decision-making layer, the passenger preferences and styles are firstly recognized via the analytic hierarchy process and fuzzy inference system. Then, a pure strategy game approach is developed to generate decisions by balancing riding safety, overall traffic efficiency and riding preferences. In the planning layer, a hyperparameter optimization-based neural network is established, with the incorporation of passenger preferences, to predict the proper lane change duration. A fused method combining polynomial curves and quadratic programming is employed to generate optimal lane change trajectories in real time based on passengers' preferences on safety, comfort and efficiency. Simulation results demonstrate that the proposed method can not only produce personalized decisions, but generate LC trajectories based on different passenger preferences.