Imitation-Based Personalized Driving Decision and Evaluation Method with Experimental Studies

Developing personalized driving systems has become a critical trend in autonomous driving technology. A key focus in this domain is tailoring driving policies to individual drivers, and evaluating their adaptability-both major research challenges. This paper presents an imitation-based method for personalized driving decision and evaluation across typical driving tasks, such as longitudinal following and lateral lane-changing. The proposed personalized decision module captures individual driving characteristics from human demonstration data, while the evaluation module quantifies subjective driving preferences using a novel interactive device and two evaluation metrics. Moreover, extensive experimental studies, conducted through both the driving simulator and a real-virtual vehicle platform, demonstrate the method's superior performance in adapting to individual driving behaviors and improving both the trustworthiness and user satisfaction of autonomous vehicles.