Transforming Driver Education: A Comparative Analysis of LLM-Augmented Training and Conventional Instruction for Autonomous Vehicle Technologies

Abstract

As modern vehicles continue to integrate increasingly sophisticated Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicles (AV) functions, conventional user manuals may no longer be the most effective medium for conveying knowledge to drivers. This research analysed conventional, paper and video-based instructional methods versus a Large Language Model (LLM)-based instructional tool to educate 86 participants about the operation of specific ADAS and AV functionalities. The study sampled participants aged between 20 and over 40, with driving experience ranging from one to over six years. The first group was educated using the conventional methods. In contrast, the second group received instructions via an LLM, i.e., users learn via ChatGPT interaction. Our goal was to assess the efficiency and effectiveness of these teaching methodologies based on the reaction times participants required to activate ADAS functions and the corresponding accuracies. Our findings revealed that the group trained via ChatGPT demonstrated significantly improved learning outcomes compared to conventional training. This included shorter activation times, higher consistency, and higher accuracy across examined functions. This study further proposed a framework to effectively use ChatGPT for different training scenarios and education purposes, offering a valuable resource for leveraging Artificial Intelligence (AI) in training users to handle complex systems. The framework empowers educators to tailor ChatGPT’s interactions, ensuring efficient, guided learning experiences for learners. For researchers, this study lays the foundation for exploring the role of LLM-based instructional tools in a broader range of applications.