Preparing Next-Gen Drivers: The Roles of Standardised Terminology, Regular Training and Large Language Models in Driver Education for Autonomous Vehicles

Title: Preparing Next-Gen Drivers: The Roles of Standardised Terminology, Regular Training and Large Language Models in Driver Education for Autonomous Vehicles Abstract: The advancement of vehicular technologies towards autonomy necessitates a re-evaluation of driver interaction with Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicle (AV) technologies. This thesis presents an integrated examination of the challenges and opportunities in enhancing driver understanding and interaction with these systems, focusing on terminology standardisation and transparency, the effectiveness of driver training, and the application of Large Language Models (LLMs) in driver education. Through a series of studies, this research seeks to enhance the integration of ADAS and AV technologies into society, ensuring safer roads and more informed users. The research identifies a critical gap in the standardisation and transparency of terminology and operations in ADAS, highlighting a widespread disparity across manufacturers. This investigation reveals that terminologies that describe ADAS functions vary widely and often lack intuitiveness, contributing to confusion and underutilisation of safety features. The study advocates for the adoption of standardised terminology and clear operational guidelines to improve user comprehension and vehicle safety. These recommendations aim to provide essential insights for Artificial Intelligence (AI) experts and vehicle manufacturers, assisting the transport industry, regulators, and technology developers in designing comprehensive frameworks and guidelines to ensure consistent and effective communication of ADAS functionalities. Further investigation evaluates the role of structured training programs in preparing drivers for the complexities of ADAS and AV technologies. Simulation-based studies show significant improvements in drivers’ operational proficiency post-training, emphasising the necessity of formal education in enhancing road safety in the autonomous era. Training sessions’ results involving novice and experienced drivers indicate that appropriate training significantly enhances drivers’ accuracy and reaction times when interacting with ADAS and AV systems. This underscores the importance of tailored and continuous training programs. The final area of study explores the efficacy of LLM-based instruction versus conventional training methods in educating drivers about ADAS and AV functionalities. Findings reveal that instruction via LLM, specifically through interactions with ChatGPT, significantly outperforms conventional training approaches regarding learning outcomes. Participants trained via LLM demonstrated quicker comprehension and higher consistency in activating and utilising ADAS functions than those trained through conventional methods. The findings further indicate that participants who engaged with ChatGPT-based training scored higher (on average, 21% higher) in accuracy in activating ADAS functions. A statistically significant reduction in activation time for all the functions was also observed, such that P<0.05. This highlights the potential of integrating AI into educational frameworks for complex system operations. Collectively, these research areas contribute to a comprehensive strategy to improve driver safety and efficacy in the era of autonomous vehicles. By advocating for standardised terminologies, effective training programs, and the innovative use of AI in education, this thesis lays the groundwork for future initiatives to ensure drivers are well-prepared for the evolving demands of vehicular technology. These insights offer valuable implications for policymakers, educators, the automotive industry, and the transportation industry, potentially improving global road safety and regulatory practices.