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

Murtaza, Mohsin; Cheng, Chi‐Tsun; Fard, Mohammad; Zeleznikow, John · 2024 · OpenAlex-citations

DOI: 10.1007/s40593-024-00407-z

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of effectively training drivers to operate Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicles (AVs), a critical issue given that human error contributes to over 90% of accidents. As vehicles integrate increasingly sophisticated automation, conventional training methods—such as static user manuals and video demonstrations—may be insufficient for fostering the accurate mental models required for safe interaction. The research investigates whether Large Language Model (LLM)-augmented instruction, specifically using ChatGPT, offers superior learning outcomes compared to traditional methods. The experimental design involved 86 licensed participants aged 20 to over 40, with varying levels of driving experience and ADAS usage frequency. Participants were randomly assigned to two groups: a conventional learning group (n=46) and a ChatGPT-based learning group (n=40). Statistical tests confirmed no significant differences in demographics or experience between the groups. Training occurred in a high-fidelity driving simulator equipped with a Logitech G27 racing wheel interface, where specific buttons controlled five ADAS functions: Lane Keeping Assist, Autopilot On/Off, Collision Avoidance, and Adaptive Cruise Control. The conventional group received a 10-minute video demonstration and 15 minutes to read an 1100-word printed manual. The ChatGPT group engaged in a 15-minute interactive session where ChatGPT, constrained to use only the provided manual content, delivered personalized, conversational instruction via pre-designed prompts. Performance was measured by accuracy and reaction time when activating ADAS functions in response to audio triggers during simulated city and highway driving scenarios. The results demonstrated that the ChatGPT-trained group achieved significantly improved learning outcomes compared to the conventional group. Specifically, participants trained via the LLM exhibited shorter activation times, higher consistency, and greater accuracy across all examined ADAS functions. The study also introduced a framework for optimizing LLM prompts in educational contexts, distinguishing between general prompting principles (e.g., goal setting, structured instruction) and application-specific pedagogical principles. This framework ensures that LLMs provide accurate, context-relevant guidance without relying on external knowledge, thereby mitigating risks of hallucination or inconsistency. The findings suggest that LLM-augmented training is a more effective method for imparting complex technical knowledge than static instructional materials. By facilitating interactive, two-way dialogue, ChatGPT helps learners develop robust mental models of vehicle systems, potentially enhancing safety and efficiency in an era of increasing automotive automation. The proposed framework provides a scalable resource for educators and industry stakeholders to tailor AI-driven training for diverse applications, laying the groundwork for broader integration of LLMs in professional and technical education.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• generative ai voice assistants