Project R-16793

Design and Development of an Adaptive Digital Coaching Avatar Using Large Language Models for a School Ecosystem. (Research)

Road traffic injuries represent a critical global health challenge, particularly for young road users who are vulnerable due to developing risk perception and decision-making skills. In Belgium, nearly 40% of traffic accidents involving children occur during the commute to school, highlighting the urgent need for effective safety interventions within these dynamic mobility contexts. While road safety education is a recognized approach for improving awareness, traditional programs often lack the individualized guidance and sustained engagement necessary for long-term behavioral change. This research proposes the design and development of an adaptive digital coaching avatar that leverages Large Language Models (LLMs) to provide personalized road safety education. The project aims to enhance engagement and encourage safer decision-making among students, parents, and drivers interacting within school environments. The system's core intelligence will utilize Retrieval-Augmented Generation (RAG) to ensure dialogue is grounded in authoritative safety guidance, combined with prompt engineering to deliver empathetic and persuasive feedback. The methodology integrates psychological frameworks, specifically the Theory of Planned Behavior (TPB) and the Transtheoretical Model (TTM), to profile users based on their interaction history and scenario performance, allowing the avatar to tailor its coaching strategies to an individual's specific stage of behavioral change. Through a multi-stage approach involving requirement gathering, system design, and pilot deployment in Belgian schools, the research will evaluate the avatar's impact on safety knowledge, risk awareness, and decision-making. Ultimately, this project contributes to broader policy initiatives like "Vision Zero" by addressing the human behavior component of traffic safety through human-centered, emotionally intelligent technology. Beyond its immediate impact, the project aspires to create a scalable framework that can later be adapted for diverse mobility contexts internationally. It further seeks to strengthen collaboration between academic, governmental, and educational stakeholders, ensuring long-term sustainability and practical relevance. The insights gained may also support future integration of AI-driven coaching systems into national road safety strategies and digital learning ecosystems.

18 June 2026 - 30 June 2030