| As advancements in machine learning (ML) continue to drive the evolution of Human-Robot Interaction (HRI) in driver monitoring systems, there remains a pressing need to address persistent challenges, including latency, privacy concerns, and maintaining real-time accuracy across diverse driving conditions. This paper critically reviews existing ML-based HRI frameworks for driver monitoring systems, underscoring their limitations in detecting subtle driver states and ensuring user privacy and fairness. In response, we propose an innovative framework that leverages offline, user-specific retraining on local devices, enabling models to adapt to individual drivers' behaviors without requiring data transmission. This approach reduces latency, enhances model accuracy over time, and preserves user privacy. This also offers a structured pathway for developing adaptive, privacy-focused HRI systems that are responsive to individual needs. Our framework provides a significant step forward in creating robust, user-centric driver monitoring solutions for future transportation systems. For evaluation, we focus on a driver drowsiness monitoring system using deep learning models trained on facial images. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.