Driving ability in older adults is essential for independent mobility and social participation, yet declines under high cognitive load or distraction often lead to visual attention failures such as "look-but-fail-to-see," increasing crash risk. Older adults and individuals with mild cognitive impairment (MCI) show impairments in visual attention, executive control, and visuomotor integration, which are not adequately captured by conventional assessments. Virtual reality (VR) integrated with eye-tracking and upper-limb motion analysis enables ecologically valid simulation of driving scenarios and precise quantification of visuomotor behavior. However, current studies are limited by single-scenario designs, unimodal AI models, and insufficient integration of action-related data.
This study proposes a multi-phase framework: Year 1 develops an eye-movement-based AI model for MCI identification; Year 2 integrates multimodal data in VR driving tasks; and Year 3 establishes an explainable AI system with longitudinal validation. The study aims to advance cognitive assessment and develop a digital tool for early MCI detection and driving risk prediction.