Impacts of AI-Role-Based Multimodal Takeover Messages and Affective State on Automated Driving Takeover Performance

Automated vehicles may still require drivers to resume control when system capability is exceeded. This study examined whether AI-role-framed takeover requests (TORs) and multisensory cues influence driver decision time under different affective states. Twenty-two licensed drivers completed 18 trials in a medium-fidelity SAE Level 3 driving simulator. A 3 × 3 × 2 within-subjects design manipulated AI role (Advisor, Co-pilot, Guardian), multisensory TOR condition (auditory + visual, auditory + tactile, auditory + olfactory), and affective state (positive valence/high arousal vs. negative valence/low arousal). Decision time and subjective ratings of satisfaction and usefulness were collected. Results showed significant main effects of AI role and multisensory condition on decision time. Co-pilot produced the longest decision time, while Guardian produced the shortest. Visual cues were associated with shorter decision times than tactile and olfactory cues. Affective state interacted with AI role. Subjective ratings showed no significant differences across conditions. Findings inform role-based, multisensory TOR design for automated vehicle systems.