Simulating Edge AI Interventions: An HCI Proof-of-Concept for Dynamic AR-HUD Decluttering Based on Physiological Monitoring

Advanced Driver Assistance Systems (ADAS) face competing demands among environmental perception, physiological monitoring, and cognitive overload. This paper proposes a sequential Edge AI software architecture capable of executing Traffic Sign Detection and Recognition (TSDR) alongside incabin biometric analysis. Deploying these models on constrained hardware introduces latency bottlenecks and resource contention. To evaluate the human-computer interaction (HCI) requirements of this system, we conducted a Hardware-in-the-Loop simulation using a Wizard of Oz methodology. Six participants navigated complex routes using a simulated Augmented Reality Head-Up Display (AR-HUD) and optical heart rate monitors. Empirical data indicate that static, data-dense interfaces induce cognitive saturation and navigational panic. By dynamically decluttering the AR-HUD during high-stress events, the system managed driver arousal and mitigated visual masking. These findings detail a closed-loop system design that balances Edge AI scheduling limitations with cognitive ergonomics.