Eyes Off the Road: Where Drivers Look When ADAS Control Authority Changes

As Level 2 driver assistance becomes standard, drivers supervise automation that varies in feature bundling and system-state communication, which we term ADAS architecture. But does this architecture shape where drivers look and how they manage control transitions? This study examined three architectures: Modular (features activated separately), Integrated (functions bundled under single control logic with context-based transitions), and Hybrid (longitudinal functions combined while lateral control remains separate). Thirty-six drivers completed a simulator drive while gaze was tracked using a deep-learning pipeline synchronized with telemetry. Functional capabilities were held constant; only the coupling and communication of control authority varied. Results showed that Hybrid produced stable, road-focused gaze and less unnecessary interface checking, while Modular drivers spent significantly more time verifying HUD status. Lateral control elicited stable monitoring, suggesting steering automation is sensitive to architectural design. These findings position ADAS architecture as a key design criterion for visual attention in Level 2 systems.