Human-Aware Agentic Driving: Rethinking Driver State, Trust, and Intervention

Abstract

Agentic driving brings goal-directed reasoning, memory, tools, and adaptive interaction into automated vehicles. A vehicle can clarify a request, explain a maneuver, or prepare a driver to take control, then adapt its next action to the response. These capabilities make the vehicle an active participant in shaping the human behavior it observes. We examine how agentic interaction affects driver-state inference, trust and reliance, and human intervention in conditional automation. Established findings on attention, takeover prediction, trust-aware planning, and adaptive warnings provide useful foundations, but their applicability depends on how observations are elicited and how assistance changes the driving task. We distinguish what remains applicable, what changes under flexible interaction, and what remains unresolved across these three areas. The analysis identifies three research priorities: establishing the effects of interventions, jointly designing interaction and driving policies, and maintaining useful adaptation across repeated use and system changes. Progress should be measured through driver comprehension, appropriate reliance, and the ability to perform the required driving action in time.