A Driver-in-the-Loop Study of Human-AI Joint Replanning

Abstract

This study investigates human-AI teaming requirements in the context of ill-structured replanning tasks, where evolving goals, ambiguous constraints, and shifting environmental states create coordination demands with AI agents. We conducted a high-fidelity staged world experiment using a driver-in-the-loop simulation, wherein participants completed two event-driven driving scenarios to explore how AI capabilities can aid cognitive work of drivers. These prototypes embodied distinct functionality and coordination strategies grounded in a Flexecution model of replanning. Our analysis draws from post-hoc interviews to evaluate how participants navigated the trade space between the cognitive benefits of AI support and the coordination overhead of engaging with the system. Results reveal that environmental events modulate interaction strategies and suggest design implications for AI teammates that better support human adaptation under varying workload conditions and temporal constraints.