Overloaded, underloaded or in control: How many automated vehicles can one person supervise?

Despite extensive research over the past two decades, the question of how many automated vehicles (AVs) or robots an individual can effectively supervise remains unresolved, with estimates ranging from as few as two, to as many as 12. Most prior studies have conflated monitoring and direct interaction tasks, leading to inconsistent findings largely driven by variations in interaction complexity and duration. This research study addresses this issue by isolating the monitoring task from the interaction task to establish a more precise baseline of supervisory capacity for AV systems.

A rigorous experiment was conducted wherein 24 participants monitored between three and nine simulated AVs operating within a realistic sub-urban environment modelled on Coventry, a mid-sized city in the UK. Unlike experiments in previous studies, participants were tasked exclusively with monitoring AVs to identify those requiring potential manual intervention, subsequently delegating interaction to a separate remote operator. Performance metrics, perceived workload, situation awareness, and decision-making efficacy were systematically measured and analysed.

The results reveal situation awareness (SA) was maximised at when supervising five Avs, and optimal monitoring occurred when supervising 5–7 AVs, with the capacity to temporarily manage surges of up to 9 AVs without significant performance degradation. However, supervisors assigned to monitor as few as 3 AVs exhibited tendencies toward micro-management, often misidentifying situations requiring manual intervention and unnecessarily escalating control handovers. These findings have significant implications for developing scalable AV supervision systems, where appropriately calibrated monitoring loads can enhance performance and decision-making while minimising erroneous interventions.