Hidden Cost of Re-Engagement: Workload in Dynamic Multi-Operator Supervision of Autonomous Systems

Autonomous systems are increasingly shifting from single-agent oversight to distributed fleet supervision, where operators alternate between monitoring, waiting, and brief intervention. We examined whether multi-operator, multi-agent supervision preserves objective performance while increasing subjective workload. In a 2 × 2 × 2 between-subjects online experiment, participants completed supervisory tasks after passing a proficiency threshold. Participants were assigned to either single-operator/single-agent supervision or three-operator/six-agent supervision with a shared FIFO task queue. Workload schedules varied by intervention intensity and task-distribution evenness. Participants completed landing, flying, supervisory flying, and monitoring tasks. Performance was measured using baseline-normalized accuracy and task duration; subjective workload was assessed using NASA-TLX subscales. Linear mixed-effects models tested effects of supervision paradigm, workload intensity, task distribution, and interactions. Multi-operator supervision did not degrade normalized accuracy, but operators reported lower perceived performance and greater effort. These findings suggest that distributed supervision can maintain short-term performance while imposing hidden workload costs.