Scenario-based robust scheduling for collaborative human-UAV visual search tasks

Many decision support algorithms used to aid human decision making provide guarantees of optimal performance when the optimization parameter are perfectly known. However, algorithm performance degrades when the parameters are inappropriately estimated, and furthermore, algorithm performance is also sensitive to the uncertainty arising from human oversight and interaction with the algorithm. This paper discusses decision support in the form of a scheduling algorithm designed to support human search of imagery collected by unmanned aerial vehicles (UAVs). We demonstrate the sensitivity of the algorithm to uncertainty in human search times, and present a new robust formulation for the search recommendation using data obtained from a previous human-in-the-loop experiment. We show that this robust formulation results in fewer constraint violations in the search task recommendations, as well as increased average performance than an algorithm that does not take this uncertainty into account. For the final draft, the goal is to present human-in-the-loop results that confirm the predictions obtained in the simulations.