Task switching and single vs. multiple alarms for supervisory control of multiple robots

Foraging tasks, such as search and rescue or reconnaissance, in which UVs are either relatively sparse and unlikely to interfere with one another or employ automated path planning, form a broad class of applications in which multiple robots can be controlled sequentially in a round-robin fashion. Such human-robot systems can be described as a queuing system in which the human acts as a server while robots presenting requests for service are the jobs. The possibility of improving system performance through well-known scheduling techniques is an immediate consequence. Unfortunately, real human-multirobot systems are more complex often requiring operator monitoring and other ancillary tasks. Improving performance through scheduling (jobs) under these conditions requires minimizing the effort expended monitoring and directing the operator's attention to the robot offering the most gain. Two experiments investigating scheduling interventions are described. The first compared a system in which all anomalous robots were alarmed (Open-queue), one in which alarms were presented singly in the order in which they arrived (FIFO) and a Control condition without alarms. The second experiment employed failures of varying difficulty supporting an optimal shortest job first (SJF) policy. SJF, FIFO, and Open-queue conditions were compared. In both experiments performance in directed attention conditions was poorer than predicted. A possible explanation based on effects of volition in task switching is proposed.