Effects of System Reliability on Workload and Performance in Image Recognition Tasks

Abstract

Autonomy has found wide-ranging applications, yet its imperfect nature necessitates human oversight and intervention. Investigating autonomy's impact on the operator is pivotal for enhancing human-machine system performance and safety. This study analyzes the effects of autonomous system reliability on operator task performance and mental workload in the context of vehicle type recognition. Experimental findings reveal that autonomy with 90% reliability significantly reduces task completion time and lessens subjective workload. Autonomy with 70% reliability supports the participants, while 50% reliability hampers them, although insignificantly. The reliability threshold for autonomy to have no effect on the participants is around 55%. Autonomy reliability's influence on the operator lies in altering task completion strategies — an all-or-none approach that accelerates task processing speed without improving overall response accuracy. The experiment yielded insights applicable to the design of assistive autonomous systems and the allocation of human-machine functions in real-world tasks.