User Performance and Trust with Network-Connected Augmented Reality Head-Worn Displays in Unstable Network Environments

Real-world augmented reality (AR) systems are increasingly network connected in order to succeed in operational spaces dependent on cutting edge sensor and computing technologies, all connected together via the network. This dependency inevitably introduces new sources of error to the AR system through instability in network performance. These errors create familiar problems, such as registration errors, on scales unfamiliar to modern AR systems. The effects of these problems on user performance are difficult to investigate as user performance involves many interrelated factors, some of which are difficult to measure. In this work, we control for user behavior to reduce the complexity of investigating the effects of real-world scale latency and signal dropouts on users’ objective task performance and subjective assessments of the AR system. Behavior was controlled by instructing participants to adopt specific behaviors based on different levels of trust and reliance on AR tags and environmental visual cues. We found significant negative effects of both network error sources on subjective assessments of both the AR system and task, as well as objective task completion time and errors committed. Participants’ subjective assessments and objective task also varied unexpectedly with increasing network error levels, suggesting that human perception of the effects of network errors are not well calibrated to the actual effects on performance. We also found significant interaction between behavior and network errors on task performance, drawing implications not just for AR system design, but also for AR system operation guidelines.