Goal

Determine whether matching real-world and virtual-world latency levels in AR systems produces better task performance than simply minimizing overall latency — with direct implications for AR training system design.

Challenge

Comparing real AR systems experimentally is impossible without losing control over confounding variables. Prior work couldn't isolate latency type (real-world vs. virtual) independently — so the relative effect of each on registration quality and performance was unknown.

Approach

• Built an AR simulation inside VR (CAVE system) to precisely control latency parameters that would be uncontrollable in actual AR hardware

• Tested 7 conditions across 30 participants (12F, 18M), simulating optical see-through (OST) and video see-through (VST) AR at matched and unmatched latency levels:

• Task: military forward observer simulation — tracking aerial drops onto targets at 400–650 ft using both unaided view and 4× virtual binoculars

My Role

Lead researcher responsible for study design, AR simulation platform development, experiment execution, and analysis.

Key Findings

• Matched latency conditions produced significantly better accuracy than unmatched ones (p<0.0001) — even when total latency was higher

• High VST matched (75ms/75ms) outperformed low OST (0ms/25ms) — a higher-latency system with perfect registration beat a lower-latency system with misregistration

• Misregistration error grew with later targets — users in unmatched conditions fell progressively behind, unable to recover

• 3 clearly separated accuracy groups emerged: perfect registration (error ~0.4–1.2), some misregistration (~3.5), significant misregistration (~10)

• Subjective ratings showed significant interaction effects on difficulty, irritation, naturalness, and perceived precision (all p<0.05)

• Key design implication: for registration-critical tasks, AR designers should consider deliberately adding real-world latency to match virtual latency rather than optimizing total latency alone