complicated reasoning and was by far the most difficult task. The real AR portion revealed differences in task performance in three task questions (Q3, Q6, Q8), which could either be due to the lighting artifacts in the high-dynamic range scene or actual differences in scene realism. Overall, these minimal differences are promising for the validity of MR Simulation, and the difficulties of conducting outdoor real-world experiments demonstrate the usefulness of MR Simulation. In future work, we would like to clarify through follow-up studies what the exact cause of the differences between the AR case and the simulated AR cases was. Since it became very clear to us that high-dynamic range lighting coupled with outdoor HMD us uage had a detrimental effect on distinguishing small physical objects in the real
scene, we will look at better ways to present AR imagery to users, i.e. at improving the actual AR experience. Maybe an interesting intermediary simulated AR condition could also be using a (well-lit and adjusted) panoramic image backdrop with the same annotations as in this study, sacrificing motion parallax and real sensory input (being situated in an actual outdoor environment) for better image visibility
and experimental control. This would give evidence if the differences in task performance were indeed due to lighting and resolution differences or due to the modeling abstractions.