This paper has gone through several iterations, each one better than the last. The version being published here has a learning slant, and although this wasn't the focus when we started, I'm happy it ended up here given my passion for education.
But that's not to say that the ideas are not applicable to other domains. In fact, I'm hoping that some of you will take the next step. I'd really love to see future work build on this, applying the advice we put forward to future projects and digging deeper into these cognitive theories and others. If you're interested in moving the research forward, feel free to contact me and I'll share some more ideas (I won't be doing this line of research for my thesis anymore).
Without further ado, here is the abstract and a link to the preprint version of the paper.
Abstract: Augmented reality has recently become a popular interface for various learning applications, but it is not always clear that AR is the right choice. We provide a theoretical grounding that explains the underlying value of AR for learning and identify when it is a suitable interface. Our list of operational design advantages includes AR's use of reality, virtual flexibility, invisible interface, and spatial awareness. This list is backed by four underlying cognitive theories: mental models and distributed, situated, and embodied cognition. We argue that the more design advantages a learning system incorporates, the better AR works as an interface. We also identify a set of questions to be used in the design and evaluation of AR projects. With this, we can begin to design AR for learning more purposefully.