sentience is a software library that allows for robotic stereo vision using stereo webcams (like Minoru 3D, which is of British origin despite its Japanese name), and is written in C#. Meanwhile, ARToolKit is one of the most widely-employed augmented reality software frameworks (and is also, like sentience, FOSS). 

A pet project that I want to do starting at the beginning of Spring semester is to hook up sentience with one of the various language-specific implementations of ARToolKit with sentience so that ARToolKit can make use of a stereo camera to plot a depth map in which to place virtual objects. The idea is that, since traditional AR has issues with telling a real foreground from a real background so as to place the virtual objects somewhere in the middle, a stereo webcam and a depth-perceptive plugin can allow for ARToolKit to accomplish this feat.

This, I believe, will result in tru-er AR interfaces which are blended into – not onto – the real world. Virtual keyboards, for example, will be usable as floating visual tools by which text communication can be typed into practically "thin air"; heck, that means that, if AR glasses or contact lenses ever take off, laptops will be replaced with virtual, thin-air equivalents in terms of their purpose as mobile, compact desktops-on-the-go.