We are a fair way from the genuine 'magic helmet-mount surveyor', but there is steady progress in this direction so I have reason to believe it may become practical quite soon and maybe even affordable and useful a few years after that.
The kinetic games thingy is actually quite a useful surevyor - it's already been used for office surveys. It has limited range (about 5 metres), will quickly go off track, and a has a somewhat limited cone of view that works well in square corridors but will probably miss a lot of cave features unless you go forward and backwards, but as a way of filling in shape around an existing survey it's very good.
Anna Mason did an MSc on using video plus film tracking software to generate underground surveys and essentially showed that it could be done with expensive cameras/lights and software.
http://www.soi.city.ac.uk/~dmm/research/pubs/2008-MasonA-gisruk.pdf] [url]http://www.soi.city.ac.uk/~dmm/research/pubs/2008-MasonA-gisruk.pdf[/url] Working out how to do this to adeqaute quality using cheap cameras/lights and software would be a very fruitful area of research - I'm pretty sure it could be done. Again, this technique probably still needs a centreline to work on for good-enough accuracy, although it can in theory gfenerate it's own if you just stick identifiable markers on the wall somehow.
Meanwhile people are doing a lot of accelerometer and computer vision mapping work. I'ts quite true that accelerometers we can afford have apalling accuracy and will drift completely out of whack in about 5 seconds, but you can do things to dramatically improve this. Someone at a talk I went to found that attaching one to the foot works wonders, because it stops every time to put the foot down for a step - this allows a calibration (detecting zero vertical momement and very little forward movement) every couple of seconds. That probably still works underground most of the time. Caving has an advanatage over most uses of this tech, in that it is easy for us to stop, put the magic box down and let it recalibrate to 'still'. This is the only way that cheap inertial tech is going to be any use anytime soon.