MIT and the National University of Ireland have come up with a new algorithm to use a Kinect camera to 3D map passages as you go.

Will be limited to the effective range of the Kinect and to spaces where you can return to the start point. Would be interesting to see if it will work in a cave.

http://web.mit.edu/newsoffice/2013/3d-mapping-in-real-time-without-the-drift-0828.html

At the ICS in Brno this year a paper by Matt Covington, Jason Gulley, and David Ochel was presented, showing the use of an Xbox Kinect for detailed mapping of glacier caves in Nepal and Svalbard.

I believe their work was specifically around the rate of melting of the ice walls throughout the summer, so they needed something to map the walls in a quick and detailed way...

I spoke with Matt afterwards, he said it worked but was still very difficult because you can get drifting of the surveyed tubes as they get built. To correct this they tied the mapping into a "permanent" centreline. Unfortunately the results weren't ready for display in Brno, but they still managed to fill the largest lecture theatre there!

Maybe the algorithm you link to has overcome some of the problems, but i think there is still quite a way to go, especially if you are not able to close loops. The example on the video at 3:00 minutes just shows how bad it can be if there is no loop closure! Maybe if someone clever can build in a compass and/or clino...

Also the Kinect camera runs on a IR splay matrix, with a short (~5m) range. Perfect for most UK caves, but maybe a bit tedious not for any big stuff....

I wonder what the minimum Kinect range is? When gaming it seems to be about 5' but I guess that is to fit a person in the camera frame.