In August of last year,
after the Quecreek mine accident, US Senators called for robots that
could map mines. Mining engineers claimed the idea was "impossible" and
November, CMU had completed the first test of Groundhog,
an autonomous mine mapping robot. Groundhog can take photos, make 2D
maps, 3D maps and 3D animations. Examples
can be found on CMU's website. This
Friday, Groundhog will take on its hardest test yet: mapping a
3,500-foot corridor in an abandoned coal mine near West Elizabeth, PA.
More details about the development of
Groundhog can be found in this Carnegie
Mellon Magazine article.
Does anybody know how their mapping software works? I wish they'd
explain that better, there are hardly any technical details on the
site. There are two ways to make a map (that I can think of) from a
robot. First is to use wheel encoders to give you your absolute
position and then overlay your laser data (they're using a SICK
laser/ladar) on that. The problem is that wheel encoders won't be
completely accurate, and that inaccuracy adds up the farther you go.
The way to get around that is to find some feature, like a corner, that
is easy to locate and you relate your data back to that point. This is
the harder way, but more accurate if you do it right.
Does anyone know how CMU is doing it? If not, what other sorts of
mapping software have people written using similar sensors?
They probably don't have complete accuracy, just more accuracy than
previous maps. My guess is that they did use wheel encoders. A lot of
the inaccuracy in wheel encoders can be eliminated with some
redundancy in the sensory. It would be far to hard to make reference
to a common point like a corner in this situation. You would have to
have multiple points all along the mine and know the distances between
those points which may not be hard in an ideal situation. I don't know
what your experience of being in a mine is but from my experiences
there are times where there isn't a place that you could make
reference to a point time after time.
Another idea would be to place some kind of beacon every so often on
the wall or ceiling of the mine and use these as reference points. But
this method would add far to much complexity to the system in my
Neat application. We built a similar-looking robot at the University of Alberta back
in 1998/99 for use in the IGVC and for a
geophysical company. It was called Polar Bear, was powered by a gas engine and
instead of the
relatively complicated transmission used on Groundhog, used hydraulic motors. We
used it to track and follow people outdoors (eventually with video in addition to
Videos (1), (2) and (3) of
Bear in action are available. Also, there are a
bunch of photos here.
You can read up on Polar Bear by reading my thesis (available on my McGill webpage). A technical report is
available at the IGVC website.