Occupancy Grid SLAM

[motters]

Some initial examples of the occupancy grid mapping algorithm. This
is based upon DP-SLAM, but extended to 3D and with the use of sensor
models (the original algorithm described by Austin Eliazar and Ronald
Parr didn't use any kind of sensor model since they used a scanning
laser rangefinder and just assumed that the returned range values were
completely accurate).

Initially I thought that 3D would be too much for this algorithm to
handle, since it's already keeping track of multiple grid map
hypotheses. At each point in time the robot exists in a superposed
state, which collapses to a single result only after enough
observations have been made (similar to quantum superposition).
Fortunately though using three dimensions instead of two actually
speeds up the algorithm, since it makes searching for matches more
localised.

http://sentience.googlegroups.com/web/dpslam1.jpg

http://sentience.googlegroups.com/web/dpslam2.jpg

There's still a lot of drift going on and more bugs to be fixed before
this is a practical system capable of being installed onto a robot,
but the results so far look encouraging.