Hi Aaron,

I just replied to your question in answers.ros.org. I suggested two
things, and it looks like we are thinking along the same lines.

I strongly agree that the best way to do this is to modify
slam-gmapping to accept point clouds. The benefiits of this are many:
 - doing slam with mutliple lasers
 - doing slam with lasers with varying tilt/roll, for example, on quadrotors
 - doing slam with non-laser sensors directly (kinect)

In my research, I have had to hack slam-gmapping to go around these
limitations. If you are interested, perhaps we should join efforts,
and maybe get the gmapping- developers on board, to rewrite the
gmapping wrapper to provide for PointClouf input efficiently.

Ivan