Revision history [back]

Sorry for the delay, ROScon and DRC qualification took their toll ;) So the internal state basically boils down to the state estimate of the robot and that of the map. The former ist just a 2D/3DOF pose and the map is represented by multiple grid map layers.

If you look at HectorSlamProcessor.h (note we recently moved to github, too) it boils down to keeping a copy of the MapRepresentationInterface and the last scanmatch pose somewhere. If something goes wrong, this data could be retrieved and thus be used to revert the broken map. It likely isn't too hard to do, but still requires changing some of the internals. Also, reliably detecting things going wrong might or might not be easy.