Visual SLAM and Cartographer

All SLAM implementations use an algorithm to match the current map features with previously seen features. Although many videos on the internet show this in action, many engineers have a hard time to trigger this condition and sometimes have to give up. The amount of parameters and combinations makes it very hard to find a stable configuration without expert knowledge.

Even if an initial map can be created, a second effect then kicks in which is the aging of the map features, meaning, the features seen a long time ago are no longer present in the environment. So in changing environments, remapping must be done on a regular basis, but this will deteriorate the quality of the initial map if not done properly.

SLAM algorithms have to calculate the current position and orientation by comparing the features detected by the sensors with those in the previously built map. This calculation can not be done with absolute accuracy due to noise in the measurements. This is worsened the more you deviate from a previously mapped path.