Revision history [back]

GMapping (and, in fact, all 2D-scan-based mapping techniques) are deeply dependent on the quality of the data that they work with; in particular, it's very important that the scans be unambiguous, which is quite hard when using a Kinect for gmapping (where the short range and narrow field of view make it very hard to get unambiguous scans).

While mapping, do your best to always be able to see things (say, a wall) and have non-trivial structures, like wall corners, in view while you do this. Anecdotal evidence suggests that mapping anything larger than a medium-sized room with Turtlebot (using the Kinect in this way) is impractical, because of the sensor limitations.

On the other hand, if you mount up a laser rangefinder, that's an entirely different story...