Re: Persistent local scan matching ignored error for Lidar + robot + Odometry + RGBD mapping

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Persistent-local-scan-matching-ignored-error-for-Lidar-robot-Odometry-RGBD-mapping-tp3911p3914.html

Do you have the database without "RGBD/LocalLoopDetectionTime" to share? It seems that is an optimization problem, in particular g2o can be sensitive to covariance set by odometry. Try with TORO or GTSAM: "Optimizer/Strategy" to 0 or 2 respectively. I don't recommend using "RGBD/LocalLoopDetectionTime=true", as it generally doesn't add very much accuracy comparing for as many links added (which increases quite a lot graph optimization time).

If you are going to do multi-session mapping, I also don't recommend g2o as it optimizes poorly multiple maps. GTSAM is the better, however, with many maps, it can happen that GTSAM returns no solution, so loop closures are rejected. TORO is the most stable between the three, though not giving as good optimizations, it will always find a solution in multi-session mapping. A small issue with TORO is when the map is getting very big, it puts sometimes all errors in one link even if the new loop closure is good, thus "RGBD/OptimizeMaxError" should be set to 0 to avoid rejecting good loop closures.  

For the sensor, the map should be better with a 20 m than a 4 m. The robot mapping demo (video) uses a 30 meters UTM-30LX, launch file here.

cheers,
Mathieu