How does Rtabmap propagate the error back after detects loop closure

[dizeng]

Hi Mathieu,

I am trying to understand how does Rtabmap propagate the error back after detects loop closure?
I mean after you detects two similar images and find there is an error between the pose it estimates, how do you update the past states and update the map?
Is there any paper talks about this in detail?

Thanks.

[dizeng]

Is it using first-order-propatation?
http://ais.informatik.uni-freiburg.de/teaching/ss10/robotics/slides/11-error-propagation.pdf

[matlabbe]

RTAB-Map uses TORO, g2o or GTSAM libraries to correct the graph after a loop closure. Here some links:
TORO: http://openslam.org/toro.html, paper: http://www2.informatik.uni-freiburg.de/~stachnis/pdf/grisetti09its.pdf
g2o: http://openslam.org/g2o.html, paper: http://ais.informatik.uni-freiburg.de/publications/papers/kuemmerle11icra.pdf
GTSAM: https://borg.cc.gatech.edu/projects/gtsam?destination=node%2F245

cheers