map_optimizer node: iterations and strategy

[KamalDSOberoi]

Hello,

In the map_optimizer node, we have the option to chose the optimization strategy and the number of iterations. For g2o strategy, what is the ideal number of iterations we should use? And what is the effect of using less or more number of iterations on the optimization process?

Thanks for your help!!

Regards  

[matlabbe]

Hi,

For GTSAM and g2o, they converge quite fast (under ~5 iterations). You can open a database in rtabmap-databaseViewer to play in the Graph View panel with the number of iterations and see live the map changing. There is a parameter called "Optimizer/Epsilon" that can be used to stop optimizing when global error decreases under it (if you set to 0, all iterations set by "Optimizer/Iterations" will be done):

$ rosrun rtabmap_ros rtabmap --params | grep Optimizer/Epsilon
Param: Optimizer/Epsilon = "0.0001"                        [Stop optimizing when the error improvement is less than this value.]

You may look at the corresponding g2o or GTSAM papers for more info on the convergence.

cheers