wh200720041 / floam

Fast LOAM: Fast and Optimized Lidar Odometry And Mapping for indoor/outdoor localization IROS 2021

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Another optimizer

EPVelasco opened this issue · comments

Hello.
Thank you very much for the FLOAM code.
I am working with a UGV traveling at 10 km / h that has a Velodyne VLP16 and a Jetson Xavier AGX. I have tried other optimizers from the Ceres library.
I modified the code (odomEstimationClass) on line 63, changing ceres :: DENSE_QR to ceres :: SPARSE_NORMAL_CHOLESKY; and I had shorter times than with the previous optimizer.

Why does the loop on line 52 of the code (odomEstimationClass) start to decrease to only 2 iterations?
I have removed this and make the loop always do only 2.

Hi @EPVelasco

More iterations are needed for better initial state estimation, since in the beginning we do not know the status of the robot. Hence we decrease the iterations when system start.

You can make that fixed at 2 if there is no issue on your robot