Pose graph SLAM, pose error

84 views
Skip to first unread message

L D

unread,
Feb 19, 2021, 10:11:13 AM2/19/21
to gtsam users
Hi,
I am applying pose graph slam on 2D lidar data. I cumulate 2D lidar scans using ICP initialized with odometry.
I detect loop closures based on 2D lidar features matching and Ransac. This works well but sometimes I get wrong loop closures due to wrong matchings.
I want to reject these false alarms based on geometric distance between matched pairs. I have difficulties figuring out the distance threshold to apply.
I want to be able to decide a matching is clearly impossible due to relatively big distance separating matched points.
Since there is a drift between the current scan pose P_i and the matched scan pose P_j I can't figure out this threshold.
I have tried using the marginals covariances provided by GTSAM, but I get relatively big sigmas (>> 100 meters) from P_j to P_i.
I think I am not using these covariances the right way.
Please, can someone show me what I am doing wrong ? and what am I supposed to do with these marginals ?

Thanks you.





Reply all
Reply to author
Forward
0 new messages