search

gisbi-kim / SC-LIO-SAM

LiDAR-inertial SLAM: Scan Context + LIO-SAM
619 stars 174 forks source link

Question about the parameter historyKeyframeFitnessScore #2

Closed Xiangzhaohong closed 3 years ago

Xiangzhaohong commented 3 years ago

you set the parameter historyKeyframeFitnessScore: 1.2 But, it may bring some wrong loop factor, why you choose 1.2 ? thanks a lot!

gisbi-kim commented 3 years ago

no particular reason for that :) but I just wanna use a slightly larger value than the original repo, because I found some actually true loops sometimes have ~1 fitness scores so I didn't wanna miss those loops.

If wrong loop factors occured, it's okay because we use a robust kernel for loop factors, (see below lines) https://github.com/gisbi-kim/SC-LIO-SAM/blob/6829a5ba3104096d43fd30233929d5e0b821b4eb/SC-LIO-SAM/src/mapOptmization.cpp#L790 so we can reduce those wrong loops effects when enough true loops exist.

ps. empirically, I tested the harshest cases where the number of wrong loops was maximized when the scan context descriptor threshold = 1.0 (i.e., all loops were accepted). below is the result and the trajectory still looks good.

cauchy_good1

Xiangzhaohong commented 3 years ago

ok！thanks ! I just noticed that your video have many loops in the begining , but the results is good! image