Re: Improving a lidar-only setup

Posted by Robin on
URL: http://official-rtab-map-forum.206.s1.nabble.com/Improving-a-lidar-only-setup-tp11071p11152.html

Hi Mathieu,

Thanks for the time and effort!

The original problem I had with sideways deviations depending on speed seems to be resolved. Not sure yet exactly why though. I've adjusted some settings, but I've also very accurately calibrated the scaling of the wheels for wheel odometry and the 6D pose of the livox to the baseframe. The previous values for those were already pretty accurate to like 5mm or 0.5 degrees. I'll be testing reverting some of those changes and seeing when I can reproduce the problem.

My current parameters:
params.yaml

Most, if not all, of my parameters in my original post were based on your lidar3d.launch.py. So if you're saying that result looks good, maybe my problem was more related to the localization.

The robot is 2m tall, with the livox on top, so yes, it will have noticeable oscillations. About using the IMU for deskewing though, I'm wondering about the effect of velocity (and acceleration). Using imu_to_tf + lidar_deskewing will only account for angle changes, right? So constant robot speed is not accounted for and non-constant swaying of the sensor neither. Only the impact on roll/pitch oscillations would be captured? Would using something like robot_localization to combine wheel odometry + imu be a good approach for the deskewing?

Some other questions:
- Every time I do a mapping, I afterwards have to do rtabmap-recovery, as the database consistently is corrupted. I don't know if there's something I'm doing wrong here. When done mapping, I pause in rtabmap_viz and then I ctrl-C the launch. ("Database is indeed corrupted, found one or more neighbor links missing.")
- What metrics do you generally look at and where (rtabmap-databaseViewer?). Specifically how you would compare the quality of the mapping or localization when changing parameters when there's no ground truth available. Currently I try to use wheel odometry as the ground truth, but that's not that great of course.

Next week I probably will have time to make new ros bags that would be including the IR and RGB images of the two cameras and also calibrated wheel odometry and calibrated TF transforms for livox and cameras.