Hi, Mathieu,
I am appreciating your work on Rtabmap, just want to ask a question on better 3D map quality. I am currently using RGBD with lidar setup. The map is just a bit messy, I wonder where is the noise coming from. I have a few things in mind:
1. The tf between RGBD and Lidar is not perfect, so when using both for loop closure, there might be some error. Also, a question i have is that, it seems lidar are only used when loop closure are detected from RGB image, then it does more pose adjustment. There was one instance where the robot got stuck around a corner, so it is unable to get loop closure, and the robot just drift further away from the "actual position". However, from grid map and lidar value, I can clearly see the robot is delocalized. Is there a way to use ICP between neighboring links and map?
2. Our RGBD camera is publishing at 6 Hz, and lidar at 8 hz. We set approx_sync=True, but there are still discrepancy between the two sensor message time-stamp. How big of a problem is this going to cause? And how should i approach it?
3. RGBD camera does not work that well beyond 4m, and it seems Rtabmap is still assembling the map using further away points. Is there a way to only use closer points? I can see how it would make pose adjustment harder.
This is an example of my databse:
https://drive.google.com/file/d/15pcpFikW2yROFGiiKHEgz_ekB63UYmsE/view?usp=sharing
Locked
