Parameter Tuning

[frapit]

When I move the L515 through the hall, I notice very inconsistent behavior with the node. In the beginning it seems to work but when the sensor is almost standing still, the predicted position and map starts to drift away spirally until I get a
[pcl::VoxelGrid::applyFilter] Leaf size is too small for the input dataset. Integer indices would overflow.
from PCL library:

The scenario you showed in the video does not suffer from such problems. Did you perform some additional parameter tuning or can the node only be used in a small amount of robust scenarios?

[wh200720041]

Hi @rapit :

The cause of "Leaf size is too small for the input dataset. " is that you set the map resolution too small. However, I can't tell the exact problem since the map is too messy. Maybe a photo of camera view helps.

As you may know, the SLAM relies on the matching of geometric features, the localization fails when there is not features detected. For example, you put the sensor in front of a white wall and move the sensor parallel to the wall, the matching easily fails since it is hard to tell the changes. This is my guess based on the photo you provided

You are welcomed to leave any rosbag so I can have a quick test.

Regards
Han