I stumbled across a small issue:
When the SLAM mode is activated, aka when one of the related topics is subscribed to (/pose,/quality...) The memory usage of the node starts growing when the slamdunk is moved.
But the memory usage never diminishes, even when the SLAM mode is deactivated.
And after a few minutes (7 in one case), memory space is full and the node crashes.
This happens even when the node is the only process launched (apart from a small rostopic echo or hz).
In this context, the slamdunk is pretty much unusable for its intended use for a meaningful length of time.
What could we do to solve this?