MultiThreadedSpinner move base

[Maciej Stępień]

Hi,
I noticed that there is a commented out definition of MultiThreadedSpinner in the move_base_node.cpp. I'm wondering if there is any particular reason that it is commented out and if you are aware of any risks that come with using MultiThreadedSpinner in move_base?

To give a bit more context, in our application map can get quite large and I noticed that when the global costmap's inflation layer updates, it somehow blocks the local costmap's sensor callbacks (ObstacleLayer with LaserScan and PointCloud2) . As observation buffers are not updated, it results in an emergency stop of the robot. I noticed that using even two threads for spinning solves the problem. I'm not yet sure what exactly can block the callback queue - I haven't found anything suspicious in the inflation layer (we also have some custom costmap layers, but it seems to happen during the inflation layer update).
Thanks,
Mac

[Michael Ferguson]

move_base is already multi threaded in a few spots - the planner spins up a separate planning thread and each costmap has it's own update thread. I think adding additional threads for the main loop probably has some risk since it is likely there are things touching the same data that are not appropriately locked.

[Michael Ferguson]

Here is an example: #1173

[Maciej Stępień]

Understood, thanks for your response!