Science sensor system causes glitches in sim

[braanan]

Hi @mabelzhang,

I've been testing the empty_world env with the science sensor system in the loop and it seems like the queries are causing significant slowdowns. As a result the vehicle appears to be glitching along the navigated track.

empty_world_tethys.mp4

The temperature values produced for temperature seem to be correct.

To reproduce run the lrauv-application with the following inputs:
./bin/LRAUV -x "run Science/profile_station.xml"

Thanks

[mabelzhang]

Hey Ben, I was able to reproduce this on main.
However, when I switch to my trilinear interpolation branch in #127, the lag disappears. So, I recommend getting my PR merged so we have all the improvements.
I haven't dug into what is the problem on main causing this. If #127 solves it for us, then that seems the easiest solution.

I also noticed main is still printing a lot of garbage from my old (3 months now) interpolation code, which has been refactored into its own class in #127 and has cleaner printouts now. Definitely a better experience.

Could you try this out with #127 and see if the lag disappears for you?

[braanan]

Thanks for looking into this issue. Yes, I'll merge in #127 to my local and give it a shot.

[mabelzhang]

I did a quick test to switch to barycentric interpolation in #127 by this line in ScienceSensorsSystem.cc:

  this->dataPtr->interpolation.SetMethod(BARYCENTRIC);

to try to isolate if the disappearance of lag comes from the TRILINEAR method, or the infrastructure improvements.
Still no lag.

So I'm guessing the lag might come from infrastructure.

That of course says nothing about whether interpolation is correct, which we should check separately. But seems like things are better with the new code, not worse. In any case, to test whether interpolation is correct, which is probably the next step, we should start from the new code.

[mabelzhang]

I added a few IGN_PROFILE lines. See timelines produced below.
You were right about the kNN search being the bottleneck. FindTrilinearInterpolators() takes 800 ms.
In comparison, publishing point clouds take 30 ms.

[braanan]

resolved in #191