ArduPilot EU Dev Call 2026-07-29

Development Team
1

Attendees (unique): 10

UTC0802

Andrew: We need to compare build size against the PPP build.

  • Also what is the achievable bandwidth.
  • PPP is the natural thing to compare against. It needs to beat PPP in some way in order to be considered for merging.
  • It is crucial to know if we need special configuration in the host to connect to the device. If it works plug-and-play, then this will also be a benefit.

UTC0810

Andy: I’ll review it, it slipped past me.

UTC0812

Peter: Do we lie about arming stage?
Andy: By default the chip will look at the arming state to go from low power to high power. So we lie in order to keep a high power output always.
P: It’s good that this feature is opt-in.

Merged!

UTC0816

P: It doesn’t build, let’s take a look and rebase.

UTC0818

Andy: We are have a new KF for the optical flow distance, but to switch to another hight source we are still using the raw rangefinder measurements.

  • We should use the KF output.

A: terrainHgtStable is the wrong trigger to start fusion of optical flow.

  • It broadens when EKF accepts optical flow data, which is likely to cause problems.

Andy: This is mostly for indoor flying.
P: Then let’s have a boolean in the EKF that flags this.
R: I think it would be holding position well over a sloped surface.
A: Sure, but the EKF HAGL estimate deriving from optical flow might still be wrong, when flying above a slope.

  • I can test it over a slope.

UTC0838

Approved!

UTC0847

P: Bob Long pointed out that a back-transition into QLOITER has started swirling around the target.

  • The rate PIDs seem to be happy, yet we have a lot of roll error.

UTC0849

P: The nulldereference will now trigger an error.

  • But sometimes the null-dereference warnings are wrong, and we have to make the compiler happy in these cases.

A: The resulting size if smaller, even though we added more checks? This is worrisome.
P: gcc when it thinks there can be a nullptr derefence, adds some code in to handle the potential. When you check it manually, that extra code goes away, saving size.

  • I’ll bring this back when I make sure this is what happens.

UTC0859

P: More tests for triggering all camera backends at once.

UTC0901

A: I fixed a threading issue on a branch of my own.

  • We have to be careful when trapping on reads, to avoid falsely trapping on innocuous reads.
  • I’ll push mine and we can compare notes.

UTC0906

P: Apparently Bob thinks the back-transition funkiness is related to wound-up state during forward-transition.

UTC0908

Andy: Allow for more total notches when configuring quintuple notches.

  • Will allow for more notches in general.

P: Why not add more, since there’s no compile overhead?
Andy: We try to protect the user from accidentally configure too many and overload the CPU.

Merged!

UTC0913

A: As long as we don’t try to allocate the mission at runtime, where we risk a nullptr.
P: It is always non null.

Merged!