Copter: Land Detector: Include angle checks

[lthall]

This PR fixes #26864 where the aircraft can detect a landing when at full throttle in stabilize and upside down. Now the aircraft will correct detect a crash.

Sorry I didn't explain what is happening here and why this fix works.

The land detector checks that the throttle mix value has dropped to near throttle_mix_at_min as part of the landing checks. To set the throttle mix value to throttle_mix_at_min we have two key checks:

1. Angle requested < 15 degrees
2. Angle error < 30 degrees

It then sets throttle_mix to throttle_mix_at_min.

The land detector checks to see if throttle_mix is close to throttle_mix_at_min before proceeding but this is not the same as checking that it has been set to throttle_mix_at_min. This check passes all the time with the default settings as throttle_mix_at_min = throttle_mix_at_man.

We are then left in a state where we are able to detect a landing without passing the two checks above.

Why does this fix the problem above? The problem above goes through a number of steps:

1. throttle goes high in stabilize
2. stabilise correctly detects a take off
3. aircraft is inverted and left stationary
4. the angle boost takes the throttle to zero
5. the land detector checks throttle_mix and finds that it is equal to throttle_mix_at_min and throttle_mix_at_man so it passes
6. a landing is detected with a high throttle request
7. the takeoff flow of control error is triggered because the aircraft should not have detected a landing

Crash detection:

The crash detection now works correctly because we are not detecting the landing. As Pete says below this is a race and we simply need to not detect a landing to allow the crash detection to work correctly. In this case the aircraft will repeatedly detect a landing and a takeoff meaning the aircraft would never disarm.

Tricky little bugger.

[MattKear]

Those checks seem sensible for checking for a successful landing. However, I think that this indirectly fixes the issue. The problem to me seems to be that we cannot reliably infer if we are taking off from flightmode->is_taking_off() in stabilise because that mode doesn't have the user_takeoff flag. I think a more direct fix to the issue would be to bring the has_user_takeoff() check, up to line 56 of land_detector.cpp: https://github.com/ArduPilot/ardupilot/blob/3c8b3be7a730255927d64265bcaf0d98ebb3c0b1/ArduCopter/land_detector.cpp#L56

So it would look something like this: if (has_user_takeoff(false) && !flightmode->is_taking_off() && motors->get_throttle_out() > get_non_takeoff_throttle() && motors->get_spool_state() == AP_Motors::SpoolState::THROTTLE_UNLIMITED) {

This would mean that we can then only use this check for the modes that support the user_takeoff() function.

[rmackay9]

@Gone4Dirt,

Thanks for the feedback. This is Leonard's work but this Internal Error and the set-land-complete() after it is quite important because it is essentially an independent check that the flight modes have fulfilled their responsibility to clear the land-complete flag. If we add the "has_user_takeoff" condition then I worry we are basically giving a get-out-of-jail-free pass to those modes that don't "have user takeoff" (which includes Stabilize and Acro)

For reference, each mode has this "has_user_takeoff()" method available and will apparently return true if it will accept a MAVLink command to takeoff to a specified altitude. So even in Loiter and AltHold modes a GCS can send a takeoff command to ask the vehicle to climb to X meters and it will do it without the pilot touching the sticks. You can try it yourself in SITL by doing this:

1. start SITL copter
2. loiter
3. rc 3 1500
4. arm throttle force
5. takeoff 10

[rmackay9]

By the way, this PR is failing the Heli autorotation autotest which I'm sure we're going to need to look into. A random guess is that during auto rotation its not landing flat and so the new landing-detector conditions are not being met.

FAILED STEP: test.Helicopter at Tue Apr 23 06:40:59 2024 Failure Summary: test.Helicopter: AutoRotation (Check engine-out behaviour) (Failed to DISARM within 120.000000s) (see /tmp/buildlogs/HeliCopter-AutoRotation.txt) (duration 10.23026156425476 s) FAILED 1 tests: ['test.Helicopter']

[IamPete1]

The new checks added here are the same as those in update_throttle_mix here: https://github.com/ArduPilot/ardupilot/blob/c9f7a3c03d7ef58e328c7ebbe6b4be0d8b030394/ArduCopter/land_detector.cpp#L230-L235

So if they are set then throttle mix will also be set to max, so the existing update_land_detector check for throttle_mix_at_min will fail. I don't think there will be any change in behavior in none manual throttle modes? The only change would be in a manual throttle mode with throttle at zero? That would be this case in update_throttle_mix: https://github.com/ArduPilot/ardupilot/blob/c9f7a3c03d7ef58e328c7ebbe6b4be0d8b030394/ArduCopter/land_detector.cpp#L222

This PR fixes #26864 where the aircraft can detect a landing when at full throttle in stabilize and upside down. Now the aircraft will corret detect a crash.

I'm a little confused by this description. I would have expected full throttle in stabilize to result in throttle mix man so the throttle_mix_at_min would not pass and landing would never be detected.

You mention that the aircraft now correctly detects a crash, but there is no change to crash detection here? Maybe there is a race and currently landing detection triggers first? So by preventing the landing detection the crash detection will trigger after slightly more time?

The TLDR is that I would have expected the existing throttle mix check to protect against this case.

[lthall]

Those checks seem sensible for checking for a successful landing. However, I think that this indirectly fixes the issue. The problem to me seems to be that we cannot reliably infer if we are taking off from flightmode->is_taking_off() in stabilise because that mode doesn't have the user_takeoff flag. I think a more direct fix to the issue would be to bring the has_user_takeoff() check, up to line 56 of land_detector.cpp:

https://github.com/ArduPilot/ardupilot/blob/3c8b3be7a730255927d64265bcaf0d98ebb3c0b1/ArduCopter/land_detector.cpp#L56

So it would look something like this: if (has_user_takeoff(false) && !flightmode->is_taking_off() && motors->get_throttle_out() > get_non_takeoff_throttle() && motors->get_spool_state() == AP_Motors::SpoolState::THROTTLE_UNLIMITED) {

This would mean that we can then only use this check for the modes that support the user_takeoff() function.

I finished the PR description and I explain there why this is actually a landing detection problem. In this case the takeoff is being correctly detected in both Stabilize and by the flow of control error.

[lthall]

The new checks added here are the same as those in update_throttle_mix here:

https://github.com/ArduPilot/ardupilot/blob/c9f7a3c03d7ef58e328c7ebbe6b4be0d8b030394/ArduCopter/land_detector.cpp#L230-L235

So if they are set then throttle mix will also be set to max, so the existing update_land_detector check for throttle_mix_at_min will fail. I don't think there will be any change in behavior in none manual throttle modes? The only change would be in a manual throttle mode with throttle at zero? That would be this case in update_throttle_mix:

https://github.com/ArduPilot/ardupilot/blob/c9f7a3c03d7ef58e328c7ebbe6b4be0d8b030394/ArduCopter/land_detector.cpp#L222

This PR fixes #26864 where the aircraft can detect a landing when at full throttle in stabilize and upside down. Now the aircraft will corret detect a crash.

I'm a little confused by this description. I would have expected full throttle in stabilize to result in throttle mix man so the throttle_mix_at_min would not pass and landing would never be detected.

You mention that the aircraft now correctly detects a crash, but there is no change to crash detection here? Maybe there is a race and currently landing detection triggers first? So by preventing the landing detection the crash detection will trigger after slightly more time?

The TLDR is that I would have expected the existing throttle mix check to protect against this case.

Sorry Pete, I addressed your questions in the description above. In short you are correct with everything you said but there are two additional points

1. throttle_mix_at_min = throttle_mix_at_man.
2. angle boost means the output throttle is zero.

[lthall]

I have confirmed the failing test is due to a problem with the heli autorotation code rather than the landing detector. We should not be detecting a landing with a lean angle of 50 degrees:

@bnsgeyer What do you think?

[bnsgeyer]

@lthall first of all, I did not spend much time reading thru all of the posts but this PR has me concerned with inverted flight for heli which currently doesn’t have a test in the autotest.

As for the failing autorotation test, this may be due to the fact that I am only looking for the declaration that it is in glide phase. The code as it currently stands doesn’t have the ability to autonomously complete the landing. So I was probably just letting it hit the ground and declare landed to stop the test. I will need to look into this further.

[lthall]

@lthall first of all, I did not spend much time reading thru all of the posts but this PR has me concerned with inverted flight for heli which currently doesn’t have a test in the autotest.

As for the failing autorotation test, this may be due to the fact that I am only looking for the declaration that it is in glide phase. The code as it currently stands doesn’t have the ability to autonomously complete the landing. So I was probably just letting it hit the ground and declare landed to stop the test. I will need to look into this further.

There are a number of other checks but yes, this PR should make that safer.

[MattKear]

I had a bit of a look at the autorotation autotest. The test is failing because the aircraft it not disarming after landing. But I am really not sure why this PR is causing it to not disarm though. Desired and actual roll and pitch angles are all close to 0 and the aircraft is neither climbing or descending. So I don't really see how the changes in this PR are causing this to fail.

[bnsgeyer]

@lthall I fixed the AutoRotation autotest. For some reason the autorotation mode caused the aircraft to pitch forward on landing. The changes that I pushed to your branch will fix the test failing. I had to push a second commit to cleanup changes I made to the file for testing it. Anyway, it should pass CI now.

[lthall]

Legend Bill, Thanks!