Flow out-of-focus takeoff support

[LorenzMeier]

Current discussion state:

• Report flow as valid or special mode if landed (if flow data is coming in)
• Report sonar / altimeter as zero if landed (if altimeter data is coming in)
• Once non-landed time-out after 2-3 seconds if either flow or sonar / altimeter do not become valid
• Ensure position controller goes into failsafe if timing out
• Takeoff should takeoff action, not user-driven altitude (so the system clears the minimum altitude quickly)

[LorenzMeier]

@ChristophTobler @priseborough @Tumbili

[ChristophTobler]

About flow failsafe:

• In the simulation it does switch to WARN [commander] failsafe mode on after not getting valid measurements (fly outside of asphalt) after a few seconds. But then it just switches to altitude mode (obviously it can not return when it does not have a position information). I guess it should handle that case by just landing?
• Do what is mentioned above. Then there seems to be a bug when returning to the asphalt manually. It does not always switch back to WARN [commander] failsafe mode off (this is ekf2 specific)

[LorenzMeier]

Yep, we should add auto-land as an option instead of falling back to manual and we should find and fix the issue which is not making it return.

[mhkabir]

@jgoppert You will be interested in this. We should also consolidate the hack in LPE which we currently use to do the out-of-focus takeoffs.

[jgoppert]

Any reason not to implement a fake landing measurement like LPE does. This isn't really a hack since it is fusing the extra information that we know when landed. This gives agl = 0, vx=0, vy=0 when landed. During takeoff the measurement goes away and the vehicle is in the air. Then the covariance matrix for x, y is growing. If it gets too large it publishes that position is invalid. So you get a timeout that is adaptive based on the measurements that you have. I think this is the best option. We do need to handle the data better and make sure the vehicle lands.

[jgoppert]

@priseborough That change looks reasonable to me. The only other thing you might want to do is add a measurement of agl = 0 to keep the terrain state and altitude state the same.

[priseborough]

What I did with the APM implementation was add a parameter which was the expected on-ground range measurement. In the absence of range data it would use the expected on-ground value if not in-air. This allowed for the variation in landing gear height between copters.

[jgoppert]

I think both estimators have working solutions for this now.

[mhkabir]

I'm not sure if this is solved. @ChristophTobler?

[ChristophTobler]

Yes, this is solved. Thanks