Fix the correction yaw gives in 2pass (Motor Output Mixer) (Thrust Linearization)

[Quick-Flash]

Previously yaw correction would change based on how much roll and pitch was mixed. When mixing an even amount of roll and yaw you would end up with the motors producing more roll than yaw. This problem only even shows up during yaw movements that involve pitch or roll.

example from simulated roll and yaw pidsum with current and new code.

pidsum inputs = 0.2roll, 0.0pitch, 0.2yaw

current code
m1 = -roll+pitch-yaw = 0.09 = -0.03roll+0.03pitch-0.3yaw
m2 = -roll-pitch+yaw = 0.33 = -0.11roll-0.11pitch+0.11yaw
m3 = +roll+pitch+yaw = 0.73 = +0.243roll+0.243pitch+0.243yaw
m4 = +roll-pitch-yaw = 0.49 = +0.163roll-0.163pitch-0.163yaw

roll = -0.03-0.11+0.243+0.163 = 0.266
pitch = 0.03-0.11+0.243-0.163 = 0.0
yaw = -0.03+0.11+0.243-0.163 = 0.16

new code
m1 = -roll+pitch-yaw = 0.01 = -0.003roll+0.003pitch-0.003yaw
m2 = -roll-pitch+yaw = 0.41 = -0.136roll-0.136pitch+0.136yaw
m3 = +roll+pitch+yaw = 0.81 = +0.27roll+0.27pitch+0.27yaw
m4 = +roll-pitch-yaw = 0.41 = +0.136roll-0.136pitch-0.136yaw

roll = -0.003-0.136+0.27+0.136 = 0.267
pitch = +0.003-0.136+0.27-0.136 = 0.0
yaw = -0.003+0.136+0.27-0.136 = 0.267

This code makes sure that yaw will not be shrunk during roll or pitch movements and should improve handling during quick yaw movements.

[Quick-Flash]

@tylercorleone perhaps you can find a way to code this more efficiently, but long story short, I found that in 2pass the yaw axis would have less authority during pitch/roll moves. This change gives it the same authority as pitch and roll, but requires a fair bit more logic to make work :(

[nerdCopter]

I flew this and it flew good. submitted logs for review in discord.

[nerdCopter]

• approved based on conversation
• will defer to @Quick-Flash to merge or not

[nerdCopter]

@tylercorleone , are you interested in reviewing this PR? I've flown about 9 packs on it.

[tylercorleone]

@tylercorleone , are you interested in reviewing this PR? I've flown about 9 packs on it.

Hi guys, honestly lately I lost some interest in quads but above all I've been for so long away that it's hard for me to understand the change.

I would ask only two questions:

• the current version already has a change compared to what was my initial release. Wasn't that a fix? This one tries to fixes another problem? I already have some difficulties understanding in dept the code after the first fix so it's even more difficult to understand this one
• the problem here described has been reported by many users? I fear that a fix could mess up the initial idea, and again there's some correlation with the first fix? I don't remember any problem (considering my flight style) with my initial version so it's difficult to me to follow these fixes.

I'll try to look again at this PR anyway.
I'm sorry to not be that helpful 🙁

[Quick-Flash]

@tylercorleone , are you interested in reviewing this PR? I've flown about 9 packs on it.

Hi guys, honestly lately I lost some interest in quads but above all I've been for so long away that it's hard for me to understand the change.

I would ask only two questions:

* the current version already has a change compared to what was my initial release. Wasn't that a fix? This one tries to fixes another problem? I already have some difficulties understanding in dept the code after the first fix so it's even more difficult to understand this one

* the problem here described has been reported by many users? I fear that a fix could mess up the initial idea, and again there's some correlation with the first fix? I don't remember any problem (considering my flight style) with my initial version so it's difficult to me to follow these fixes.

I'll try to look again at this PR anyway. I'm sorry to not be that helpful 🙁

So the change to your PR that was made changed the mixer slightly so that yaw axis moves would not cause more throttle movement unless it had to. Basically me and a few others complained that it seemed to sort of "float" during sharp turns due to yaw adding thrust. Without the yaw compensation if you just hovered and gave a little yaw you would see the drone move up due to this. Added some code to correct and compensate for that.

Later I found out that this compensation would create a yaw value to the motors that would shrink as more roll/pitch was applied to the mixer. So you would end up with less yaw than your controller was commanding during roll/pitch moves. This fixes that so that yaw control stays the same regardless of pitch/roll (unless clipping starts to be involved in which case roll/pitch/yaw are all clipped). I should probably update this code so that these code changes are only applied if you have the mixer_yaw_throttle_comp turned on.

Looking over the code again I would hope that there is a more cpu effective way to achieve this effect. Thanks for taking a look @tylercorleone and thanks for the original implementation! I've been flying this change for a while and don't notice a huge flight performance difference, but it does seem to be a little mathematically better way to mix the pidsums while the mixer_yaw_throttle_comp is enabled.

[nerdCopter]

would constrainf help here in any way? or constrain_boundaries ?

[nerdCopter]

• @Quick-Flash , please check last comment.
• i've been flying this exclusively on 5", 6", 7" without any errors -- 8K and 10.67 K loops.

[Quick-Flash]

• @Quick-Flash , please check last comment.
• i've been flying this exclusively on 5", 6", 7" without any errors -- 8K and 10.67 K loops.

Looks good!

[skylupop]

Hi @Quick-Flash, @nerdCopter, reading the descriptio:
m1 = -roll+pitch-yaw = 0.09 = -0.03roll+0.03pitch-0.3yaw
the contribute is not "distributed" in that way, on every single motor but the total motors' contribute is what makes the quadcopter roll, pith and yaw.

For example, if I'm not wrong, given:

m1 = -roll+pitch-yaw = 0.09
m2 = -roll-pitch+yaw = 0.33
m3 = +roll+pitch+yaw = 0.73
m4 = +roll-pitch-yaw = 0.49

by looking at signs I can deduce that motors 1 and 2 will contribute "negatively" to roll movements (it would be better if they stay still or reverse), so the total roll contribute will be:
m3 + m4 - m1 - m2 = 0.73 + 0.49 - 0.09 - 0.33 = 0.8
while for yaw:
m2 + m3 - m1 - m4 = 0.33 + 0.73 - 0.09 - 0.49 = 0.48

total roll contribute is higher because the "0.2" input to the controller was thrust difference and not motor output, while yaw-related input is already proportional to motors' RPM.

P.S. I'm tylercorleone using my work account, sorry :)

[Quick-Flash]

Hi @Quick-Flash, @nerdCopter, reading the descriptio:
m1 = -roll+pitch-yaw = 0.09 = -0.03roll+0.03pitch-0.3yaw
the contribute is not "distributed" in that way, on every single motor but the total motors' contribute is what makes the quadcopter roll, pith and yaw.

For example, if I'm not wrong, given:

m1 = -roll+pitch-yaw = 0.09
m2 = -roll-pitch+yaw = 0.33
m3 = +roll+pitch+yaw = 0.73
m4 = +roll-pitch-yaw = 0.49

by looking at signs I can deduce that motors 1 and 2 will contribute "negatively" to roll movements (it would be better if they stay still or reverse), so the total roll contribute will be:
m3 + m4 - m1 - m2 = 0.73 + 0.49 - 0.09 - 0.33 = 0.8
while for yaw:
m2 + m3 - m1 - m4 = 0.33 + 0.73 - 0.09 - 0.49 = 0.48

total roll contribute is higher because the "0.2" input to the controller was thrust difference and not motor output, while yaw-related input is already proportional to motors' RPM.

P.S. I'm tylercorleone using my work account, sorry :)

I got the data without using any thrust linear, where roll/pitch/yaw should keep their ratios regardless of extra roll or pitch input. It's not to tricky to pull out the mixer code and run simulations based on the desired r/p/y and then back calculate things based on the motors to see the actual outputs it gives.

[nerdCopter]

i can produce BBL files upon request. i'm unsure how to analyze them as proof or disproof however.

[nerdCopter]

all these are from this branch
https://drive.google.com/drive/folders/1fpohLiyvhKU7pLrtJgJOxbeqMP0c7PNo?usp=share_link

[skylupop]

Hi @Quick-Flash!

It's not the thrust linearization thing the main "problem" (although it should be kept in consideration it is the only reason because the 2PASS mixer exists: handling roll and pitch separately from yaw because of their different nature :) ) but the steps like this: m1 = -roll+pitch-yaw = 0.09 = -0.03roll+0.03pitch-0.3yaw
Looking at a single motor output cannot give you any hint of what is happening if you don't look at the opposite motors.
E.g. a motor could be at 100% but not producing any roll, that is m1 = 100 doesn't mean 33.3% roll, pitch and yaw!

As I wrote you in a message we could play with the very first code proposed, e.g. here https://github.com/emuflight/EmuFlight/blob/master/src/main/flight/mixer.c#L1015. We could stard from there and tweak the very basic way we mix values avoiding additional "fixes".

"When mixing an even amount of roll and yaw you would end up with the motors producing more roll than yaw" this is due to thrust normalization. Also it's difficult to compare roll and yaw :) you could compare the percentage of what the user is requesting and how many of that request is being applyed instantaneously.

@nerdCopter it would be nice to see a log describing the issue of the initial code and then we could try to "fix" it!

[nerdCopter]

merging. i rather fly it than not fly it.
i do prefer proper analysis as @tylercorleone suggested, but no effort is being made.