For an X-Quad copter configuration, will tilting motors in the direction of motor torque (only circumferential tilt, no radial tilt), work out-of-the-box with pixhawk + ardupilot? If not, what tuning or modification is needed?
From a torque / yaw perspective it should not be a control problem, but I am not sure if circumferential tilt is a problem for lateral control. Also the tilted motors introduce a relatively tiny amount of pitch / roll moment from tilted torque vector.
Seems like it can work out of the box. All of the examples I see, there is no mention of tuning or configuration changes.
https://ardupilot.org/plane/docs/quadplane-building.html
Here is what ArduCopter’s lead control loops designer @Leonardthall thinks about a similar motor twist:
and
that is totally different. looks like a radial tilt inwards they are calling ‘tilt up’, not a circumferential tilt…
i think circumferential tilt is legitimate and not a hack. just a question of whether additional tuning or configuration is needed to accommodate it.
You are still loosing efficiency in order to gain some yaw control authority.
But are you yawing so often that it makes sense to tradeoff efficiency for it?
Anyway, back to your question: to do it 100% correctly you will need to change the source code loops.
But yes, it will work just fine without any changes, if the twist is kept small. Just don`t ask how small is small 
the efficiency loss can be negligible for a huge gain in yaw authority control. its very easy for it to be worth it, especially if you consider yaw control without tilt consumes lots of power. examples:
1 deg: 99.984769515%
2 deg: 99.939082701%
3 deg: 99.862953475%
5 deg: 99.619469809%
10 deg: 98.480775301%
2 deg is recommended in the link above for huge gain, and that means 0.06% loss. Is that even worth complaining about? Like I said, given how inefficient pure-torque-based yaw control is, you might even save power over the course of a mission.
So you say we need to change the source code loops; got any examples?
if your target is 2 deg I do not think it is worth the trouble of changing the source code.
Just follow all the tuning procedures as normal. And make sure there is no yaw imbalance (when hovering all motors must get the same PWM value).
If there is imbalance, adjust arm twist to compensate.
And please post the results, it sounds interesting.
The only change in approach is you start to benefit from doing an additional tuning step after autotune and adding some D term.
great, that is what i hoped to hear. whats the typical amount of tilt?
There isn’t one but I would recommend between 3 and 5 degrees.
I am not making any recommendations on that. They are all engineering decisions to be made by the designer, in context.
From your previous answer it looks there are no adverse effects so i guess it would be nice to have a better yaw at the expenses of a negligible efficiency drop.
what if the tilt of each motor are slightly different (due to adjustment imprecision)?
yeah if you have an otherwise normal build motor tilt seems like a no-brainer.
Can someone help me understand the difference between a “radial tilt” and a “circumferential tilt”? Maybe a little diagram or link?
From reading this post it seems a radial tilt is discouraged and a hack for crappy flight controllers. But circumferential is completely valid for increasing yaw authority. Am I understanding correctly?
yes you understand correctly
Thanks. I can’t see clearly in the videos above, I’m still a bit confused what the physical difference between a radial and circumferential tilt is. Do you have a picture you could post that helps explain this?
oh sorry. @dkemxr’s paper shows circumferential tilt. radial tilt would be towards the center of the drone.
