[lesto]

@Kevin: I am not a good pilot at all; actually i can barely take off, that is why I write code to make it for myself 🙂 After all who build race car, plane and stuff rarely fly them.

@david: I know it can be difficult, and if you can post some slowmo of the reaction and the blackbox we can analyze data.

Some things i want to check:
1. estimation jerkness. When I analyze the data for the Tilt Drone, one big issue was that the great acceleration was actually giving a bad time to the angle estimation, and the rotation was exaggerated until the error was corrected. At the time i pushed for the DCM (see https://github.com/cleanflight/cleanflight/issues/1253) but at the time i left the company and never had the chances to test the new cleanflight/betaflight algorithm in real scenario; and nobody did, AFAIK.

2. Roll pitch yaw interference. There of course by tilting the motor you will have some roll/pitch/yaw mixing together. This mixing can be programmatically estimated and cleaned out.
If you mount a camera and you mix only pitch and yaw, or pitch and roll or other combination, using some dedication and software used for VFX we can estimate motion camera (well, i dont know how to do that) and check if the estimation for mixing is matching the real case scenario.

The estimation should be pretty easy for 2 axes, as component in pitch influence yaw based on the sine/cosine of the pitch angle.

Of course there may be other effect, and would be nice to see what you think is the most problematic; some stuff, like compensating the throttle when pitching forward, is so automatic in pilot’s finger they actually disliked it at the time i tested.

[Author]

Posts