Tuning Challenges

🧪 Flight Testing & Log Analysis
1

Hi All,

Has anyone used this https://github.com/dronecrew/ulog_tools for computing multicopter gains? I tried using it for a small 10" quad rotor with what seemed to be good results. For a much larger frame (~35" span) I get results that are too twitchy.
Anyone have some advice? Are there some parameters I can adjust in the LQR synthesis?

This was the first flight using roughly default gains:
https://logs.px4.io/plot_app?log=5829e4ff-ccf3-4704-865d-234826ebd8b1
And the flight with LQR gains:
https://logs.px4.io/plot_app?log=f83a8ebe-fce4-425b-8093-482c2df61371
When using nearly default gains, the quad would get growing oscillations hovering almost still. It seems as though the roll/pitch rate P terms were too high. Any thoughts on how to improve the stability? I don’t really care about how responsive the system is, I only care about stability.

Thanks!

1 Like
2

@jgoppert Look nice!
but does it really does the work?
Is that the same algorithm like on the ‘auto tune’ of ardupilot?
I couldn’t understood if i need to do spacial maneuvers, and how/if need to select praticular section of flight

3

I believe it worked for my small 10" frame. For my large 35" frame, the LQR consistently resulted in twitchy gains…I’m not sure if there is a problem with the system ID or the LQR. In the case of the LQR, I don’t know how the balance between tracking performance and actuator energy is established.

4

Are there parameter of the algorithm that need to be changed according to the platform?

BTW, I tried it on the Iris simulator, some parameter seems to be changed a lot, here are my result after the first iteration: