I'm trying to use Webots software to simulate a quadcopter. For now, I just want my drone to be able to stay in a fixed position after take off. I have created 6 PD controllers for (x,y) position, roll, pitch, yaw and altitude. So anyone can share with me how to choose the suitable values for the P, D constants for all 6 controllers, please? Thank you!
The base way to tune your PD controller, first simulate your PD controller using software like Matlab or ..... and fine-tune it. Then try it on hardware.
You can use evolutionary algorithms to tune the parameters of the PID controller, also tuning can be done using the built-in PID controller tuner app in MATLAB.
Taimoor Ahmad Khan Hapuarachchige Don Nelaka Shayamal Priyankara If I can't determine the transfer function of the system, can I still use Matlab to tune all my 6 PD controllers which work together?
Mr. Nguyen, I think this project may be a good starting point for learning to choose the best parameters for the 6 PD controllers.
https://cyberbotics.com/doc/guide/mavic-2-pro
Ideally, you would want to tune the parameters one axes at a time.
This is a technique I used many years ago to learn how to do PID tuning of a quadcopter when simulation software were not available. The idea is to consider only one axes of the quadcopter can change while keeping the other two fixed.
I think you would need to define a speed to rotation equation which would take some RPMs as inputs and generate, say rotation angles at a given timestamp. You would then iteratively change the values of P first and then D to find the best combination of PD. Repeat the process for the other two axes and this should in 'theory' give you the best parameters. However, a quadcopters motion in real life is a 'on-the-fly' optimization due to the fact that, a set of static PD values cannot account for different external factors (wind condition, variance in BLDCs RPM at a given throttle, etc.)
I hope this helps.
Good luck
Azmyin Md Kamal as I said I'm simulating this done via software so I don't know how I can fix one axis. Besides, can you explain to me more clearly what you mean in " I think you would need to define a speed to rotation equation which would take some RPMs as inputs and generate, say rotation angles at a given timestamp ", please? Thank you so much!
I believe it should be possible. My suggestion to find out how you may be able to artificially lock the drone at a fixed height and one of the roll/pitch axis would be to post a question on StackExchange tagged with the Webots tag https://stackoverflow.com/questions/tagged/webots
Regarding the linear equation consider the plus shaped quadcopter in this picture https://qph.fs.quoracdn.net/main-qimg-006f50df374059a3fd371cb0ae152f6e.webp
The scenario is you are tuning the PD values for the roll axis i.e. Motor 1 and Motor 2 RPMs can vary. The Pitch Axis and Height of the quadrotor are locked
You have N timestamps (think about the 1,2,3,4,.... N on a X axis). Thus, to see what would be correction angle proposed by your PD, you would first need to generate some 'disturbance'
angle_pitch_disturbance = some_factor * RPM_motor_1(t) + some_factor*RPM_motor_2(t) ......................... (1)
If the set angle is set to 0, your PD equations would then subtract this disturbance from 0 to find the error the value against which your PD would generate a throttle signal to bring the quadcopter back to '0' angle.
This is what I meant when I wrote "I think you would need to define a speed to rotation equation which would take some RPMs as inputs and generate, say rotation angles at a given timestamp "
For the scenario described above, values for RPM_motor_1 and RPM_motor_2 would vary with time. Unfortunately I have experience only with a real-life drone so I never needed such equations. However, I am quite confident that a disturbance equation of this form is needed for the simulation. I would encourage you to look into papers first and focus on how researcher generated the dynamic scenario against which you would then tune your quadcopter's PD values. My second suggestion would be not to go with Webrot first but to use MATLAB codes and tutorials found online to get some basic understanding on solving this problem first. Jumping directly into the a complex simulator such as Webot may have been adding more complexity to your problem then benefits at this stage of the algorithm development.
With best regards,
Azmyin