About porting in a controller from Crazyflie_nmpc

[Yiqing-Gu]

Hi,

I am currently trying to port in a controller from crazyflie_nmpc to CrazyS.

Now, I am looking at controller.cpp in nmpc (https://github.com/bcbarbara/crazyflie_nmpc/blob/master/crazyflie_controller/src/controller.cpp).

However, it is my first time to port in a controller. I wonder whether I am on the right track or anything suggestions for me?

Thanks.

[gsilano]

Well, if I were you, the first step would be to understand the control chain and the control signals involved. You could start reading the two conference proceedings attached to the repository. You can also use the repository's GitHub issue tracker to learn more. Simply, open a thread if you need any kind of assistance (only serious questions and issues).

Not so sure, but if I'm not mistaken, the controller was made to control a real Crazyflie quad-rotor. In that case, the control signals should be limited in the range [-1,1] for the roll, pitch and yaw rate and between [0,1] for the trust. Therefore, these values ​​should be mapped to work in CrazyS where we assume we are able to control the propeller angular velocities ranging between [0, 2300] rad/s (check, I don't remember the correct values). However, the control signals are normalized quantities that are easy to remap. Also, you should understand the frequency each block works. This could affect the stability of the system.

In summary, here's how I would approach the problem:

1. I would read the papers related to the repository. From here, I would try to sketch up the control chain, detailing the signals and frequencies involved. To each block I would associate the corresponding cpp file, so as to create a map between paper and code.
2. I would understand what are the functions in CrazyS, how information is exchanged, from waypoint generation to vehicle control. Here too, a chain of control, signals and respective frequencies would help a lot.
3. I would try to understand if the controller can be taken as it is, by simply changing the parameters, or if it needs further modifications or an upstream or downstream interface block.
4. etc

I hope this helps. Please, keep me posted on the updated. If I can, I will try to help.

[gsilano]

What package are you referring to for Lee's controller?
As for the build issues, it's weird to me as GitHub Action has no issues. See the recent compilation report: https://github.com/gsilano/CrazyS/actions/runs/2605230058. Here are the ones for Melodic: https://github.com/gsilano/CrazyS/actions/runs/2605238411

I think the launch file is not well sourced in the bashrc. Could you send the contents of this file here?

[Yiqing-Gu]

What package are you referring to for Lee's controller? As for the build issues, it's weird to me as GitHub Action has no issues. See the recent compilation report: https://github.com/gsilano/CrazyS/actions/runs/2605230058. Here are the ones for Melodic: https://github.com/gsilano/CrazyS/actions/runs/2605238411

I think the launch file is not well sourced in the bashrc. Could you send the contents of this file here?

Apologize for misunderstanding.
Lee's controller I meant was branch "pull/lee_controller" in CrazyS. I cloned "pull/lee_controller" and tried to build it but I got some errors under ROS noetic. I guess "pull/lee_controller" does not fully support ROS noetic since it was released much earlier than the noetic branch?

The launch file I used was "crazyflie_hovering_lee_controller_alpha.launch". Since I moved Lee's controller components to noetic package directly, it can be something wrong with my operation.

[gsilano]

Now I see your point. Well, Lee's controller-related C++ files should be compatible with every version of ROS. However, the pull request cannot be copied and pasted from one release of CrazyS to another. Requires minimal customization. Then, you could clone the appropriate CrazyS release for your ROS version, and integrate the changes from there. Please note that this is a draft, so it may not compile or work correctly. It is definitely a great starting point, and I would be happy if you could finish my job.

[Yiqing-Gu]

Hi, Lee's controller package can be successfully compiled under ROS noetic right now. I tweaked two files in plugins and revised default python configurations.
Now after executing "crazyflie_hovering_lee_controller_alpha.launch", I found every component in simulation could be loaded. However, crazyflie always stayed on the ground but propellers were spining. I would like to make sure whether it was the bug you had before? Or it was just related to my wrong configuration?
Thanks.

[gsilano]

Yes, it's correct. This is why I haven't merged the pull request yet. You need to complete the code and do some debugging.