Sysid and control design using ulog

[mzahana]

Hi @jgoppert

Can ulog files be used in the sysid and control design instead of px4log? If yes, what is the procedure to do it?

Thanks.

[jannsta1]

Hi @mzahana,

You could try using this repo instead: https://github.com/dronecrew/ulog_tools.

I would also be interested in seeing a best practices procedure for this - e.g. how to set the parameters and if logfiles with particular flight manoeuvres produce better results.

[mzahana]

@jannsta1 Thanks for pointing me out to that repo. I will try it and provide my feedback.

[mzahana]

I am getting the following error.

AttributeError: module 'scipy.linalg' has no attribute 'solve_lyapunov'

I am using

• anaconda 3
• Python 3.6.3

Any idea how to fix this?

EDIT:

doing conda list | grep scipy gives

scipy                     1.0.0            py36h1de22e9_0    anaconda

[mzahana]

Ok. I got it working by installing an earlier version of scipy=0.19.1

[jannsta1]

Glad you got it working @mzahana. Would be interested to know if you do manage to get some good results from the package? By naively running I don't seem to get a particularly good tune, especially in yaw and pitch fails entirely! Planning to play around with the matlab toolbox a bit so I understand the parameters better and then coming back to it.

[mzahana]

Same here. I am not getting good identification of the control gains and I fall back to old set of gains that work much better. I will study more how this can be improved. It also would be good to know what you exactly plan to use (algortihms) to solve this issue.

[DanAbara]

@mzahana and @jannsta1 I have also used the repo suggested on this thread, however I get the following error when I run the module to obtain the sample frequency:
KeyError Traceback (most recent call last)
in
6 comb = ut.sysid.prepare_data('log_48.ulg')
7
----> 8 sf = ut.ulog.sample_frequency(log)

~.conda\envs\ut_env\lib\site-packages\ulog_tools_ulog.py in sample_frequency(data)
28 def sample_frequency(data):
29 """Compute sampling frequency"""
---> 30 return 1.0e6 / np.diff(data['f_timestamp']).mean()

KeyError: 'f_timestamp'

I have scipy=0.19 so that is probably not the problem. Any suggestions would be greatly appreciated. I struggled with the px4tools and got stuck so decided to try the ulog_tools repo. Thanks in advance.

[DanAbara]

Update:
@mzahana and @jannsta1
So I sorted it out and its all working now. I have another question though, how can I export data to csv. For instance, if I use the prepare_data module in the '_sysid.py' file, I obtain a data structure which holds the re-sampled input/output data in a tabular form. I plan to export this to a csv and perform identification in matlab, to see how that goes, but I have little or no idea how to do that. I'm quite new to python, just learning the ropes. Any advice to ease my struggle will be helpful.

Thank you.

[jannsta1]

From what I remember it's a pandas data frame which has a built in to_csv() method. I think there are also libraries that convert .ulg files directly into MATLAB data structures.

[DanAbara]

@jannsta1 if i understand correctly, you're saying the output data structure of the resampled data is a pandas data frame with a built in to_csv() method (don't know what that means though I'll look it up), so that means there should be some way to open it as a csv or convert it to csv etc right? Thanks by the way for responding.

Update:
So I managed to get it sorted as well. Thank you very much @jannsta1. Best wishes.

[DanAbara]

@jgoppert , @jannsta1 and @mzahana have to trouble you guys again. I have a question about the sysid process in the ulog_tools package.

Why is the output signal (angular velocity) differentiated to get accelerations before the sys id?

Why is the signal not used directly?

Thank you