Client freeze

Post here to get support
Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Client freeze

Post by Mitteau » Mon Dec 18, 2017 11:00 am

Hi,
I mentionned somewhere, and I read other posts considering the freezing of the crazyflie python client after a crash. I discover a reason of this freezing: the controller output shows hard oscillations of roll rate:
P_20171216_164934_r.jpg
Thus the commands of the joypad are absolutely inactive.
I manually changed the position and the controller went back to a normal behaviour:
P_20171216_165140_r.jpg
What part of the firmware do I have to investigate?
Regards
Jean-Claude

arnaud
Bitcraze
Posts: 1738
Joined: Tue Feb 06, 2007 12:36 pm

Re: Client freeze

Post by arnaud » Mon Dec 18, 2017 1:13 pm

How do you get this behavior? Is there a way to reproduce it?

All that has to do with the stabilization happens in the stabilization loop in stabilizer.c. So I would start to look there and then dive in the modules that outputs questionable values.

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Tue Dec 19, 2017 5:09 pm

I don't like to reproduce this behaviour because it appears after hard crashes... I will investigate stabilizer.c

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Wed Dec 20, 2017 8:56 am

New: something like the discontinuities of the roll rate appears when cf2 is on the back after a crash. But it differs from my graph because it stops as soon as cf2 is put back on its feet...

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Fri Dec 29, 2017 5:18 pm

Hi,
I confirm the test made on roll rate. This graph shows the beginning of the freeze.
roll_r.jpg
I also confirm that a move by hand of the cf stops the phenomenon.
This instability does not appear on pitch control, until now.
Where does this behaviour comes from?
Regards
Jean-Claude

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Tue Jan 02, 2018 9:17 am

Hi
I confirm the phenomenon on my cf. For some reason, mainly a crash, but not only, the controller output instability on roll rate, as in preceding figures. I did not succeed trying to enter the instability by rapid turns around roll axis. Even rough and 180°.
Once the desequilibrium is attained, it do not stops, the only way to get stabilized is being give a large move to the cf.
During desequilibrium time, there is no active command from the client.
Gyro and acc measures are OK during this period. The commands from client are obtained by firmware (they are visible on the plotter tab).
Regards
Jean-Claude

tobias
Bitcraze
Posts: 1718
Joined: Mon Jan 28, 2013 7:17 pm
Location: Sweden

Re: Client freeze

Post by tobias » Tue Jan 02, 2018 12:29 pm

A quick question. Is the kalman filter or the complementary filter running?

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Thu Jan 04, 2018 9:38 am

Hi
In the starting log I can see "estimator 2". I suppose this means Kalmann filter. The cf is eqipped with the flowdeck for a perfect stability, except fast horizontal run, or the unstabilities we are talking about here.
Regards
Jean-Claude

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Thu Feb 22, 2018 9:43 am

Hi!
For two weeks, I fly crazyflie without vertical assistance: flowdeck or z-ranger. I neither had the freeze problem nor roll rate instability. I verified that estimator is the estimator 1 on connection console message. So the problem would be linked to estimator 2?
I will make new test adding the vertical assistance deck.
BR
Jean-Claude

Mitteau
Expert
Posts: 248
Joined: Fri Jan 29, 2016 6:00 am

Re: Client freeze

Post by Mitteau » Fri Feb 23, 2018 8:52 am

And, this morning, I installed the flow deck... At first run, the freeze of the crazyflie occured... Estimator 2 is used. Sure it is responsible for this behaviour. Do I have to write an issue for this which looks like a bug?
BR
Jean-Claude

Post Reply