With what firmware version have you observed this?
We have seen some occurence of the EKF diverging mostly when starting the Crazyflie well outside the TDOA system. We have tried to isolate the points where a NaN could appear and solve it over time but maybe there is still some left.
Can you see a pattern of for when this kind of problem happen?
Hi arnaud, thanks for your reply,
So the matrix multiplication from CMSIS library is known to have some specific values that do not work?
I'm using a custom firmware to understand and maybe test sightly changes of the TDoA2 localization algorithm, can you explain (or point me where to look at) how you isolated the points and solve it?