Resetting Kalman filter with invalid quaternion results in crash

[bgromov]

Setting initial quaternion (kalman/initialQ[x,y,z,w]) to zero and resetting the filter (kalman/resetEstimation = 1), results in immediate firmware crash. CF spits out a part of the regular boot log, but then freezes and does not respond to commands anymore.

[knmcguire]

Hi! Interesting that this happens. Could you tell a bit more how you set this? Through the parameters or hard coded in the firmware?

[bgromov]

Through the parameters. Initially I thought it has something to do with the convergence of the filter (see #402), but then I noticed I am actually setting a malformed quaternion (all terms set to 0). Everything is fine when the quaternion is set to [0, 0, 0, 1].

[knmcguire]

Hmm okay. Did you make your own parameters for the initial quaternions? It doesn't exist in the current master, and it is set starting from this line:

crazyflie-firmware/src/modules/src/kalman_core.c

Line 189 in c376b10

initialQuaternion[0] = arm_cos_f32(initialYaw / 2);

Just making sure what is changed in your version before we try to recreate your problem.

[bgromov]

Seems like we are on a pretty old version indeed. Sorry for this. We will try to merge from the upstream and see if we have any related issues. Thanks