1D EKF tests

[arnauochoa]

The following figures have been obtained with the code from branch feature/backTo1D (commit ID: 0dd672eb66284e0f7d1c910f7878751094b73251).

We have implemented the 1D EKF, position error, velocity error and bias converge to 0 when we generate the accelerometer measurements with 0 bias (only Gaussian noise). We think that these results are the ones that we should expect for this configuration. Then, we tested the algorithm considering a bias of 0.01 m/s^2 in the accelerometer measurements (every iteration). In this case, we observe in the position and velocity estimates that the bias is compensated but the errors of these estimations do not converge. Furthermore, the estimation of the bias does not converge to the true value. We are not sure if the problem is the definition of the F regarding the evolution in time of the bias (we haven't changed F between tests). Or maybe we are missing something else.

Test 1: No accelerometer bias Config File: ConfigNoBias.txt

Test 1: With accelerometer bias Config File: ConfigWithBias.txt

[jtec]

Hi guys,

Thank for the update!

I tried running the filters with a large accelerometer bias (1 m/s^2) and could see the same behavior: the estimated bias goes to zero; when looking at the estimated position error, one can see that the error follows a zig-zag pattern - so I think the EKF does not actually compensate the bias (how would it, if it's not correctly estimated), but corrects its effect at each GNSS update;