Question about the Kalman Filter

[sandyUni]

I try to run dimple\demo\16_KalmanFilter\run.m, then from the results, I conclude that there is some slight difference between the Kalman guess and Factor graph guess. However, theoretically, These two results are supposed to be same for both are using same parameters describing Gaussian distribution.

So could you give me some advice or explanation?

[sandyUni]

Besides, I don't think it is caused by quantization error. The difference is significant regard to that level.

[analog-cbarber]

I think there may be numerical accuracy issues with Dimple's gaussian solver when used with joint (multivariate Guassian) variables.

[sandyUni]

So, will you fix it in the future?