Should KalmanFilter implement KalmanTypeFilter?

[truher]

hey, is there a reason that KalmanFilter doesn't implement KalmanTypeFilter?

I was going to fix up the state space arm example for angle wrapping, but it would be a lot cleaner if the LinearSystemLoop took a KalmanTypeFilter instead of KalmanFilter itself.

?

[calcmogul]

KalmanTypeFilter is currently only used for KalmanFilterLatencyCompensator. KalmanFilterLatencyCompensator requires setP() and getP() functions, which wouldn't have sane behavior for KalmanFilter. Linear Kalman filters with consistent periodic predictions and updates have steady-state error covariances and Kalman gains which are dependent on the A, C, Q, and R constructor arguments.

[truher]

Ah, ok. Is there better way to make ExtendedKalmanFilter work smoothly with LinearSystemLoop? Or for my angle-wrapping goal, would it be better to add residual functions to the normal KalmanFilter?

[calcmogul]

ExtendedKalmanFilter is for nonlinear systems, so you won't be able to use it with LinearSystemLoop.

[calcmogul]

We've considered adding residual overloads for KalmanFilter, but it makes the implementation annoyingly more complex.

[truher]

ah, so the recommendation is to make something like NonlinearSystemLoop that does all the right things? i'll try that.