The algorithm in performUTforUKF

[svendvn]

First off, great document! - it has been very helpful to me. I have just started learning about Kalman filters and I was trying to implement the UKF myself, and I met some inconsistencies trying to simultaneously follow the code in this document and "Unscented Kalman Filter Tutorial" (Terenaju, 2011) and "On Unscented Kalman Filtering for State Estimation of Continuous-Time Nonlinear Systems" (Saarka, 2007);

In contrast to Terenaju and Saarka, the function performUTforUKF calculates the covariance without adding the covariance matrix of the dynamic equation, say, Qk. That is, performUTforUKF computes
covdyn = errdyn.dot((self.wc * errdyn).T)
instead of
covdyn = errdyn.dot((self.wc * errdyn).T)+Qk

Relatedly, In contrast to Saarka (and perhaps also Terenaju), the function performUTforUKF propagates the already transformed sigma points in the update step. That is, performUTforUKF computes
SPobs = obsfun(SPdyn)
instead of
SPobs = self.constructSigmaPoints(meandyn, covdyn)

Is it two different algorithms for the UKF? Or are some of them specified wrongly? Any insight to this would be useful.