narykov
commented
2 months ago OK, found a semi-decent conference paper with Bar-Shalom among the authors, which exposes the $F$ and $Q$ matrices. Indeed, the element in $Q$ that describes the turn rate noise is time-dependent, which means that the following is brought into question, i.e., q**2 may need to be turned into dt * q**2:
P.S. It would be good to find reference for q**2, since in the publications I've come across either $q$ or $\sigma^2$, but never $q^2$.
The description of the nearly-constant turn rate model in here doesn't look right.
The expression of $F(x)$ here is a vector, not a matrix as implied by $F$. An appropriate $F$ matrix can be seen, e.g., on p. 16 in here, but a bit struggling with finding a good reference. Regardless, in the entries to this vector above, it should be $x{pos}$, not just $x$ (and the same, respectively, for $y$). Finally, a bit surprised to see $q{\omega}^2$ not scaled by the time interval, which seems contradictory to the SDE in the model's description that states $d\omega = q_\omega dt$.