When to Choose the C-D EKF

1. The system is Non-Linear.
2. We know that the pdf of $f(x(t)|Y^t)$ is approximately a Gaussian distribution.
3. The EKF (Extended Kalman Filter) gave bad results, or failed, and we can suppose that the bad result were given to a not-so-good linearization of the function $f$.

NOTE: $f(x(t)|Y^t)$ means: pdf (probability distribution function) of $x(t)$ knowing the measurements ($Y^t$) up to time $t$.