When to Choose the EKF

1. The system is Non-Linear but Locally Linear, the function of state-update $f$ and measurement $h$ will be locally linearized.
2. We know that the pdf of $f(x(t)|Y^t)$ is approximately a Gaussian distribution.

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

---

• In case the approximation is not satisfactory, for example: real distribution (in black), resulting distribution (in red). We can use more accurate alternatives: The UKF (Unscented Kalman Filter). Or the PF (Particle Filter).

---