Investigation of the Rao-Blackwellized particle filter for two jump-Markov inertia measurement models

In the importance sampling step, the Rao-Blackwellized particle filter makes use of the probability density functions (pdf) of the Kalman filter for the computation of likelihood functions and sample weights. Whether these pdf are separated or overlap each other has a significant effect on the estimation of the posterior distribution of the non-marginalized states. This papers presents simulation results demonstrating the above phenomenon; the models considered are two simple jump-Markov inertia navigation measurement models. It is revealed that when the Kalman filter pdf for different jumps states are separated despite measurement noise, tracking of the jump state is more effective.