Fast computation of look-ahead unscented Rao-Blackwellised Particle Filters

In this paper, we explore a methodology for fast computation of the look-ahead unscented Rao-Blackwellised Particle Filtering (Fast la-URBPF) algorithm. We show that the complexity of the existing la-URBPF algorithm can be substantially reduced by restricting the unscented Kalman filtering prediction and updating step to only a representative particle of a group of particles having the same discrete state or mode. Not only can Fast la-URBPF achieve equal or much higher performance than the existing unscented Kalman filtering based algorithms, but simulation results also show that its time usage is substantially lower than those algorithms. The real data test shows its superior estimation accuracy as compared to the standard particle filtering algorithm.