Fig. 1

Architecture of the proposed robust Bayesian filter. The filter takes a prior probability distribution of the linearization points and state and uses a system transition model and measurement model to estimate the posterior probability distribution of the state. The linearization point probability distribution is represented as weighted samples, and the state probability distribution is represented by a Gaussian distribution tracked by the underlying robust Kalman filter. The filter first applies the prediction step to each linearization point in the prior distribution. Using this predicted point for linearizing the robust filter system transition, measurement and cost models, we then apply the predict and update steps in parallel to update the state probability distribution for each linearization point. Based on the updated state probability distribution, the linearization point samples are refined and their weights are updated using the available measurements. Finally, the weighted linearization point samples are resampled and combined with the updated state probability distribution to represent the posterior state probability distribution