Particle filtering for dynamic systems with future constraints

This work focuses on a new state estimation problem for dynamic systems with future constraints. It is well known that constraints, as a priori, should always be satisfied by states of some dynamic systems, and thus sufficiently fusing the valuable information into filters will produce more accurate estimates. To this end, a particle smoothing framework is proposed by incorporating future constraints, instead of only current constraints. Next, we construct a criterion function by making the center value of dynamic model satisfy constraints to improve dynamic modeling accuracy, and investigate a cost function to compensate for the impact of the modified dynamic model on the original predicted distribution. Based on this, we present a new particle filter that future constraints are applied in importance sampling and updating particle weights. Especially, the modified dynamic model is chosen as the importance density function so that the sampled particles satisfy constraints at consecutive times as much as possible. Finally, two simulations verify the superiority of the proposed filter.