Abstract: Methods, and systems, for Multiagent Pathfinding for Non Dynamic Programming Problems, including a CE method which provides for sampling from a complex probability distribution that is not necessarily known in a closed form. The applications of this method include rare-event simulation, variance reduction for estimation problems, and stochastic optimization. The method iteratively searches for a probability distribution that is “close” to the intended distribution, where the closeness of distributions is measured using the Kullback-Liebler (KL) divergence between the distributions. At each step, the method generates samples according to a current candidate distribution from the family. Next, it uses those current candidate distribution samples to move the distribution toward a new candidate distribution that is closer in the sense of KL divergence to the target distribution.

Abstract: Methods, and systems, for Multiagent Pathfinding for Non Dynamic Programming Problems, including a CE method which provides for sampling from a complex probability distribution that is not necessarily known in a closed form. The applications of this method include rare-event simulation, variance reduction for estimation problems, and stochastic optimization. The method iteratively searches for a probability distribution that is “close” to the intended distribution, where the closeness of distributions is measured using the Kullback-Liebler (KL) divergence between the distributions. At each step, the method generates samples according to a current candidate distribution from the family. Next, it uses those current candidate distribution samples to move the distribution toward a new candidate distribution that is closer in the sense of KL divergence to the target distribution.