Junction Tree Factored Particle Inference Algorithm for Multi-Agent Dynamic Influence Diagrams

摘要

As MMDPs are difficult to represent structural relations among Agents and MAIDs can not model dynamic environment, we present Multi-Agent dynamic influences (MADIDs). MADIDs have stronger knowledge representation ability and MADIDs may efficiently model dynamic environment and structural relations among Agents. Based on the hierarchical decomposition of MADIDs, a junction tree factored particle filter (JFP) algorithm is presented by combing the advantages of the junction trees and particle filter. JFP algorithm converts the distribution of MADIDs into the local factorial form, and the inference is performed by factor particle of propagation on the junction tree. Finally, and the results of algorithm comparison show that the error of JFP algorithm is obviously less than BK algorithm and PF algorithm without the loss of time performance.