Graphical Approach to Representation and Inference in Multi-sensor State Estimation

A graphical approach to design of an IMM-PDAF (Interacting Multiple Model – Probabilistic Data Association Filter) is presented. In particular, a Dynamic Bayesian Network (DBN) is used to model the temporal evolution of an INS/GNSS integration system that will be used in self-driving cars. Conditional independence embedded in the network is utilized to obtain the system and measurement model that will be the basis of prediction and update steps of the IMM-PDAF. Performance evaluations show that the resulting filter produces acceptable Root Mean Squared (RMS) errors in attitude, position and velocity of a car. A further discussion reveals that this approach is a Plug-and-Play All Source Positioning and Navigation (ASPN)-capable candidate.