Detection and Compensation of Landmark Errors in Monte Carlo Localization

This paper studies a new Monte Carlo vision-based localization algorithm which performs on-line detection and compensation of measurement biases. For that purpose, the state vector is augmented to include the mobile coordinates and orientation, but also discrete latent variables indicating the validity of each landmark angular measurement. An appropriate particle filter is then proposed to solve the resulting non-linear filtering problem. The efficiency of this filter is guarantied by using relevant models for the different kinds of systematic errors corrupting the angular measurements. Simulation results illustrate the gain of the proposed approach when compared to a more conventional method.