State Estimation for Vision-based Localization under Uncertain Conditions

TL;DR

This paper develops and tests advanced EKF-based estimators for vision-based robot localization in environments with semi-static landmarks and partial visibility, enhancing robustness and accuracy.

Contribution

It introduces an observer framework incorporating landmark position changes and a multi-rate estimation approach for improved localization under uncertain landmark visibility.

Findings

A-EKF and PI-EKF effectively handle landmark position changes.

Multi-rate estimation improves localization when landmarks are intermittently visible.

Simulation results demonstrate robustness of proposed methods.

Abstract

Vision based localization is a popular approach to carry out manoeuvres particularly in GPS-restricted indoor environments, because vision can complement other activities performed by the robot. The objective is to estimate the current location with respect to a known location by matching the bearings. The problem is challenging as the known location information is in terms of the bearings of landmarks extracted from an image. We address the problem under more challenging scenario when landmarks are semi-static. In this work, an observer formulation is presented which enables to incorporate the effect of change in landmark position as parameters. The efficacy of two estimators: Augmented Extended Kalman Filter (A-EKF) and a Proportional-Integral EKF (PI-EKF) is tested under the cases where there are changes in some of the landmark positions. Morever, it is likely that not all landmarks are visible to the robot at all instants of time. A multi-rate estimation framework is proposed to mitigate this issue. Observability analysis is carried out to arrive upon a minimum number of landmarks required for such a formulation. Simulation studies are presented to test the efficacy of the formulations.