Performance characterisation of a feature-based Gaussian pose tracker for mobile robots in indoor environments

This paper introduces a redundant localisation system, combining a low-cost Gaussian pose tracker and a particle filter localiser which is more CPU intensive. Ideally, the Gaussian pose tracker can be used for most of the time and the particle filter localiser switched on only when the Gaussian pose tracker fails. The problem studied here is to determine when the Gaussian pose tracker has failed. A number of different measures are proposed which indicate the state of the pose tracker. Experimental results are presented which illustrate the relative performance of the proposed measures. The results demonstrate that a measure derived from the size of the covariance of the Gaussian pose tracker gives a good prediction of when the pose tracker has failed.