A Sensor Fusion Framework for On-Line Sensor and Algorithm Selection

Abstract

This paper presents a sensor fusion framework for mapping unknown environments for mobile robots. The proposed framework enables on-line selection of the most reliable logical sensors and the most suitable fusion algorithm. The framework is rule-based, employing the “simplest sensor fusion algorithm with the most reliable sensors” concept. This goal is achieved through measures that were developed to quantify on-line the performance of the sensors. The framework was evaluated in an experiment consisting of a mobile robot equipped with five logical sensors. The framework was compared to four other algorithms. The advantages of this new framework are presented using statistical, histogram, time series and graphical analyses.