Line based robot localization using a rotary sonar

Abstract

Ultrasound sensors, despite their low cost, are less common than other sensors used for robot localization. This is because its main drawback: The lack of reliable bearing information. We introduce a rotary ultrasonic sensor able to extract features in a 360deg-scan and classify them as wall or corner. For robot localization purposes, these features are matched with a set of line-features stored on a priori known map. The localization process is carried out through an Extended Kalman Filter that fuses data provided by a dead-reckoning system with measurements data provided by the rotary sonar. This paper also presents techniques we use to classify the detected features. The experimental test carried out in a long corridor shows the usefulness of the rotary sensor for feature extraction and robot localization process. The results show that localization error is in subcentimeter range.