In this article, we demonstrate a reliable, robust, and computationall
y efficient algorithm that uses inexpensive hardware to localize a mob
ile robot in a rather structured environment that is relatively consis
tent to an a priori map. Furthermore, the incorporation of thresholdin
g makes possible the localization of the robot even in the presence of
objects not depicted in the a priori map. An Extended Kalman Filter i
s used to combine dead-reckoning, ultrasonic, and infrared sensor data
to estimate current position and orientation. Implementation issues a
nd experimental results from experience with a mobile robot, Nomad 200
, are also presented. (C) 1995 John Wiley & Sons, Inc.