Using Geographical Data and Sonar to Improve GPS Localization for Mobile Robots

Internet technology and the availability of large public knowledge bases should enable future autonomous systems to drastically improve their perceptual and cognitive capabilities with only inexpensive sensors. In this paper we investigate this aspect with respect to robot self-localization. We present a method to improve GPS-based localization of mobile robots using geographic data from a public database. From a cadastral map a basic map of a robot’s working area is automatically created. A mobile robot is equipped with a low-cost GPS receiver and ultrasonic sensors. Then, a particle filter is used to fuse GPS position values and odometry data and to match sonar scan data with the a priori geodata map. The map is also updated with previously unknown environment features. The algorithm was tested in an outdoor environment with uneven terrain. Experimental results show considerable improvements in position estimation compared to using GPS alone.