ULTRASONIC SENSOR DATA INTEGRATION AND ITS APPLICATION TO ENVIRONMENTPERCEPTION

To move in an unknown or uncertain environment, a mobile robot must co llect information from various sensors and use it to construct a repre sentation of the external world. Ultrasonic sensors can provide range data for this purpose in a simple and cost-effective way. However, mos t ultrasonic sensors are not sufficient for environment recognition be cause of their large beam opening angles. In this article the beam-ope ning-angle problem is solved by fusing data from multiple ultrasonic s ensors. We propose two methods for sensor data fusion. One uses an art ificial neural network (ANN), and the other is based on a mathematical model. Simulations and experiments show that the mathematical model i s more accurate when there is no noise in the sensor readings, but the ANN method is better when the sensors are subject to much noise. To e xtract line segments from the ultrasonic image, we develop a line extr actor that is more efficient than traditional line fitting methods in this application. Experimental results show that this method is effect ive for environment perception in a robotic system. (C) 1996 John Wile y & Sons, Inc.