<title>Robot location densities</title>

Abstract

Any niovernent of a niobile robot increases the uncertainty in the precise position of the robot. Sensor itteasurernents of the environnent gathered on board the niobile robot help in reducing this uncertainty. The uncertainty in robot position niay be eiabedded in a probability density function which we refer to as the robot location density. This paper presents iiiethods of expressing this density in terms of sensor measurement characteristics. Vision and range data provide information about the position of the robot. We show how this information may be incorporated into the location density. The probability mass in a region corresponds to the likelihood of the robot being in that region. Sensor data can effect this likelihood. For example an ultrasonic range system may provide information that the robot is close to a region boundary. The probability mass in the inner regions would thus be decreased when this range data is used to update the location density. Motion of the robot also influences the a posteriori location density. A collision event or the absence of this event both provide information about robot position. We show how the location density may be used for collision avoidance and robot position estimation (or robot self-location). Optimal methods are presented to determine the maximum likelihood estimate of the robot location in a known polygonal environment using range data. The methods may be used with or without compass