Abstract :
The periodic re-localization of a mobile robot is necessary in order to reduce the uncertainty about its position and orientation when it moves. These uncertainties are due, among others, to integration errors of its internal sensors, such as odometers, accelerometers, etc. So, the localization of external references can be used to update the actual position and orientation of the vehicle; to this purpose, ultrasonic sensors are widely used because of its low price and ease of use. In this work, a modeling of the obstacles in the local environment of the robot is developed as a set membership in such a way that it is feasible that all of the measurements failing into a set membership come from the obstacle associated to this class. Once that an obstacle is localized, a process of self-correction can he carried out by comparing the localization saved up in memory and the measured localization by the ultrasonic sensors. Each class for each obstacle is defined from the uncertainties in the position and orientation of the vehicle and also from the uncertainties in depth and angle of the ultrasonic range sensors. The technique called of "small variations" is used in order, to propagate uncertainties from one reference frame to another.
