Possibilistic sonar modeling and localization for mobile robots

Sonar sensors are widely used in mobile robots applications such as navigation, map building, and localization. The performance of these sensors is affected by the environmental phenomena, sensor design, and target characteristics. Therefore, the readings obtained from these sensors are uncertain. This uncertainty is often modeled by using Probability Theory. However, the probabilistic approach is valid when the available knowledge is precise which is not the case in sonar readings. In this thesis a new model of uncertainty in sonar readings is proposed by using Possibility Theory. The possibilistic approach is valid when the available knowledge is imprecise and coherent as in sonar readings. It is verified experimentally that the behavior of the sonar readings obtained from a corner is the same as the behavior of these obtained from a wall only when the sensor is at distances less than 75cm. Based on this finding, a new approach for corner detection is implemented on the mobile robot, Pioneer1. Unlike signal and image processing approaches, our approach is not time consuming because it depends on direct interpretation of readings obtained from different sensors in the same time (TOF). Stationary and dynamic localization methods are presented and applied on Pioneer1. These methods can be generalized for different robots configurations, especially the ones with a ring configuration.