Kalman Filters: Linear and Nonlinear applications in Sensor Fusion

Autonomous navigation is nowadays one of the hottest topic among engineering research fields and, as a matter of facts, the number of researches which are working on this field increased a lot in last years and it is expected to increase even more in future. One of the main problem, which can be encountered when facing with autonomous navigation, is the gathering of information coming from different sensors. The autonomous devices are, in fact, equipped with a large number of perception tools useful to build a model of the surrounding environment. The main question to which this dissertation aims to found an answer, is: how can this large number of data be combined in order to exploit all the available information and to obtain an improved, unique and more reliable estimation of the interesting variable? Therefore, the objective of this dissertation is the study of different techniques of state estimation which can be used to build different data fusion algorithms. In particular the focus is on Kalman-based data fusion techniques, the theory behind this kind of algorithms is deeply analyzed and the performances of them are tested by means of a real application: a Kalman-based method is used to combine data coming from visual odometry, IMU, GNSS module and encoders in order to estimate the absolute position of a UGV within a semistructured environment, such as an orchard.