This paper deals with using of low-cost Global

Navigation Satellite System (GNSS) sensors in a localization

process for an autonomous guidance system of mobile robots.

Generally, this process is made using a Kalman Filter (KF) to

fuse information coming from different sensors. But as GNSS

error is an unpredictable stochastiscal process, the localization

estimated by the KF becomes unreliable. To solve this problem,

the error of a cheap GNSS receiver is analyzed. Then, an

AutoRegressive process (AR process) is used to establish a reliable

prediction model. A second problem of GNSS systems concerns

the disturbances in the observation of satellites. To detect this

disturbance, we use a condition based on the Mahalanobis

distance. This model and condition are taken into account in

the localization system to improve its