Internal inspection of storage tanks can be long, costly and even detrimental to the health of the inspector. An alternative to human inspection is the use of robotic systems. These systems can be teleoperated from outside the tanks, making it possible to carry out the inspection more safely, quickly and in some cases without having to empty it. In order to provide the location of any defects in the tank, the mobile robot must be able to know its relative position within it. Self-localization is of great importance for mobile robot navigation. Inspection robots are, for the most part, vehicles with wheels or tracks. This configuration adds two difficulties that need to be addressed in the localization task. Due to its configuration, in this type of vehicle, wheel slip is intrinsic to its operation, being essential to take into account its effect to model its behavior properly. Another difficulty is the use of magnetic wheels, due to the strong magnetic field generated by these elements, which interfere with the measurements of magnetic sensors, such as compasses. In this work, a Kalman filter was developed and implemented for the localization of a four-wheel fixed magnetic robot, from the fusion of inertial sensors and odometry. In the modeling of the vehicle, a kinematic model was used as the basis for a dynamic model, which allowed to consider the intrinsic slippage of the system. In the sensor fusion, measurements of the magnetometer on board were discarded, due to the great interference produced by the wheels and the great distance that would be necessary between them to be unaffected by noise. Simulations and experiments have proven the efficiency of the implemented filter.