One of the problems with solutions that involve mobility is to accurately estimate the robot s position. In an outdoor environment, the GPS sensor is the most commonly used method because it provides a global position, but with an error margin that is greater than just a few meters, and creates a dependency on the visibility of the satellites. Another solution is to use an inertial sensor, which at the beginning of the operation shows good accuracy, but the positioning error grows indefinitely because it is calculated by a double integral of acceleration and angular velocity measures. This work develops a system for localization of mobile robots in outdoor environments. The positions are estimated via GPS and inertial sensors, combined using a Kalman filter, reducing the uncertainty. The equations and two distinct implementations of the filter will be presented. A classical implementation and an extended version for low-grade inertial measurement units, which utilizes the orientation given by compasses in the filtering process. The effectiveness of the Kalman filter navigation is verified through experimental and simulation results. The performance gain of the extended filter in comparison to the classic is also verified.