The real time positioning system is necessary and essential for autonomous agricultural vehicles. In general, the external sensor used for positioning system equipped with vehicle has no practical use because of being unable to obtain the correct location information of a vehicle due to output error of a sensor. In this study, accuracy of the supersonic waves sensor which is used as the external sensor, was estimated, and further, the learning system of neural network (NN) was applied to decrease of output error of the sensor. As a result, error and fluctuation of sensor output became large according to revolutionary velocity of the sensor unit under the condition of measuring with sensor only. On the other hand, in case of revising a sensor output based on actual vehicles' location by the NN learning system, sensor output was almost in agreement with actual location without the influence of revolutionary velocity of the sensor unit. Consequently, the learning system of NN is available for improving the accuracy of external sensor as a positioning system of the autonomous vehicles.