Reducing Complexity of GPS/INS Integration Scheme Through Neural Networks

A vehicle-mounted GPS receiver used for localization can suffer from signal blockage. To remedy to this problem, GPS/INS integration can be considered as a solution in some cases. However, in the case of urban areas where there are severe multipath conditions, the performance degrades considerably. The last decade have seen many proposals of techniques aiming at improving the accuracy of GPS positions. The complexity of these techniques increases with the increase of the accuracy required. These techniques usually combine Kalman Filters (KF) with neural network or fuzzy logic. etc. In this paper, we propose a new technique based solely on neural network, which offers a better performance while presenting a lower complexity. The idea is to use a neural network, which emulates the behavior of a given estimator in order to replace it. We present simulations results, which validate the performance and the robustness of our proposed scheme in various conditions.