Robust Bayesian filtering for positioning using GPS & INS in multipath environments

While traveling on roads surrounded by high-rise buildings, positioning units that use global positioning system (GPS) data often suffer large multipath errors. To improve the accuracy of GPS units, we propose a novel robust Bayesian filtering algorithm that can distinguish outlier signals affected multipath errors. The proposed method implements sequential estimation using multiple hypothesis tracking. In the proposed method, an observed distribution of the GPS satellite signal is assumed to be a mixture of a Gaussian distribution of normal values and a Cauchy distribution of abnormal values due to multipath errors. The proposed method generates two hypotheses of the normal values and abnormal values. To limit the number of hypotheses, we introduced Gaussian mixture reduction based on Kullback-Leibler divergence. Experiments with real driving data show that the proposed method is more robust than previously reported approaches based on extended Kalman filters or optimization algorithms.