TERRAIN AIDED UNDERWATER NAVIGATION – A DEEPER INSIGHT INTO GENERIC MONTE CARLO LOCALIZATION

This paper proposes a method for terrainaided navigation in underwater environments based on the simple means of a digitized seafloor map and sonar measurements. Since the depth information is highly non-linear and non-Gaussian, a family of probabilistic algorithms known as Monte Carlo Localization (MCL) is used for the estimation process. To increase the performance of MCL under limited computational power the sample size of the classical particle filter is adapted on-the-fly. Within this adaptation process the number of samples is increased if the state uncertainty is high and decreased if the density distribution is focused on a small area. This approach is based on a method known as KLD-sampling. As the performance of the filterand therefore the accuracy of the position estimationcan be limited by the noise levels of the system, a method is proposed to adapt the noise level over time. This adaptation is made with reference to the terrain excitation in the area of the observation. The performance of this method is presented in terms of the entropy propagation of the particle distribution.