Comparison of Kalman filter and particle filter used for localization of an underwater vehicle

This paper compares filtering methods used for localization of an underwater robot: Kalman filter and particle filter. Kalman filter and particle filter are major filters for estimation of robot pose on the ground. They are adapted for underwater robot localization. While Kalman filter can be used for linear or linearized processes and measurement system, the particle filter can be used for nonlinear systems. Also, the uncertainty of Kalman filter is restricted to Gaussian distribution, while the particle filter can deal with non-Gaussian noise distribution. In cases where abrupt sensor noise is rarely observed, both filters work fairly well. However, when sensor noise exhibits jerky error, Kalman filter results in location estimation with hopping while particle filter still produces robust localization. The paper also compares performance of these filters under various measurement uncertainty and process uncertainty. The methods are compared and verified through experiments.