A soft computing method for an AUV navigation system with pseudo-real-time applicability

Abstract

This paper describes the implementation of a soft computing method based on fuzzy logic and multiobjective genetic algorithm techniques to adapt the parameters of an error-state complementary Kalman filter (ESCKF) to enhance the accuracy of an autonomous underwater vehicle (AUV) navigation system. In the ESCKF, inertially-derived quantities from an inertial navigation system (INS) sensor are combined with direct measurements of the same quantities by use of the global positioning system (GPS) and other aiding sensors. The backlash of the integration processes however, is that errors can grow rapidly and the values obtained therein can drift off the true value significantly. By contrast, the directly-measured data contain high frequency noise with bounded error. This instinctively suggests integrating the two sets of quantities, which is exactly what the ESCKF does. To maintain the stability and performance of the ESCKF, which is likely to deteriorate when the assumed error and noise characteristics do not reflect the true ones, a fuzzy logic based scheme is used to make these values adaptive. The choice of fuzzy membership functions for this scheme is first carried out using a heuristic approach and further refined using a multiobjective genetic algorithm method.