A Positioning Error Correction Method for Dead-Reckoning Navigation of Underwater Vehicles

Dead-reckoning (DR) navigation systems have been widely applied by autonomous underwater vehicles (AUVs). In DR navigation, due to the lack of direct position measurement, the AUV's position is calculated from the velocity and attitude. As a result, the DR positioning error accumulates over time. To conquer this problem, the existing researches implement sensor error corrections. By reducing the errors of velocity sensor and attitude sensor, the accumulation speed of DR positioning error is restrained. However, the sensor errors cannot be fully corrected. To further improve the DR positioning accuracy, this paper proposes a novel correction method that directly acts on the DR positioning error. The core of the proposed method is a single-step correction strategy in which a radial basis function neural network (RBFNN) is designed as an error predictor. Before underwater navigation, the RBFNN is trained on the water surface with the assistance of global navigation satellite system (GNSS). The RBFNN input is consistent with that of DR navigation. The RBFNN target output is the error between DR position and GNSS position in a single DR period. Through training, the RBFNN approximates the function between DR input and single-step positioning error. In real-time underwater navigation, it can predict the single-step positioning error to correct the DR position. Thus, the proposed method provides a new technical route to suppress the error accumulation of DR navigation. It can not only operate independently, but also be combined with the sensor error correction for multi-level correction. The effectiveness and superiority of the proposed method are verified by realistic simulations and field experiments. In the field experiments of 'Haiwei1' AUV, the DR positioning errors are reduced by 14.6%-72.1%.