A data-driven based hybrid multi-branch framework for AUV navigation

In the Autonomous Underwater Vehicle (AUV) navigation task, aiming at the problem that traditional state estimation techniques introduce multiple errors affecting the navigation accuracy, and considering the uniqueness of different navigation sensor parameters, this paper proposes a hybrid multi-branch network framework for high-precision AUV navigation that can extract local characteristics while capturing the longterm temporal dependence of the input sequences. Firstly, each input time series is separately processed by the One Dimensional-Convolutional Neural Network (1D-CNN) based feature extraction module to provide a feature representation with various parameters. After that, the extracted features are concatenated to obtain anew time series and fed into the Long Short-Term Memory (LSTM)-based feature fusion module to learn the long-term temporal dependencies in the series. Finally, the output of the network can be obtained by performing regression calculations through the Fully Connected (FC) layer. Sailfish 210 AUV actual sea trial data has been used to validate the effectiveness of the proposed algorithm.