Deep Learning-Enabled Fusion to Bridge GPS Outages for INS/GPS Integrated Navigation

The low-cost inertial navigation system (INS) suffers from bias and measurement noise, which would result in poor navigation accuracy during the global positioning system (GPS) outages. Aiming to bridge the GPS outages duration and enhance the navigation performance, a deep learning network architecture named GPS/INS neural network (GI-NN) is proposed to assist the INS. The GI-NN combines a convolutional neural network and a gated recurrent unit neural network to extract the spatial features from the inertial measurement unit (IMU) signals and track their temporal characteristics. The relationship among the attitude, specific force, angular rate and the GPS position increment is modelled, while the current and previous IMU data are used to estimate the dynamics of the vehicle via the proposed GI-NN. Numerical simulations, real field tests and public data tests are performed to evaluate the effectiveness of the proposed algorithm. Compared with the traditional machine learning algorithms, the results illustrate that the proposed method can provide more accurate and reliable navigation solution in the GPS denied environments.