Without the assistance of GNSS, errors in low-cost INS will spread rapidly resulting in a steady decline in accuracy. Yet, GNSS signals are easily blocked by various obstacles in urban environments. To solve the problem above, this paper proposes an INS error modeling algorithm for GNSS/INS integrated navigation in urban environments based on Long Short-Term Memory (LSTM). The dynamic nonlinear mapping and long-term memory ability of LSTM allows us to make full use of the relationship between past and current navigation information to realize accurate prediction. When sufficient GNSS satellites are observable, the LSTM model is trained with the specific force, angular velocity, position, velocity from the last epoch of INS, combined with the position and velocity from the current epoch of integrated system, so that it can fully learn the rules to compensate for the INS navigation error. Then, when insufficient satellites are observable, the trained LSTM model outputs the corresponding position and velocity to correct the error of INS. The proposed algorithm is verified using the RTK/INS loosely-coupled model, observing GPS and BDS satellites at the same time. When the GNSS are unobservable for 30 s, compared with the pure INS solution, the accuracy of RTK/INS based on LSTM is improved by 69.83%, 78.90% and 80.74% in three different experimental scenarios. This demonstrates a significantly better navigational ability during periods of GNSS outages.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    A Long Short-Term Memory-Based INS Error Modeling Algorithm for GNSS/INS Integrated Navigation in Urban Environments


    Additional title:

    Lect. Notes Electrical Eng.


    Contributors:
    Yang, Changfeng (editor) / Xie, Jun (editor) / Jiang, Lei (author) / Sun, Rui (author)

    Conference:

    China Satellite Navigation Conference ; 2022 ; Beijing, China May 22, 2022 - May 25, 2022



    Publication date :

    2022-05-07


    Size :

    10 pages





    Type of media :

    Article/Chapter (Book)


    Type of material :

    Electronic Resource


    Language :

    English