Learning Inertial Measurement Error Compensation in GPS Signal Shielded Using LSTM

Abstract

GPS (Global Positioning System) is an indispensable technology in vehicle positioning and navigation. Now the GPS positioning technology is very mature and is developing towards high-precision and high-reliability technology. However, the stability of GPS needs to be improved. This paper uses RTK (Real Time Kinematic) real -time dynamic differential positioning technology that can improve GPS accuracy, as well as basic simple inertial navigation components such as gyroscopes, accelerometers, and magnetic compasses as GPS. Compensation during interruption improves the reliability of GPS positioning. However, the error of the long-term inertial navigation system accumulates over time, which seriously affects the navigation accuracy, and the accuracy of the simple sensor output is not high. Therefore, this paper proposes a neural network-like learning scheme that uses LSTM to achieve high-precision and reliable positioning. We use cars to collect the XY position coordinate data of the original vehicle around the urban area without difference and with difference positioning. Use MATLAB offline operation to calculate λ (longitude), Φ (latitude) and use the data measured by integrated inertial elements to assist navigation in the road section that is shielded by GPS signals. And use LSTM deep learning to correct its errors, and then compare with and without differential positioning methods to get a more optimized path map to achieve the compensation effect.