A tightly coupled GNSS RTK/IMU integration with GA-BP neural network for challenging urban navigation

Abstract Intelligent transportation system is increasing the importance of real-time acquisition of positioning, navigation, and timing information from high-accuracy global navigation satellite systems (GNSS) based on carrier phase observations. The complexity of urban environments, however, means that GNSS signals are prone to reflection, diffraction and blockage by tall buildings, causing a degraded positioning accuracy. To address this issue, we have proposed a tightly coupled single-frequency multi-system single-epoch real-time kinematic (RTK) GNSS/inertial measurement unit (IMU) integration algorithm with the assistance of genetic algorithm back propagation based on low-cost IMU equipment for challenging urban navigation. Unlike the existing methods, which only use IMU corrections predicted by machine learning as a direct replacement of filtering corrections during GNSS outages, this algorithm introduces a more accurate and efficient IMU corrections prediction model, and it is underpinned by a dual-check GNSS assessment where the weights of GNSS measurements and neural network predictions are adaptively adjusted based on duration of the integrated system GNSS failure, assisting RTK/IMU integration in GNSS outages or malfunction conditions. Field tests demonstrate that the proposed prediction model results in a 68.69% and 69.03% improvement in the root mean square error in the 2D and 3D component when the training and testing data are collected under 150 s GNSS signal-blocked conditions. This corresponds to 52.43% and 51.27% for GNSS signals discontinuously blocked with 500 s.