INS Aided GNSS Pseudo-range Error Prediction Using Machine Learning for Urban Vehicle Navigation

GNSS is being extensively applied in different navigation applications. However, GNSS direct signals are easily affected by multipath and non-line-of-sight signals, resulting in severe deterioration of positioning. GNSS receiver output information, such as carrier-to-noise ratio (C/N ₀ ) and satellite elevation, cannot accurately reflect the pseudo-range quality, leading to a significant increase in positioning errors. This paper proposes an inertial navigation system (INS) aided GNSS pseudo-range error prediction approach based on machine learning for urban vehicle navigation. As an important feature, the pseudo-range residual estimated by INS is employed for model training, together with the carrier-to-noise ratio, satellite elevation, and pseudo-range rate consistency. The predicted model of the pseudo-range errors is obtained by an ensemble bagging decision tree learning method. Urban vehicle tests show that compared to GNSS single point positioning (SPP) with C/N ₀ -based weighting, the horizontal accuracy in the form of CEP95 of SPP with model-based weighting improves 52.81%, and the GNSS/INS horizontal positioning error in the form of CEP95 is reduced from 21.23m to 5.02m in deep urban environments.