To effectively overcome the interferences of GNSS signals and enhance the reliability of multi-sensor integration positioning in complex urban environments, a robust Mahalanobis distance statistic-based multi-sensor integration robust estimation method is proposed. With the basis of faulty measurements evaluation and typical model of variance inflation robust estimation, the robust Mahalanobis distance statistic is constructed based on the adjacent innovation sequences. The introduction of past innovation contributes to the observation redundancy. Meanwhile, the robustness of anomaly detection statistics can be improved by interacting between innovations from different measurements. According to the statistical property of this robust distance, the critical thresholds are ensured and then the measurement noise covariance can be adjusted adaptively with two traditional weighted strategies. Some experiments have been implemented on the INS / GNSS / LiDAR/ VINS vehicle positioning system in an urban canyon environment. It shows that compared with existing methods, the 3D positioning error root-mean-square of proposed method is limited within 3. 37 m. The superiority of our method is further validated by analyzing the positioning results with different significances.