In complex positioning environments, the nonlinear nature of GNSS/INS combined navigation systems, coupled with noise distributions often deviating from the Gaussian model, poses challenges to traditional Kalman information fusion techniques. This can result in a priori statistics misestimation, leading to diminished filtering accuracy or even result dispersion. To address these issues and enhance the positioning accuracy of GNSS/INS systems in such environments, this study introduces an enhanced Adaptive Robust Cubature Kalman Filter (ARCKF) fusion algorithm. The proposed algorithm begins by employing the Mahalanobis distance noise estimation method to assess the conformity of the volume measurement value's probability density function to the Gaussian distribution. If non-conformity is detected, the algorithm updates and corrects the volume measurement noise using a dynamically calculated scale factor. Subsequently, updates to the mean square deviation matrix of volume prediction and Kalman gain coefficients are performed, thereby enhancing system estimation accuracy and resolving filter convergence challenges. Simulation and comparative tests demonstrate the algorithm's effectiveness. Compared to the Extended Kalman Filter (EKF) algorithm, the proposed approach reduces the root mean square error of position in longitude, latitude, and altitude by 32.21%, 40.73%, and 44.68%, respectively. Furthermore, compared to the Cubature Kalman filter (CKF), reductions of 21.33%, 11.73%, and 4.66% are observed in the respective directions. These findings underscore the algorithm's ability to significantly enhance system reliability.