Risk-Averse Robustness-Based Intelligent Environment-Adaptive GNSS/INS Tightly-Coupled Positioning Method

In urban environments, Global Navigation Satellite System (GNSS) positioning is susceptible to outliers, which severely degrade reliability and hamper accuracy even with inertial navigation system (INS) assistance. This paper pro poses an Adaptive Risk-Averse Performance-Specified (ARAPS) method based on GNSS real-time kinematic/INS integration. This framework dynamically selects measurements to minimize outlier risk while meeting performance constraints, utilizing slack variables when strict feasibility is unattainable. Complementing this optimization, a set transformer network estimates measurement uncertainty to enable adaptive weighting. Experimental results demonstrate that ARAPS achieves 87.28% of horizontal errors below 1.5 meters and 92.68% of vertical errors below 3 meters. Compared with the extended Kalman filter, the root mean square error is reduced by approximately 88.78%. Moreover, the method achieves the lowest processing time among other learning-based approaches, satisfying the real-time requirements of on-board systems. The results indicate that ARAPS offers a feasible and robust solution for high-precision positioning in dense urban environments.