Adaptive robust SINS/DVL/USBL integrated navigation using persistent homology-based USBL state classification

For autonomous underwater vehicles (AUVs), SINS/DVL/USBL integrated navigation provides continuous underwater positioning. However, multipath propagation, non-line-of-sight (NLOS) effects, and prolonged USBL outages can degrade acoustic measurements and yield innovations with heavy tails. This article develops an adaptive robust navigation framework that integrates USBL state recognition using persistent homology with state-dependent compensation and Student’s t robust fusion. Real Songhua Lake data generate geometry-constrained USBL degradation scenarios while preserving vehicle motion and measurement geometry. Delay-embedded USBL windows are mapped to persistence images (PIs) using a Student’s t kernel, and a 100-tree Random Forest classifier uses 500 labeled windows in total; across three cross-scenario transfer tests, it achieves an average accuracy of 0.6863. The estimated state regulates compensation and robust filtering so that reliable USBL updates are used, biased acoustic updates are down-weighted, and position information is maintained during outages. Compared with USBL measurement, KF, MCCKF, SSMKF, and the previous RKF method, the proposed robust navigation framework achieves the lowest average errors over six representative scenarios, with mean absolute error, root mean square error, and axis root mean square error of 9.86 m, 27.33 m, and 19.32 m, respectively.