Adaptive cooperative localization algorithm for AUV swarms in complex underwater acoustic environments

Abstract: Objectives To improve cooperative localization of master-slave autonomous underwater vehicle (AUV) swarms under unknown time-varying noise and measurement outliers, a Mahalanobis distance-assisted variational Bayesian expectation-maximization unscented Kalman filter (MD-VBEM-UKF) is proposed. Methods Under the variational Bayesian expectation-maximization framework, prediction error covariance and measurement noise covariance are jointly estimated online. The expectation-maximization mechanism dynamically updates the prior scale matrix of one-step prediction error covariance. A Mahalanobis distance-based outlier detection and isolation strategy is adopted before measurement update to reduce outlier interference in state estimation and covariance updating. Results Experimental results show that under non-stationary noise, VBEM UKF reduces the average localization error and RMSE by 48% and 50%, respectively, compared with UKF, and delivers slightly better positioning accuracy than Sage-Husa UKF. With measurement outliers, MD-VBEM-UKF cuts the two metrics by 20% and 24% relative to VBEM UKF. Simulations using measured navigation trajectories verify that the proposed algorithm achieves stable localization with a reasonable acoustic measurement update period. Conclusions The algorithm achieves online adaptive estimation of prediction error covariance and measurement noise covariance, effectively suppresses the impact of acoustic measurement outliers, and improves the localization accuracy and robustness of master-slave AUV swarms.