Q‐Learning‐Based Adaptive Student's tt Maximum Correntropy Cubature Kalman Filter for Non‐Gaussian Noise With Unknown Noise Covariances

The cubature Kalman filter (CKF) is widely used for state estimation in nonlinear systems but struggles with noise uncertainties, non‐Gaussian measurement noise, and outliers. This paper introduces a novel Q‐learning‐based adaptive Student's ‐distribution maximum correntropy CKF (QL‐STMCCKF) to address these issues. By integrating reinforcement learning and maximum correntropy criteria with the CKF, the method improves robustness. The Student's ‐distribution kernel handles heavy‐tailed non‐Gaussian noise, while Q‐learning adaptively updates process and measurement noise covariances, enhancing estimation performance. A maximum likelihood estimation (MLE)‐based variant, MLE‐STMCCKF, is also developed for comparison. The algorithms are tested on a benchmark aircraft tracking problem and validated with offline hardware data from a grid‐connected 3‐phase voltage source inverter. Comparative analysis highlights the superior performance of the QL‐STMCCKF in challenging conditions.