As a l1/l2 norms-based estimation method, Huber's M- estimation has provided an efficient method to deal with measurement outliers for robust filtering, which has been applied to the cubature Kalman filter (CKF), namely Huber's M-estimation based robust CKF (HCKF) and its square-root version (HSCKF). To further handle abnormal measurement noise, an adaptive method is proposed in this paper to adjust the measurement noise covariance used in the Huber's M-estimation approach based on the difference between actual and theoretical innovation covariance, leading to adaptive HCKF (AHCKF) and adaptive HSCKF (AHCKF). Simulation results on a typical target tracking model have demonstrated their advantages over existing approaches in terms of estimate accuracy, outlier-robustness and reliability.