In the in-motion alignment of a strapdown inertial navigation system (SINS), the unscented Kalman filter (UKF) is usually used to solve non-linear problems. The measurement noise covariance R has a direct influence on the filtering results of the alignment of the SINS. The measurement noise is assumed to follow Gaussian distribution with a constant covariance R . However, these assumptions are often not realistic, neither the Gaussianity nor the constant covariance. This will degrade the performance of the UKF. To solve this problem, this paper proposes a novel adaptive robust UKF (NARUKF). In the NARUKF, a sliding window is used in estimating the covariance R in real-time. The NARUKF is divided into three main steps, the first step is to use the Mahalanobis distance algorithm to robustify the UKF. The second step is to use the projection statistics algorithm to reweight the abnormal stored innovations. Finally, the covariance R is adaptively estimated. The simulation and experimental results for the problem of the body frame velocity aided SINS in-motion alignment under heavier-tail distribution and/or outlier conditions demonstrate the superiority of the proposed method over the traditional ones.