Abstract: Robust beamforming on mobile platforms is challenging because platform attitude variation causes snapshot nonstationarity of the array manifold, which makes the conventional sample covariance matrix inaccurate and aggravates desired-signal leakage. To address this problem, a robust beamforming method under snapshot nonstationarity for mobile platforms is proposed. First, a snapshot-level signal model is established, and motion compensation is performed by using attitude measurements to reduce cross-snapshot manifold inconsistency. Then, a target blocking matrix is constructed from the reference steering vector, and the compensated training data are projected onto the subspace orthogonal to the desired signal to suppress desired-signal leakage. After that, a shrinkage covariance estimator is adopted to improve the robustness of covariance estimation under limited snapshots. Finally, the beamforming weight vector is obtained according to the minimum variance distortionless response criterion. Simulation results under four scenarios, including a static platform, small yaw jitter, yaw jitter with local scattering, and platform drift with IMU(inertial measurement unit)/calibration errors, show that the proposed method achieves better output SINR than Raw SMI, Raw Shrink, MC-SMI, MC-Shrink, and two covariance-reconstruction-based methods under varying input SNR, number of snapshots, and motion strength. The normalized beampattern results further verify that the proposed method preserves the mainlobe more accurately and forms deeper nulls in interference directions. Therefore, the proposed method can effectively alleviate the joint influence of snapshot nonstationarity, desired-signal leakage, and small-sample fluctuation, and is suitable for robust beamforming on mobile platforms.