Aiming at the problems that there is a lot of uncertain disturbance in the nonlinear vehicle model of agricultural machinery, and the measurement is often with noise, the moving horizon estimation (MHE) method for vehicle motion state was proposed. The state estimation problem was transformed into a fixed time domain optimization problem and the constraint conditions were fully considered. In order to improve the computational efficiency of MHE, taking into account the different sampling frequencies of sensors and the possibility of missing or abnormal measurement values, a multi-threading architecture was designed. The multi-threading architecture also can make MHE more suitable for practical applications. The automatic driving simulation system of the rice seeding machine was established by Matlab. The simulation results showed that MHE can effectively suppress system disturbance and measurement noise. The x and y positions and heading angle estimated by MHE were closer to the truth value than those estimated by extended Kalman filter (EKF). MHE was used to estimate the lateral deviation and heading angle deviation measured during the autonomous driving process of the rice seeding machine. The results showed that when the time domain window N was 3~5, the MHE algorithm had a good effect on suppressing the jump of measurement value, and it can also reflect the real trend of state value. It proved that MHE had excellent performance in suppressing system disturbance and measurement noise.