Abstract: The complex working environment, uneven road surface, obstacles, and other factors will reduce the accuracy and efficiency of agricultural robots, so the controller and control algorithm of the steering system need to be optimized and tested continuously. The development of traditional controllers and algorithms relies on field experiments, long r＆d cycles, and many interference factors. Therefore, a semi-physical simulation platform for the steering system is established, which mainly includes a steering closed-loop control system, a torque closed-loop control system, input and output modules, and a real-time system based on Windows+RTX. The steering closed-loop system is modeled, the compound control algorithm based on feedforward compensation is adopted, and the established model and algorithm are simulated and verified in Matlab/Simulink. The feedforward compensation composite algorithm is compared with the uncompensated algorithm. The simulation results show that the maximum absolute error is reduced by 72.3%. Further experimental verification shows that the response time of the system is 0.380 s, the overshoot is 0.296%, the maximum tracking error is 0.15°, and the mean absolute error is 0.04°. When the amplitude is 30°, the mean absolute error is reduced by 94.23%, and the maximum absolute error is reduced by 88.68%.