Abstract: Aiming at the problems of the current differential vehicle navigation algorithm, such as large turning radius, unable to track long-distance path, complex navigation logic, and difficult development, this study uses the idea of modular and hierarchical development, and designs a navigation control algorithm based on ROS (robot operating system). And through the transformation of four wheel differential speed vehicle for testing. The algorithm describes the target path by the way of target point plus target yaw angle, and the target path is an infinite straight line through the target point whose direction is the target yaw angle. The vehicle can start from any position and approach the target path, and finally reach the target point. The kinematic equation of the vehicle is obtained by the kinematic analysis of the four-wheel differential speed vehicle, and the relationship between the forward speed and steering speed of the vehicle with the speed of the left and right wheels is obtained. Then the lateral deviation between the vehicle and the target path, and the longitudinal distance between the target point and the vehicle are obtained by coordinate transformation. The required forward speed is calculated based on the longitudinal distance, the driving angle close to the target path is determined based on the lateral deviation. Then the required steering speed is obtained after the difference between the driving angle and the yaw angle of the vehicle itself. Based on this idea, it brings two advantages. First is that it improves the robustness and flexibility of the algorithm, which can be stably solved under any size of lateral deviation and simplifies the complexity of path planning. It can focus on the path planning between lines and ignore the path planning between lines, and let the algorithm automatically close to the target path of the job. Second, the internal parameters of the algorithm all have specific meanings, which are easy to understand and adjust, and can be easily adjusted according to different needs to adapt to different environments and operations. The above calculation is realized by the double variable limiter PID algorithm. The forward speed and steering speed obtained are converted into the left and right wheel speed through the vehicle's kinematic equation. Then the wheel speed is input into the closed-loop PID of the lower computer to realize the accurate control of the left and right wheel speed, so as to stably control the vehicle to move according to the calculation results of the algorithm and reduce the disturbance caused by the vehicle itself. Through simulation experiments, cement ground experiments and field experiments, the accuracy and adaptability of the algorithm are verified in three aspects: the average value of lateral deviation, the coefficient of variation and the maximum deviation. The simulation results show that the average error is 1.146 cm, the coefficient of variation is 47.535%, and the line distance is less than 4.092 m when the four-wheel differential vehicle controlled by the path following algorithm runs at the speed of 0.4 m/s. The navigation system has good stability and adjustment ability. At the same time, it can also be observed from the driving trajectory that the driving angle close to the target path is determined based on the unique lateral deviation of the algorithm, so that the vehicle can drive perpendicular to the target path when it is far from the target path, and approach the target path quickly, avoiding the defect of the traditional algorithm that needs to slowly adjust the direction to the target path in the same environment. From the comparison of different parameters in the simulation, the influence of different parameters on the control effect of the algorithm can also be observed, which also shows that the parameter adjustment of the algorithm is simple and clear to a certain extent. In the cement ground test, the average error was 2.833 cm and the coefficient of variation was 68.478% when the vehicle was driving at the speed of 0.4 m/s on the cement pavement environment. When the vehicle was driving at the speed of 0.4 m/s on the fruit tree pavement environment in the field test, the average error was 5.249 cm and the coefficient of variation was 62.804%. It can be seen that with the increase of the ruggedness of the ground environment, the average error rises slightly, and the variation coefficient changes little, which indicates that the robustness of the algorithm is strong and the anti-interference ability is excellent. The above test results show that the navigation system and the test vehicle have better adaptability in different terrain environments, and the research results can provide reference for the navigation technology of differential steering chassis.