4cm。 An autonomous navigation system based on machine vision for agricultural vehicle was studied. A novel line detection algorithm was proposed for navigation directrix detection, which could decrease the cost of memory and time considerably. A 2-D fuzzy controller with the inputs of lateral error and heading error was built to determine the expected angle of turning wheel, and a turning controller based on PID was constructed to control the turning of front wheel. The simplified kinematics model of two-wheel vehicle was adopted for simulation. The results of simulation and experiments indicated that the navigation system could track the desired beeline path effectively. The maximum tracking lateral error was less than 5cm and 8cm at the speed of 0.3m/s and 0.6m/s respectively, and the average lateral error was less than 2cm and 4cm respectively.