Abstract: In agricultural field, wheeled mobile robot navigated by machine vision is researched widely at present. An optimal lateral control algorithm is presented in this paper, but whose performance can be deteriorated by long delay feedback produced by machine vision recognizing the characteristics of complicated field environment. So based on its kinematical model, behavior of wheeled mobile robot is predicted so as to modify feedback information sampled by machine vision. In the simulation, it is showed that this algorithm can carry out the lateral control of agricultural wheeled mobile robot precisely and reliably, and can adapt different longitudinal velocity excellently.