In this article, we present different works we have done on a guidance-assistance system for two agricultural vehicles. We show different difficulties found in our projects. Firstly the environment perception described by means of image processing algorithms which use supervised or unsupervised methods. In this part we notice the need to qualify the reliability of the perception information in order to guide correctly the vehicles. To do that, we give a solution using the theory of evidence which formalizes the uncertainty and computes a reliability criterion associated with the estimation of the crop edge. Secondly, two control laws determine the orders of the machine steering angle to guide them along the crop edge. The first one consists to generate a reference trajectory and the second one, which is calculated in the image space, can be used in very difficult situation like sloping ground. The result satisfies most situations and the precision of the system is better than 10 cm. Nevertheless, there are still very difficult situations in which it is impossible to correctly guide agricultural vehicles (sun direction at the end of the afternoon, shadow of the machine in the image, lack of vegetation). Future works will increase the environment perception using other sensors like laser telemeter and fuzzy based theory to combine laser and video information. We will also work on guidance-assistance system by CPDGPS and new control algorithms in 3D space.