Abstract: This study presents a solution for the problem of learning and controlling a mobile robot in the presence of fixed obstacle. The objective is to move the robot from an initial position (source) to a final position (target) without collision. Thus, we propose an approach based on potential fields principle, we define the target as an attractive pole (given as a vector directly calculated from the target position) and the obstacle as a repulsive pole (a vector derived by using fuzzy logic techniques). The linguistic rules, the linguistic variables, the membership functions... etc are the parameters to be determined for the fuzzy controller conception. A learning method based on gradient descent for the self tuning of these parameters is introduced. Therefore, it is necessary to have an expert person for moving the robot manually. During this operation of training, the robot moves and memorizes the data (inputs and outputs). This operation is used to find the controller parameters in order to reach the desired outputs for given inputs.