In this thesis, a new navigation method is developed and implemented for wheeled mobile robots. The robot navigation problem has traditionally been decomposed into the path planning and path following. In contrast the proposed method unifies these two steps into a unified navigation method. The proposed navigation method uses two kinds of fields, potential and centripetal fields. To overcome non-holonomic constraint, the sliding mode approach is applied. The potential field makes the robot to avoid obstacles. The centripetal field makes the robot to move to the destination and does not disturb the potential field. So, the optimal centripetal field can be found by Evolutionary Programming. To apply the proposed navigation method to arbitrary polygon-map environment, the path generalization algorithm is proposed. This algorithm decomposes the path into sub-paths which are classified into the several general sub-paths. By this algorithm, the on-line navigation is possible. This thesis shows the feasibility and the effectiveness of the proposed method from the behavior obtained by different experimental conditions through the computer simulation and real experiments.