In this paper, an obstacle avoidance problem for an autonomous mobile robot (AMR) is considered. The proposed control system mainly consists of a decision-making block, a tuning block and a danger- degree evaluation block. In the decision-making block, an artificial neural network is used to select the suitable avoidance action under the measured situation. The neural network is adjusted by genetic algorithm so that both the dynamic and the static danger-degrees are minimized during the transfer period of avoiding obstacles on the way to the goal. Both danger-degrees are obtained by the fixed fuzzy rules in the danger degree evaluation block based on the measured situation. Experimental results demonstrate that the proposed system is effective to generate the suitable avoidance action of the autonomous mobile robot for the obstacles.