Abstract: A feedforward multilayered neural network controller is proposed. This controller can learn the inverse of a plant off-line or on-line. It can "associate" the correct control information that is to be put into the plant with the feedback information from input and output of the plant as well as the future output desired. Two links of RM501 robot in realtime are controled by this neural controller in experiments. The experimental results show the controller is satisfactory and has practical value.