This paper addresses the problem of mobile robot navigation in indoor cluttered environments. A new algorithm for both longitudinal and lateral real-time control of wheel-based mobile robots has been proposed. Its main characteristic is smooth and stable following of the on-line replanned path. Our control method is actually extension of the virtual vehicle method proposed by M. Egerstedt et al., which is based on the introduction of a look-ahead point on the path that serves as reference point for the control algorithm. While the original virtual vehicle method uses only error feedback for reference point movement along the path, our method uses also a feedforward component based on path curve characteristics between the robot and the reference point. In this way stable robot movement is achieved also in the presence of obstacles that are critical with respect to the path following error. Experimental verification is provided for a differential drive robot within an on-line global path planning framework.