Time-optimal trajectory planning algorithms have been widely adopted to minimize the motion time by exploiting the dynamics and joint allowable torques of a robotic manipulator. However, the actual joint torques may exceed the joint allowable torques because of modelling errors or disturbances in the control system. When the torque limit is added for actuator safety, the controller will have no margin to deal with modeling errors or disturbances, which may lead to large path tracking errors. An on-line trajectory time scaling method called path velocity controller can improve path tracking performance by modifying the path velocity when torque saturation occurs. However, the path velocity controller is based on a feedforward or computed torque controller, so the dynamic modelling errors will worsen the path tracking performance. In addition, the motion time may also be increased because the dynamic modelling errors could result in longer duration time of torque saturation. To further improve the path tracking performance of a path velocity controller, a path velocity controller with an on-line parameter estimate mechanism is proposed. The simulation results show that the proposed method can achieve a better path tracking performance and shorter motion time.