Abstract: For an uncalibrated visual servoing system with dynamic uncertainties of a robot, a joint torque controller was designed based on robot dynamics by selecting several feature points on the end-effector, which made the selected feature points converge to their desired positions on the image plane. On the basis of a depth-independent image Jacobian matrix and a set of robot dynamics parameters properly selected, linear parametric models of the camera and robot dynamics were established. Then, the online adaptive updating laws the camera parameters and the robot dynamics parameters were designed, and the stability of the closed-loop system was analyzed. Finally, simulation experiments on a three joint manipulator show good performance of the control scheme.