In order to solve the problem of robot system control stability, based on the classical passive observer method, an improved passive observer was adopted to control the robot system.The system structure model of the robot was created, and the input signal force was added as the computational signal of the controller. Aiming at the problem of energy accumulation of classical passive observer, an improved new passive observer method was proposed. By using the window passive observer(WPO) and reset passive observer(RPO), the gain of PID controller was effectively reduced, and the energy consumption was reduced, so the instability of the system was faster detected. In MATLAB, the velocity, position and force feedback tracking of the robot system model in x and y directions were simulated and compared with the classical passive observer method.The results show that the speed of the robot system with the improved passive observer tends to be stable after a short overshoot; the position tracking curve is less jitter and smoother; the maximum feedback is about 33% lower than that of the classical passive observer method. The results show that the improved passive observer can improve the control stability of the robot system.