Abstract: A dynamic model based on generalized coordinates is proposed to solve the free-force control problem in drag teaching of 6-RUS parallel robot. Firstly, the dynamic friction model is introduced, and a dynamic model is established in the form of generalized coordinates; the actual output torque of the motor in a workspace position is obtained by the global area method, and the Savitzky-Golay (SG) algorithm is used to smooth the noise and analyze the influence of the mechanism's own noise on the torque data. Then, the expected output torque of the motor is constructed based on the dynamic equation of 6-RUS parallel robot, the error between the expected motor torque and the actual torque is compensated, and the motor torque is tracked in real time to achieve free-force control. Finally, experiments are carried out on the 6-RUS parallel robot platform. The results show that the errors between the predicted and actual output torques of the motor are less than 8.25% in different load conditions. In the same teaching process, the end torque required by the proposed method is only 2.2 N·m, and 3.4 N·m is required by the free-force control method based on the compensation of gravity and friction torques, which verifies the effectiveness of the proposed free-force control method. Comparing with the traditional free-force control method, the proposed method simplifies the dynamic modeling steps, improves the efficiency of constructing the expected motor torque, and solves the problem of inaccurate prediction of the expected motor torque in the stationary, starting, and reverse motion phases of the robot. The torque error is compensated in real time, so that a better drag effect can be obtained.