Abstract: In this study, we propose an improved robot joint model to carry out robot dynamics modeling and identification work. For this, we establish a robot dynamic model, analyze the joint structure of the robot, improve the joint model, and then use the input torque of a harmonic reducer to approximately estimate friction torque. We select the Fourier series as the excitation trajectory and optimize its parameters. Relevant data are collected and processed by controlling the movement of joints according to the obtained trajectory. We then identify and link the robot joint model parameters based on the weighted least squares method. The obtained parameters are verified using the joint predicted torque. Our experimental results show that the robot dynamics model based on the improved joint model has significantly improved accuracy.