Friction affects control accuracy of robot during precision assembly.When LuGre friction model is used for joint torque calculation,the robot joint friction force has periodic ripple errors.An improved LuGre friction model was proposed to address this problem,including a steady-state friction force represented by the LuGre friction model,and a velocity-dependent position-dependent term.The friction model was identified in steps,and the steady-state friction parameters were identified using the features of the LuGre friction model,and the position-dependent terms in the model were parametrically identified by the SVM multi-class classification algorithm,support vector regression (SVR) and least squares method to solve the system equations.The experimental results show that the error can be reduced by more than 50% when the robot is operated under different loads using the improved model and the identification method to calculate the joint friction torque.