Parameter Identification of LuGre Friction Model for Ball Joint Motor based on DDPG Algorithm

Due to nonlinear disturbances like friction, the novel three degree of freedom ball joint motor may encounter issues, such as climbing and chattering, during trajectory tracking control process. For the ball joint motors, a parameter identification method based on reinforcement learning algorithm for the friction model is proposed in this paper. To identify LuGre friction model’s static parameters for each motor, actions, environments, and rewards of Deep Deterministic Policy Gradient (DDPG) are designed. A compensation mechanism leveraging the active disturbance rejection control strategy is designed as well. A speed-friction torque identification experiment is designed for three degree of freedom ball joint motor. It is results show the DDPG algorithm outperforms traditional neural networks and genetic algorithms in parameter identification accuracy. An active disturbance rejection controller with friction torque compensation is then applied, mitigating friction disturbances and enhancing tracking precision. The experimental and simulation results show the average error of the three axes is kept within 2%.