DeepServo: Deep learning-enhanced state feedback for robust servo system control

Abstract

Servo controllers are essential components in robotics, manufacturing, and various industrial applications. However, achieving fast and accurate reference tracking in servo systems remains challenging due to modelling uncertainties and external disturbances. In this paper, a hybrid control strategy is proposed that combines a Linear Quadratic Regulator (LQR) state-feedback controller with deep learning to address these challenges. The LQR controller utilises system state measurements to optimise the control input, while the integration of a deep neural network enhances accuracy and dynamic response by adapting to changing system conditions. This approach provides robust control performance, effectively mitigating the impact of uncertainties and disturbances on servo system behaviour. The proposed method was validated using AC servo motors, among the most common servo systems, though the approach is adaptable to other servo-like systems. Comparative evaluations are conducted against existing methods, including SIMC-SMC, 2DOF-IMC-SMC, 2DOF-IMC-PID, and SIMC-PD controllers, focusing on the angular position control of a servo motor. Simulation results demonstrate that the proposed controller outperforms these methods in terms of robustness, precision, and disturbance rejection. These findings highlight the potential of the proposed LQR-deep learning framework to significantly improve servo system performance across a wide range of applications.