Adaptive Sliding Mode Disturbance Observer and Deep Reinforcement Learning Based Motion Control for Micropositioners

Area of Science:

• Control Systems Engineering
• Robotics
• Artificial Intelligence

Background:

• High-precision electromechanical systems, like micropositioners, face challenges due to nonlinearity, external interference, and complex model identification.
• Existing control methods struggle to achieve the required robustness and precision for demanding micropositioning tasks.
• Reinforcement learning (RL) offers potential for optimal control but its application in micropositioning is limited.

Purpose of the Study:

• To develop a robust and precise motion control strategy for high-precision electromechanical systems.
• To address the limitations of current control methods in handling nonlinearity and external disturbances.
• To explore the application of deep reinforcement learning in micropositioning control.

Main Methods:

• A disturbance observer-based deep reinforcement learning control strategy was investigated.
• Deep deterministic policy gradient (DDPG) with an integral differential (ID) compensator was employed to reduce state error and improve transient response.
• An adaptive sliding mode disturbance observer (ASMDO) was proposed to mitigate lumped disturbances.

Main Results:

• The proposed control algorithm demonstrated high robustness and precise tracking performance.
• Simulations and experiments showed the micropositioner achieving tracking errors of less than 1 μm.
• The controller significantly improved accuracy and performance in challenging micropositioning scenarios.

Conclusions:

• The disturbance observer-based deep reinforcement learning control strategy offers a viable solution for high-precision motion control.
• The integration of DDPG with ID compensation and ASMDO effectively handles system nonlinearities and disturbances.
• The approach significantly enhances the accuracy and robustness of micropositioning systems.