Scanning control of atomic force microscope based on deep reinforcement learning

• Published in: 2021 International Conference on Artificial Intelligence and Electromechanical Automation (AIEA) pp. 19 - 22
• Main Authors: Lv, Hongfu, Xu, Hongmei, Wang, Lequan, Li, Hongyang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.05.2021
• Subjects: Adaptation models; Adaptive systems; atomic force microscope; Atomic force microscopy; deep reinforcement learning; Force; Nanometer measurement; parameter tuning; Process control; Shape; Training
• DOI: 10.1109/AIEA53260.2021.00012

Atomic force microscope can use the force between atoms to scan the morphology of samples at the micro-nano scale. However, the control problem of atomic force microscope faces the problems of complex control environment and high precision requirements. Therefore, an adaptive PID controller that introduces deep reinforcement learning technology is proposed. The control problem of the atomic force microscope is described as a Markov decision process, with the DDPG algorithm framework as the main body, and the reward function is designed according to the actual control requirements. DDPG algorithm takes error as observation input, PID parameter as action output, realizes adaptive controller design, and obtains the control parameters that meet the requirements after the training is completed the experimental results show that the controller can meet the control requirements during the learning process, improve the control accuracy of the atomic force microscope system, and improve the imaging quality. The research results can provide references for researchers in the same field.