In the field of optical engineering, the laser position control system has important role in many applications, such as measurement, communication, fabrication. Traditional methods to solve laser position control system often face the problems of insufficient generalization, such as configuration or singular solution. In this study we proposed a novel model- free reinforcement learning approach based Proximal Policy Optimization (PPO) for laser position control system. To control the position of laser, we develop an efficient representation of environmental inputs and outputs. Position error of Position Sensing Detector (PSD), and three kinds of distance parameters are applied our environmental parameters. To overcome the challenges associated with training in real worlds, we developed training environment in simulation. The simulation to evaluate performance of our approach, we perform several times of experiments in both simulated and real world system.

The fast steering mirror is now being used in industries beyond precision processing, such as space and defense. The piezoelectric fast steering mirror (PFSM), which utilizes a piezoelectric actuator, is particularly suitable for these industries as they often require devices like electro-optic devices to withstand external vibrations and impacts. While the PFSM has inherent high stiffness, its complex structure makes it difficult to control. To address this, an accurate dynamic model is necessary. In this paper, we derived a dynamic model for the PFSM using a two-inertial system model that takes into account its structural characteristics. This dynamic model consists of both a mechanical system model and an electrical system model. We measured the frequency response function from electrical input to mechanical output and compared it with the derived frequency response model to verify its accuracy. The derived model can be used not only for control design, but also for instrument design and interpretation.