Solving Actuator Selection Problems with Q-learning

In modern large-scale radio telescopes, active reflector control technology is commonly employed. For example, the Five-hundred-meter Aperture Spherical Telescope (FAST), the largest single-dish radio telescope in the world, features a reflective surface composed of multiple sub-panels. Each sub-panel typically requires three micro-displacement actuators to provide support and facilitate rapid adjustments. These actuators enable precise control of the sub-panel’s pitch, yaw, and translation, ensuring the high-precision alignment of the overall reflective surface. This paper proposes a methodology for selecting the most reliable actuator from a range of available options. To address the optimal selection of actuators in active reflector control systems, the $\mathbf{Q}$ learning algorithm is employed. The selected actuator’s performance is validated through physical experiments conducted on an active reflective surface control module, which is integrated into the active reflective surface measurement and control digital twin experimental platform of the Joint Laboratory. The final actuator selection is subsequently applied to a physical system implementation. Compared with traditional selection methods, the proposed approach demonstrates improved selection efficiency, accelerates project completion, and significantly reduces costs associated with trial-and-error processes.