posted on 2023-09-19, 04:08authored byzhang yang, yang xing, Yi Huang, Jun Chang, ze wu, jia song, Ziwen Duan
Stray light suppression constitutes a vital aspect in the development of opto-mechanical systems, but its complexity and the uncertainty surrounding scattered light necessitate intricate mathematical calculations and ample simulation iterations, along with extensive expertise and time. Consequently, researching stray light suppression in opto-mechanical systems becomes a time-consuming and challenging endeavor. To validate the feasibility of using reinforcement learning for stray light suppression, this study adopts a model-based deep reinforcement learning approach within a Monte Carlo ray-tracing environment to devise suppression strategies. The experimental results demonstrate that the model-based deep reinforcement learning method can propose effective stray light suppression measures tailored to various optical system configurations, resulting in significant improvements in suppression efficiency.
History
Funder Name
National Key Research and Development Program of China (2021YFC2202100); National Natural Science Foundation of China (62205027)