Optimal control of wide field small aperture telescope arrays with reinforcement learning

In recent years, time domain astronomy has become an active research area. Thanks to its low cost and moderate observation ability, wide field small aperture telescopes are commonly used to observe celestial objects for time domain astronomy. We would use several wide field small aperture telescopes to form an array to observe celestial objects continuously. Because there are many celestial objects for telescope arrays to observe, such as obtaining positions or magnitudes of celestial objects or discovering new transients, it would be necessary to investigate an optimal control strategy to maximize their scientific outputs. To achieve this target, we need to make trade-offs between observations of different targets and define appropriate tasks for each telescope. In this paper, we propose a framework, which includes a simulator and a reinforcement learning based algorithm, to obtain optimal control strategy for wide field small aperture telescope arrays, according to predefined scientific requirements. Our method could achieve better performance than ordinary sky survey strategies and has good generalization ability after training.