A simulation framework for telescope array and its application in distributed reinforcement learning-based scheduling of telescope arrays

Abstract

Time-domain astronomy necessitates continuous observation of celestial objects across the entire sky, with specific observation depth and cadence requirements. Telescope arrays, comprised of numerous wide-field optical telescopes, have emerged as a novel observational instrument for time-domain astronomy. However, the observation capabilities of ground-based optical telescope arrays are often constrained by various dynamic factors such as clouds, satellites, and sky background. To meet the observation requirements, active scheduling of telescopes within a telescope array is essential, leveraging telemetry data from the environment. However, due to the complexity and cost of telescope arrays, it is impractical to directly design or test algorithms using a physical telescope array. In this paper, we propose a framework for simulating telescope arrays that incorporates the majority of real observation effects. Building upon this framework, we further introduce a scheduler based on a distributed reinforcement learning framework to optimize the observation strategy of telescopes within the array. The scheduler is trained and evaluated using the simulator. Results demonstrate that the distributed control framework-based scheduler enhances the observation efficiency of the telescope array.