Observation Strategy Optimization for Distributed Telescope Arrays with Deep Reinforcement Learning

Abstract

Abstract Time-domain astronomy is an active research area now, which requires frequent observations of the whole sky to capture celestial objects with temporal variations. In the optical band, several telescopes in different locations could form a distributed telescope array to capture images of celestial objects continuously. However, there are millions of celestial objects to observe each night, and only limited telescopes could be used for observation. Besides, the observation capacity of these telescopes would be affected by different effects, such as the sky background or the seeing condition. It would be necessary to develop an algorithm to optimize the observation strategy of telescope arrays according to scientific requirements. In this paper, we propose a novel framework that includes a digital simulation environment and a deep reinforcement learning algorithm to optimize observation strategy of telescope arrays. Our framework could obtain effective observation strategies given predefined observation requirements and observation environment information. To test the performance of our algorithm, we simulate a scenario that uses distributed telescope arrays to observe space debris. Results show that our algorithm could obtain better results in both discovery and tracking of space debris. The framework proposed in this paper could be used as an effective strategy optimization framework for distributed telescope arrays, such as the Sitian project or the TIDO project.