Autonomous Sensor Tasking for Space Situational Awareness ​using Deep Reinforcement Learning

Abstract

The Joint Space Operations Center of the United States Strategic Command’s Joint Functional Component Command for Space is responsible for detecting, tracking, and identifying all artificial objects in the Earth’s orbit. Over 39,000 man-made objects are cataloged and over 16,000 objects are being tracked. It also tasks the Space Surveillance Network, a network of 30 space surveillance telescopes observing these space objects, which results in approximately 400,000 observations daily. There must be an autonomous system to manage resources generating the catalog of space objects and update information on these space objects for catalog maintenance. Maintaining the growing space object catalog will become more complex due to sensors ever increasing capability to detect a larger number of objects. This work will focus on the development and analysis of a physically-based machine learning algorithm for real-time inference of Space Objects energy parameters and states to predict the space objects orbits for the purpose of sensor tasking system to meet the high-level goals of Space Situational Awareness. By using a predictive algorithm, the system only needs to track periodically instead of continuously, which minimizes system utilization. Summarily, the system will need to be able to track existing objects already in the catalog. The sensors tasking problem will be devised as a Markov Decision Process and solved using Reinforcement Learning to design a deep neural network to generate the optimal actions for the sensors for tracking space objects.