Optimal Sensor Tasking through Deep Reinforcement Learning for Space Situational Awareness

Abstract

Recent advancements in Deep Reinforcement Learning (DRL) have demonstrated groundbreaking results across a number of domains. In particular, DRL has been used to successfully develop an artificial intelligence approach that can defeat expert human GO players. The generality of DRL and the groundbreaking results motivate the exploration of these approaches for SSA applications. Deep learning approaches mimic the function of the human brain by learning nonlinear hierarchical features directly from data where each hierarchy builds in abstraction. In an end-to-end fashion, DRL approaches can be used to process data directly to learn a control policy from training data. Under this work, DRL approaches are developed for space situational awareness (SSA) applications. Our research is divided into five interrelated research areas; development of DRL algorithms for SSA sensor tasking application, usage of deep learning techniques for space object (SO) classification and characterization, development of optimal constellation architecture for improved Low Earth Orbit's orbital capacity, development of DRL techniques for SO intent identification and anomalies detection, and development of Koppman Operator theories for space applications.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 11, 2022
Accession Number
AD1230931

Entities

People

  • Demoz Gebre-egziabher

Organizations

  • Regents of the University of Minnesota

Tags

Fields of Study

  • Computer science

Readers

  • Aerospace Engineering.
  • Distributed Systems and Data Platform Development
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Neural Networks
  • Space
  • Space - Space Objects