Optimal Sensor Tasking through Deep Reinforcement Learning for Space Situational Awareness
Abstract
Modern society relies heavily on space infrastructure for various technologies, including communication and navigation. Space assets are constantly threatened by collisions with debris, while satellite-debris and debris- on-debris collisions can result in a cascade effect, rendering space inaccessible for future generations. Even the smallest of debris objects such as paint flakes, traveling at 7 times the speed of a bullet, can incapacitate satellites through collision, rendering them useless. Presently, millions of debris objects bigger than paint flakes orbit Earth with a significant portion predicted to stay there for over a century. To protect our assetsand the space environment for future use, it is imperative to prevent collisions by tracking all such objects for comprehensive knowledge (past, present, and future) of their position. Currently, there are more objects than sensors to track them, and therefore, it is important to carefully select which objects to track through the use of sensor planning algorithms. Most existing sensor planning algorithms are overly simple, making planning decisions that only consider the short-term benefits for each observation. The focus of this work is the development of an intelligent system for sensor planning and coordination for tracking active satellites and debris objects. The challenge that this work seeks to address is the ability of such a system to “learn”from past experience and improve over time. The proposed approach is scalable to the whole catalog. The developedsystem will provide optimal tasking for sensors through a revolutionary artificial intelligence approach called Deep Reinforcement Learning.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 09, 2018
- Source ID
- FA95501810115
Entities
People
- Richard Linares
Organizations
- Air Force Office of Scientific Research
- Regents of the University of Minnesota
- United States Air Force