On Tuesday September 11th, seven technical short courses were taught by highly regarded industry experts in a variety of topics in the context of Space Situational Awareness. Over 150 AMOS and EMER-GEN attendees participated in the short courses which provide opportunities for working professionals to upgrade their technical job skills and remain abreast of recent developments in their respective fields. They are also an excellent opportunity for personalized instruction.
Over the past year, the field of Machine Learning (ML) has experienced incredible improvements in the applicability and accuracy of its techniques. ML software can predict threats and synthesize data, making it sought after for the rapid automation of predictive analytics. In cybersecurity terms, ML provides the fastest way to identify new attacks, draw statistical inferences and push that information to endpoint security platforms.
Operational Analytics: Demystifying Machine Learning
Presented by: Joseph Coughlin, Senior Aerospace Engineer, L3 Defense Solutions; Rohit Mital, Chief Technology Officer; Stinger Ghaffarian Technologies.
The course explained that operators and analysts are being overwhelmed with the amount of data available from both existing and new classes of sensors. The magnitude of data becomes too great to analyze by conventional means. ML has often been proposed as a solution to the “big data” problem which enables analysts to evaluate and determine courses of action based on the information.
Coughlin explained, “Our course presented an overview of current technologies and software and hardware architectures for aspiring or current users to utilize Operational Analytics and ML in their exploitation of SSA data. We also discussed building future processing architectures. The different approaches and applications of ML such as supervised learning, unsupervised learning, clustering, and deep learning, were presented with practical examples using SSA data.”
Coughlin added, “ML for SSA applications shows great potential to derive useful results from large amounts of data while saving time and money. Optical sensors are producing a significant number of observations per day and only a fraction of the information contained in the data is being extracted and exploited for SSA. ML algorithms can determine if a satellite is rotating, if it has maneuvered, is visible to a sensor, what type of satellite it is, and many more applications. Recent work in neural networks has shown how to predict maneuvers, detect patterns of life, and classify satellite types. A plethora of ML algorithms exist which can be used to solve SSA problems. However, choosing the right algorithm, features, target variables, and other algorithm parameters can be extremely difficult.”
Wes Faber, L3 Applied Defense Solution and course participant said, “The major take-away from the course was that ML is an excellent tool for predictive analytics and SSA in general. However, the authors did a good job describing its place within predictive analytics, when it is usable and when not, hence the demystifying of ML.”
Machine Learning for SSA
Presented by: Kyle Pula, Research Scientist, CACI; Richard Linares, Assistant Professor, MIT; and Roberto Furfaro, Associate Professor, University of Arizona
Richard Linares explains Machine Learning for SSA in the technical short course
This short course surveyed recent advances in machine learning and associated applications for SSA. The first portion of the course covered a broad overview of modern machine learning techniques, with an emphasis on those areas that seem most directly relevant to SSA. The second portion of the course examined a set of case studies of the techniques being applied to real SSA problems, including code examples in MATLAB (neural networks toolbox/statistics and machine learning toolbox) and Python.
Linares explained, “Over the past decade, the field of ML has experienced incredible improvements in the applicability and accuracy of its techniques. These advances present huge opportunities for the SSA community as it faces ever-increasing scope, sensing modalities, and data volumes. We created the course to highlight the opportunities for the SSA community and to provide a survey of recent advancements in ML.
Linares added, “The deep-learning portion of the course included a review of the major breakthroughs in the field, a theoretical outline of deep learning using neural networks, and examples of methods for designing deep neural networks. The interest in this course was very high, and I believe teaching the course next year, and updating it with progress in this area, would be well received by the technical community.”
Wes Faber took this course also and said, “This technical course paired well with the more conceptual description provided in the first short course ‘Demystifying ML’. After supervised and unsupervised learning, we dove into Deep Learning and approached problems that might cause poor performance, then touched on Extreme Learning Machines. I hope to see the rest of the advanced topics at next year’s AMOS.”
Statistical Orbit Determination for Space Surveillance and Tracking
Presented by: Moriba Jah, University of Texas at Austin
Moriba Jah provided a foundation for SSA and STM in his technical short course at AMOS.
Given the new space race driven mostly by commercial actors, we have a rapidly growing number of Resident Space Objects (RSOs) in Earth orbit. STM has become a critical problem to be solved. Part of what is needed is the ability to detect, track, identify, and characterize RSOs. This short course focused on providing the student with an overall understanding of the various components of statistical orbit determination within the context of SSA.
“My short course was attended by approximately 50 people with a diverse set of skills and experiences,” said Jah. “The students seemed to enjoy the class and said that it was dynamic and refreshing and brought everything together regarding estimating and predicting the locations of objects in space.
The topic is foundational for SSA and STM.” Jah continued. “This class helped those involved in space and those involved in space policy-making to better understand the nuances and difficulties and challenges involved in space object tracking and identification. This course should be a requirement for participating in this space community and most definitely should be a part of every AMOS conference from now on.”
2018 Conjunction Assessment and Risk Analysis
Presented by: Francois Laporte and Monique Moury, CNES; Matt Hejduk and Lauri Newman, NASA Goddard Space Flight Center
The threat of satellite collisions has become an increasing concern to the space-faring community both through an increasing mission risk due to a more congested space environment and through wider community awareness of the problem.
Newman explained, “NASA and CNES presented this short course to explain conjunction assessment theory and implementation, and to discuss how conjunction assessment fits as part of the big picture of the space environment. The space environment has been rapidly evolving and is expected to continue to do so in the coming years.”
Newman added, “As the new Air Force Space Fence radar comes online to detect smaller objects than currently possible, many more close approaches between space objects are expected to be predicted. As space becomes more congested and the workload increases, the concept of STM is coming to the forefront. STM is a focus of the National Space Council and the new Space Policy Directive (SPD-3). Although the implementation of STM is still in its infancy, international operators are diligently working towards improvements in collaboration.”
Observing and Characterizing Space Debris
Presented by: Thomas Schildknecht, Astronomisches Institut Universitat Bern in Switzerland
Schildknecht’s course provided a general introduction to the space debris problem, gave an overview on current space debris research activities to detect and characterize space debris, a summary of the efforts to model the future space debris population, and a description of international efforts to protect and remediate the space environment.
Schildknecht explained, “With space debris growing, this course addressed the different efforts of how to recognize, detect, avoid and overcome the challenges in managing the space environment. Space debris is an issue that concerns all nations in space and is a truly global challenge requiring international guidelines. With a continuing growth of nations operating in space, it is important to rapidly respond to newly-detected objects to avoid collisions.”
Space Debris Risk Assessment and Mitigation Analysis
Presented by: Tim Flohrer, Space Debris Analyst, SST Segment Co-Manager, ESA/ESOC Space Debris Office; Benjamin Bastida Virgili, Space Debris Engineer, ESA/ESOC Space Debris Office; both from Darmstadt, Germany.
The objective of this course was to provide an elementary introduction to ESA’s Debris Risk Assessment and Mitigation Analysis (DRAMA) tool suite. It enabled the participants to perform the analyses required to verify the compliance of mission scenarios with space debris mitigation requirements.
Flohrer said, “ESA’s space debris mitigation policy includes preventing uncontrolled growth of abandoned spacecraft, and orbital collisions. We hope that through disposal maneuvers we can prevent long-time debris residence in the protected regions and limit casualty risk to human population due to controlled or uncontrolled re-entry of space systems.”
Introduction to Theory and Application of Multi-Objective Optimization using Genetic Algorithms (GAs)
Presented by: Triet Tran, Cornerstone Consulting
Most real-world optimization problems are multi-dimensional in both the search and the objective space. It is very natural to employ the parallel-processing power of the Graphical Processing Unit to improve the execution speed of GAs.
Tran said, “This course presented the key concepts of genetic algorithm as applied to multi-objective optimization. The algorithm was explained in detail including the concepts of multi-dimensional search space and multi-dimensional objective space, with hands-on exercises to provide the students with a feel for how a GA works”.
Tran explained, “Flight dynamics software development often depends on multi-objective optimization using a GA. The full life-cycle development of ground satellite control systems, and estimation theory including Kalman Filter and Batch Least Square filters, are key to maneuver planning and station-keeping for satellites in an orbital regime. We need continued research and development of multi-objective resource optimization using genetic algorithm, flight dynamics maneuver planning, and GPS orbit and attitude determination.”
The AMOS Conference Organizing Committee accepts proposals for Short Courses when they issue the call for Papers late December.