Onboard Artificial Intelligence and Machine Learning for Enhancing SmallSat Constellations
- Paper number
IAC-19,B6,IP,4,x52274
- Author
Mr. Christopher Heistand, United States, The John Hopkins University Applied Physics Laboratory
- Coauthor
Mrs. Amy Alford, United States, JHU Applied Physics Laboratory
- Coauthor
Mrs. Elizabeth Bathrick, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Mr. Dmitriy Bekker, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Dr. Joshua Broadwater, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Mr. Adam Byerly, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Dr. Christopher Gifford, United States, JHU Applied Physics Laboratory
- Coauthor
Dr. Musad Haque, United States, The John Hopkins University Applied Physics Laboratory
- Coauthor
Dr. Amy Haufler, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Dr. Karl Hibbitts, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Mr. Michael Malinowski, United States, JHU Applied Physics Laboratory
- Coauthor
Mr. Justin Thomas, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Ms. Rebecca Williams, United States, JHU Applied Physics Laboratory
- Coauthor
Ms. Kiley Yeakel, United States, Johns Hopkins University Applied Physics Laboratory
- Coauthor
Ms. Michelle Chen, United States, The John Hopkins University Applied Physics Laboratory
- Year
2019
- Abstract
Earth orbiting missions are being designed around constellations of small satellites and CubeSats with greater heterogenous sensor capability, data volume, and complexity (of joint) operations than ever before. In parallel, sensors continue to become smaller and require less power, yet generate a much larger data volume. With all this space-based capability, it is time to apply mature Artificial Intelligence (AI) from ground systems and distribute it across the heterogenous sensor constellations. A benefit of a distributed intelligent system is that data can be triaged onboard, leaving communications links open for the time critical data to be downlinked. Understanding capabilities of neighboring satellites enables autonomous cueing of relevant sensors and coordinated data acquisition amongst the satellites. This coordinated inter-satellite operation facilitates capturing time-dependent natural phenomena ranging from understanding weather to monitoring natural disasters that could otherwise be missed or require additional spacecraft resources such as extended observation times and recorded data volume. As with all mission designs, there are trades to be analyzed. What is the optimal distribution of AI? What algorithms can actually be implemented onboard the satellite? What is the best combination of human and machine teaming? Which decisions can be made without human intervention? The Johns Hopkins University Applied Physics Laboratory is developing a capability that enables \begin{enumerate} \item these trades to be evaluated, \item rapid prototyping of these concepts to demonstrate efficacy using an infrastructure that can easily be deployed onto future space avionics and ground systems, and \item a flexible prototype that facilitates integration and evaluation of different sensor types and machine learning algorithms. \end{enumerate}
- Abstract document
- Manuscript document
(absent)