Distributed inferencing and federated learning, for distributed-edge-AI miniaturized satellite constellations
Abstract
Miniaturized satellites are capable of compute-intensive processing at the edge due to the shorter communication range and low altitude. Though orbital edge computing is in its infancy, edge computing shows promise for space mission optimisation, providing actionable insights in near-real time (NRT) and addressing the challenge of classical rate limits on the communication pipeline, while ensuring data privacy. This project seeks to address the first two of these knowledge gaps, and in the process establish the artificial intelligence test range as a research facility to support future basic and applied distributed intelligence research that would include cyber-resilient - cyber-secure distributed-intelligence space systems. In turn, the facility and the evolving body of research will be available to support both Australian and US developments, including the program at AFRL, and potential project arrangement(s) between AFRL and Defence Science and Technology Group (DSTG). The specific aims of the project are the development and demonstration of decentralized distributed-AI (federated and transfer learning) methods to promote intelligence sharing within the constellation without the reliance on the ground station (or server). The particular focus will be demonstration of the developed algorithms on the XView database with the integration of federated learning enabled object detection algorithms, the implementation of an edge-suitable astrodynamics-orbit prediction engine for each node, capable of tracking and determining the location of the satellites within a constellation to prompt the AI test range capabilities, and bringing the above aims together, determination of the performance of the algorithms under realistic orbital scenarios such that the distributed-intelligence network is dynamic in size and connectivity.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 16, 2024
- Source ID
- FA23862314099
Entities
People
- Bassel Al Homssi
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of New South Wales