Artificial Intelligence at the Edge, or Edge-AI, is a recent area of AI technology development that becomes attractive and necessary when users are confronted with connectivity limitations and data privacy issues. The space domain is arguably the ultimate edge and the application of Edge-AI to miniaturised space systems is a growing research strength for the UNSW Canberra Space team. Edge-AI is also applied to aspects of space operations such as Space Situational Awareness and Space Traffic Management.
Edge-AI research at UNSW Canberra Space focuses on the development of miniaturised satellites as edge devices, capable of performing complex tasks and analysis of data on orbit. This allows actionable information to be communicated directly and rapidly to the end user without the need to downlink large volumes of potentially sensitive data.
However, we take this further. Our research on distributed learning is designed to link the Edge-AI-enabled satellites into semi-autonomous intelligent satellite constellations. Such constellations could collectively perform complex missions with outcomes that significantly exceed that of the sum of the individual units.
At the same time, we apply machine learning to the science of Space Situational Awareness and Space Traffic Management, including collision avoidance, space surveillance and formation flying and control.
The enhanced capabilities of intelligent satellite constellations will improve traditional applications and uncover novel uses of space systems and space-derived information. Most sectors and aspects of society depend on space – this will have a significant impact on each of them. In particular, intelligent space systems offer the opportunity for the development and rapid (near real-time) delivery of information ready for decision-making, directly to the end user.
Our AI-for-space capability includes the development of advanced machine learning algorithms and approaches and their application to advanced edge-AI accelerator devices. This includes graphics processing units (GPUs) and vision processing units (VPUs). Algorithms and approaches that offer the opportunity for deployment to miniaturised satellite constellations with multi-modal sensors are a key focus.
Our capabilities were demonstrated by exceptional performance in recent global AI-for-space competitions. We placed seventh in Airbus' 2019 Satellite Pose Estimation Challenge and third in ESA's 2020 Collision Avoidance Challenge. We achieved these results despite employing AI approaches deliberately designed to be generalisable and scalable, rather than tuned for the constraints of specific competitions.
We developed and are demonstrating edge-AI hardware on our M2 Pathfinder and M2 satellites, including powerful in-house onboard computing and processing capabilities combining central processing unit (CPU) and field-programmable gate array (FPGA) technologies. We also developed a high-resolution imager/CPU/FPGA/GPU capability for on-board image processing that will be demonstrated on M2.
Most importantly we are one team under one roof. We combine space engineering and science, including space mission development, with research in the application of AI to space. This is enormously beneficial to our efforts in developing intelligent space system technologies.
Our current partners in this effort include Frontier Development Lab (in particular, FDL AUSNZ) and the SmartSat Cooperative Research Centre’s Artificial Intelligence in Space Research Network.