부산대학교 도서관

Spacecraft operating at great distances experience limited data bandwidth and high latency for communication with Earth. Data analysis algorithms that operate onboard, the spacecraft can perform detection and discovery of events of interest without human intervention. This capability serves to increase the quality and quantity of science data collected by the mission through data summarization, downlink prioritization, and adaptive instrument mode switching. However, before such technology can be adopted for use by a mission, it is necessary to characterize the required memory and computational resources. For operation in high-radiation environments, such as in orbit around the gas giants, a characterization of radiation tolerance is also important. In this article, we propose a framework to assess the resource and radiation profiles for machine learning algorithms in a simulated spacecraft computational environment. We apply this framework to several use cases designed for the Europa Clipper spacecraft, which plans to study Jupiter’s moon Europa. This approach can also benefit other remote deployments, such as the robotic exploration of hazardous environments on Earth.