Michael Lim, MDA; Payam Mousavi, MDA; Jelena Sirovljevic, MDA; Huiwen You, MDA
Keywords: Space Situational Awareness, High Performance Computing, Onboard Processing, Artificial Intelligence, Low Power GPUs
Abstract:
Onboard processing provides the reduction in latency that is critical to Space Situational Awareness (SSA) applications. By receiving data directly from the sensor and processing it in real-time on board of spacecraft, this technology enables real-time processing and response. Currently, onboard processing is a nascent technology and the capabilities that exist are limited: they are highly-customized, one-off systems typically built for large spacecraft. This is poised to change drastically in the coming decade. Within that time period, onboard processing will transform from a research topic to an essential element of most space missions.
One of the key functionalities for onboard processing in SSA domain is object classification. Current state of the art object classification algorithms are based in Artificial Intelligence (AI) technologies. Due to prohibitively high computational needs for AI applications their deployment onboard spacecraft has not been possible to date. However, the rapid advance in AI-oriented computing hardware, especially graphics processing units (GPUs), has opened the door to AI in space. In particular, low size, weight, and power (SWaP) GPU devices have been developed that would be ideal for space-based processing.
MDA is currently investigating use of state-of-the-art off-the-shelf low power GPUs for deployment of AI applications essential for real-time object identification as part of the SSA domain. This work is motivated by recent increased focus within the AI community on operationalizing AI methodology. In this paper, we will discuss the motivation behind the research, technical details of the implementation, current results as well as the future goals of this research.
Date of Conference: September 15-18, 2020
Track: Machine Learning for SSA Applications