AOSAT particle tracking software. Presented at Division for Planetary Sciences (DPS 47) 2015
Laboratory Spacecraft Data Processing and Instrument Autonomy: AOSAT as Testbed
Jack Lightholder, Erik Asphaug, Jekan Thangavelautham
Arizona State University
Recent advances in small spacecraft allow for their use as orbiting microgravity laboratories (e.g. Asphaug and Thangavelautham LPSC 2014) that will produce substantial amounts of data. Power, bandwidth and processing constraints impose limitations on the number of operations which can be performed on this data as well as the data volume the spacecraft can downlink. We show that instrument autonomy and machine learning techniques can intelligently conduct data reduction and downlink queueing to meet data storage and downlink limitations. As small spacecraft laboratory capabilities increase, we must find techniques to increase instrument autonomy and spacecraft scientific decision making. The Asteroid Origins Satellite (AOSAT) CubeSat centrifuge will act as a testbed for further proving these techniques. Lightweight algorithms, such as connected components analysis, centroid tracking, K-means clustering, edge detection, convex hull analysis and intelligent cropping routines can be coupled with the tradition packet compression routines to reduce data transfer per image as well as provide a first order filtering of what data is most relevant to downlink. This intelligent queueing provides timelier downlink of scientifically relevant data while reducing the amount of irrelevant downlinked data. Resulting algorithms allow for scientists to throttle the amount of data downlinked based on initial experimental results. The data downlink pipeline, prioritized for scientific relevance based on incorporated scientific objectives, can continue from the spacecraft until the data is no longer fruitful. Coupled with data compression and cropping strategies at the data packet level, bandwidth reductions exceeding 40% can be achieved while still downlinking data deemed to be most relevant in a double blind study between scientist and algorithm. Applications of this technology allow for the incorporation of instrumentation which produces significant data volumes on small spacecraft without comparable increases to power and bandwidth budgets.