Efficient Object Maneuver Characterization for Space Situational Awareness
Efficient Object Maneuver Characterization for Space Situational Awareness 32 nd Space Symposium Technical Track April 12, 2016 Presented by: Charlotte Shabarekh Jordan-Kent Bryant, Mike Garbus, Gene Keselman, Jason Baldwin, Brian Engberg © 2016 Aptima, Inc.
Space is Crowded Cosmos-Iridium Crash Cosmos 2421 Fengyun-1 C Congested environment leads to increased threat of conjunctions © 2016 Aptima, Inc. 2
Scaling the SSA tradecraft § Detect-Track-Characterize-Catalog tradecraft needs to be scaled to handle the increased number of objects and debris – Where will an object be in the future? – What is it’s intent? – What is it’s relationship to other objects? § Activity Based Intelligence Can you predict – Incorporate context to infer this activity… …from this event? what the object is doing instead of where it is – Activities consist of sequences of events that are predictable Perform analysis around activity, not around object © 2016 Aptima, Inc. 3
Patterns of Life (Po. L) § Po. L are predictable, repeated behaviors – “Normal” or “Baseline” behavior § Po. L are constrained by environmental factors and physical properties § Po. L vary by location, time and object Maritime Po. L for Strait of Gibraltar and Port of Gibraltar 2014 MDA shares characteristics with SSA © 2016 Aptima, Inc. 4
Passenger Ferry Po. L Additional Passenger Ferries Class Objects Ammam : Levante Nova Star Cuidad de Malaga Bahama Mama Po. L enable prediction of future states for classes of objects © 2016 Aptima, Inc. 5
Passenger Ferry Po. L Deviation from Po. L begins Actual Path Expected Path Tanger Express deviated from Passenger Ferry Po. L by making a stop at Gibraltar Port Deviations aren’t always a threat, but should be flagged for MDA/SSA operators assessment © 2016 Aptima, Inc. 6
Patterns of Life in GEO § Same basic idea as Po. L in maritime domain, but different features, terrain constraints and patterns § Po. L in GEO are highly regulated by physics – Po. L have less variability but fewer features § Station-keeping maneuvers are repeatable and predictable for a satellite, time and location Galaxy 15 Po. L Station-Keeping Maneuvers Sept 2014 - Mar 2015 © 2016 Aptima, Inc. 7
Galaxy 15 Historical Events 2005 Galaxy 15 Launch April 5, 2010 Galaxy 15 stops responding to commands July 30, 2010 Galaxy 15 passes Galaxy 14 July 12 -13, 2010 Galaxy 15 passes Galaxy 13 /Horizons 1 May 31 -June 2, 2010 Galaxy 15 passes AMC 11 within half degree December 23, 2010 Galaxy 15 recovered October 20 - 25, 2010 Galaxy 15 passes Anik F 2 August 8, 2010 Galaxy 15 passes GCI Sat December 9, 2010 Galaxy 15 causes interference of NOAAPORT Timeline not to scale © 2016 Aptima, Inc. 8
Galaxy 15 Anomalous Period April 5, 2010 Galaxy 15 drifts from 133°W April 4, 2011 Galaxy 15 returned to 133°W © 2016 Aptima, Inc. December 23, 2010 Galaxy 15 recovered near 98°W January 15, 2011 Galaxy 15 temporarily relocated to 93°W 9
Galaxy 15 Po. L Research Objective: Using just astrometric data, can we automatically learn Galaxy’s Po. L for each year between 2011 -2015? § Can we predict the next maneuver? § Can we determine when there is a deviation from a Po. L? Deviation from Normal Galaxy 15 Po. L Repositioned to 93° W Resumes normal Po. L at 133° W Begins travel from 93° W to 133° W © 2016 Aptima, Inc. 10
Probabilistic Approach Interval Similarity Model: § Probabilistically predict when the next maneuver will occur based on previously observed maneuvers § Uses Bayesian approach to calculate the likelihood that any interval between maneuvers is part of a repeating pattern § Key Differentiators – Unsupervised learning of patterns to detect previously unseen patterns – Learns patterns from relatively small amounts of observed data (shows high probabilities after three maneuvers from a given pattern) – Re-estimates patterns as new data becomes available for dynamic Po. L Consistent Station-Keeping Maneuvers Pattern Ends New Pattern Emerges Interval similarity matrix, representing strength of relationship between maneuvers Probabilistic approach allows greater variability in Po. L © 2016 Aptima, Inc. 11
Findings § In 2012 -2015 data, we were able to accurately predict maneuvers (25 x more accurate than Poisson baseline) § In 2011 data, we were able to quickly detect deviation from expected behavior which is critical for TWA Maneuver Predicted at day 208 Correct maneuver prediction on 2012 data which had regular Po. L Maneuver Predicted at day 252 Incorrect maneuver prediction on 2011 data which had irregular Po. L Key Take-Away: Rapid detection of changes to expected Po. L © 2016 Aptima, Inc. 12
Past and Future Work § Past Work – Started conversation on applying Po. L analysis and ABI to SSA with the space community – Early results demonstrated ability to predict maneuvers and flag deviations from expected behaviors – AFRL/RV has sponsored research as early stage of a threat warning and assessment (TWA) system § Future Work – Envision relevance to JSp. OC, JICSp. OC, AFSPC – Technology development for Maneuver prediction to inform UCT correlation and data association tasks – Left-of-event prediction of long-term, multi-object activities Phase II will engage new partners to advance concept © 2016 Aptima, Inc. 13
Charlotte Shabarekh Analytics Division Director Aptima, Inc. cshabarekh@aptima. com Direct: 781 -496 -2465 Upcoming Events • Space Symposium (4/12) • Space Control Conference (5/3 -5/4) • GEOINT Symposium (5/15 -5/16) • Computer Vision and Pattern Recognition (CVPR) Conference (6/26) Aptima, Inc. | www. aptima. com 12 Gill Street, Suite 1400 Woburn, MA 01801 © 2016 Aptima, Inc.
- Slides: 14