PLOT

Recover. Replay. Compare. Approve the next launch.

PLOT is a debrief desk for an unmanned collaborative mission. The public theater is Skua Strait, a Dead Index reference theater. It is not built from operational tracks.

01

Admit the packs

A-GRA-shaped XML is the baseline. A near-A-GRA log with epoch timestamps and feet, and a platform JSON with no ICD, still parse. Every canonical field is marked present, mapped, guessed, or missing, and the pack gets a normalization score. Three shapes of the same sortie reconstruct to one mission hash.

02

Scrub the sortie

The timeline is the product. Play, pause, or jump to an event. Models move. Tasked routes stay as thin lines.

03

Keep two red readings

If the evidence does not support one conclusion, both stay on the chart as ghosts. Each reading lists its supporting and contrary observations, checked on the pack, and its confidence is the formula over those counts. Open “why this number” to see it.

04

Change the mark

Drag a blue or red unit at the current time. The other side updates from that mark forward. The change is a diamond on the timeline.

05

Compare the next launches

Three courses of action. Two are immediate station and pairing changes; retraining is labeled later and scored as no change to the next launch. Two thousand sampled worlds draw the red reading, jitter poses and ranges, and score every course in the same world. You read intervals, a probability of being best, and a probability that each course improves on doing nothing.

06

Select the algorithms

The desk carries its own algorithm selection matrix: for ingest, blue, red, and re-planning, the candidate families scored on data need, explainability, laptop runtime, standalone transfer, and maturity, with the selected family and the reason. The works cited are listed beneath it.

07

Stamp a next plan

Approve stops the deck clock. Export writes a brief and an A-GRA-shaped pack that carry the course, the reading, the intervals, the evidence, and the normalization report. Reject keeps the clock running.

Select

The algorithms, chosen in the open.

One family is selected per area and the desk runs it. The rubric is on each column head. Nothing below is a model output; the desk only ever computes what these rows describe.

Select

Algorithm selection matrix

Scored 0–2 on each criterion: ○ 0 · ◐ 1 · ● 2. Hover a column head for the rubric. One family is selected per area; a paired family runs beside it; later work is named, not sold as the stop-gap.

Ingest and normalization

Heterogeneous, partly unstructured debrief artifacts (Q&A: a Phase I research task).

How does a pack that is not A-GRA become one canonical record without hiding the guesses?

FamilyData need at squadron scaleExplainabilityLaptop runtime, no networkStandalone transferMaturityΣCites

Declared schema detection + field ledgerselected

Detect the shape, map each canonical field as present / mapped / guessed / missing, apply declared unit and time conversions, score the record.

●●●●●10[1] [2] [3] [4] [5]

Fixed per-platform ICD adapters

One hand-written parser per interface control document. A-GRA as the only shape.

●●●◐●9[4] [5]

Learned schema matching

Instance- and name-similarity matchers propose field alignments from examples.

◐◐●◐◐6[1] [6]

Language-model field mapping

Prompt a model to name the meaning of each column.

●○○○○2[7] [8]

Selected · Declared schema detection with a field ledger scores full marks and, unlike fixed adapters, does not fail on the non-compliant pack the Q&A promised. Every guess is a row on the desk. A language model may describe a guess; it may not make one silently.

On the desk · Three shapes of the same sortie reconstruct to one mission hash; the Packs panel shows the field map and a normalization score per shape.

Blue Force optimization

Parameters, autonomy behaviors, task allocation, routing, tactics; expected benefit, survivability, constraints, confidence.

What can change on the deck cycle, and what has to wait for retraining?

FamilyData need at squadron scaleExplainabilityLaptop runtime, no networkStandalone transferMaturityΣCites

Geometry scoring from tracks and class cardsselected

Survivability and effect from poses at the mark; a course is a station shift and a pairing flag; immediate lever is parameter and behavior selection.

●●●●●10[9] [10] [11]

Consensus / auction task allocationpaired

Decentralized bundle auctions re-task the remaining aircraft under range and timing constraints.

●◐●●●9[9] [12] [13] [14]

Multi-objective evolutionary search

NSGA-II over station and parameter vectors to trace a survivability–effect front.

●◐◐●●8[15]

Supervised outcome prediction

Random forests or boosted trees on sortie features predicting loss and effect.

○◐●●●7[16] [17] [10]

Policy retraining (reinforcement learning)later

Retrain the autonomy behavior against the observed failure in simulation.

○○○◐◐2[18]

Selected · Geometry scoring from tracks and class cards is the only family that answers in the deck cycle with numbers a commander can check. Decentralized task assignment (the topic's reference 6) is paired for Phase II re-tasking. Supervised prediction and policy retraining need sortie counts a squadron does not have and are named as later work.

On the desk · Each course states benefit, survivability implication, constraint, and confidence, all computed from the same maneuver.

Red Force modeling

Adversary tactics, capabilities, likely course of action; state uncertainty and offer alternatives when evidence does not support one conclusion.

How does a reading get a number a commander can trace, and when do two readings stay on the desk?

FamilyData need at squadron scaleExplainabilityLaptop runtime, no networkStandalone transferMaturityΣCites

Competing hypotheses with Beta-binomial evidence countsselected

Each reading lists supporting and contrary observations as predicates over the pack; confidence is the posterior mean under a Beta(2,2) prior.

●●●●●10[19] [20]

Bayesian networks over adversary intentpaired

Directed graph from observed behaviors to threat profiles; posterior by exact inference.

◐●●●●9[21] [22]

Dempster–Shafer belief functions

Belief and plausibility intervals over the set of readings.

●◐●●●9[23]

Multiple-hypothesis tracking

Track-to-measurement association hypotheses with pruning.

◐◐●●●8[24] [25] [26]

POMDP adversary modelslater

Adversary as a partially observed decision process; belief updates per event.

◐◐◐●●7[27] [22]

Selected · Competing hypotheses with evidence counts is the one family whose confidence is a formula over listed observations. Bayesian networks are paired for Phase II once SMEs give the structure. Multiple-hypothesis tracking needs measurement density the debrief pack does not carry; POMDP adversary models are later work.

On the desk · Two readings per story, each with supporting and contrary observations and (F + 2) / (F + A + 4) printed beside the percentage.

Autonomous re-planning

Generate and compare follow-on plans; multiple feasible courses with confidence, tradeoffs, and traceable rationale for operator approval.

How do three courses get intervals and a probability of being best instead of three point scores?

FamilyData need at squadron scaleExplainabilityLaptop runtime, no networkStandalone transferMaturityΣCites

Monte Carlo course evaluationselected

Seeded worlds draw the reading, red poses, and ranges; every course scored in the same world; p10 / p50 / p90, P(best), P(improve).

●●●●●10[28] [29] [22]

Weighted multi-criteria comparisonpaired

Declared weights on survivability and effect rank the courses; the weights are pack data, shown on the desk.

●●●●●10[30] [31]

Global sensitivity analysis

Variance-based indices on the pack weights and noise settings.

●●◐●●9[32]

Monte Carlo tree search over plan brancheslater

Search station and tasking branches against the sampled red picture.

●◐◐●●8[33]

Mixed-integer routing and assignmentlater

Exact optimization of routes and pairings under range, fuel, and rules-of-engagement constraints.

●◐◐◐●7[34]

Selected · Monte Carlo evaluation with common random numbers turns the same geometry function into distributions, a probability of being best, and overlap flags, in milliseconds. The declared weighted utility inside it is the multi-criteria layer. Tree search and mixed-integer optimization are later work for Phase II plan generation inside a Navy simulation.

On the desk · Dot-and-whisker intervals per course, P(best) that sums to 1, P(improve) as the course confidence, and one mcHash per run.

Works cited

  1. [1]Rahm, E., and Bernstein, P. A. “A survey of approaches to automatic schema matching.” The VLDB Journal 10(4), 2001.
  2. [2]Rahm, E., and Do, H. H. “Data Cleaning: Problems and Current Approaches.” IEEE Data Engineering Bulletin 23(4), 2000.
  3. [3]Doan, A., Halevy, A., and Ives, Z. Principles of Data Integration. Morgan Kaufmann, 2012.
  4. [4]Autonomy Government Reference Architecture (A-GRA), standard repository. gitlab.com/open-arsenal/a-gra/standardtopic reference
  5. [5]Agile Mission Systems Government Reference Architecture (AMS-GRA), standard repository. gitlab.com/open-arsenal/ams-gra/standardtopic reference
  6. [6]Bishop, C. M. Pattern Recognition and Machine Learning. Springer, 2006.
  7. [7]Vaswani, A., et al. “Attention Is All You Need.” Advances in Neural Information Processing Systems 30, 2017.
  8. [8]Rudin, C. “Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead.” Nature Machine Intelligence 1, 2019.
  9. [9]Alighanbari, M., and How, J. P. “Decentralized task assignment for unmanned aerial vehicles.” Proc. 44th IEEE Conference on Decision and Control, 2005.topic reference
  10. [10]Kelleher, J. D., Mac Namee, B., and D’Arcy, A. Fundamentals of Machine Learning for Predictive Data Analytics, 2nd ed. MIT Press, 2020.topic reference
  11. [11]Rizk, Y., Awad, M., and Tunstel, E. W. “Cooperative Heterogeneous Multi-Robot Systems: A Survey.” ACM Computing Surveys 52(2), 2019.topic reference
  12. [12]Choi, H.-L., Brunet, L., and How, J. P. “Consensus-Based Decentralized Auctions for Robust Task Allocation.” IEEE Transactions on Robotics 25(4), 2009.
  13. [13]Gerkey, B. P., and Matarić, M. J. “A Formal Analysis and Taxonomy of Task Allocation in Multi-Robot Systems.” International Journal of Robotics Research 23(9), 2004.
  14. [14]Dias, M. B., Zlot, R., Kalra, N., and Stentz, A. “Market-Based Multirobot Coordination: A Survey and Analysis.” Proceedings of the IEEE 94(7), 2006.
  15. [15]Deb, K., Pratap, A., Agarwal, S., and Meyarivan, T. “A fast and elitist multiobjective genetic algorithm: NSGA-II.” IEEE Transactions on Evolutionary Computation 6(2), 2002.
  16. [16]Breiman, L. “Random Forests.” Machine Learning 45(1), 2001.
  17. [17]Hastie, T., Tibshirani, R., and Friedman, J. The Elements of Statistical Learning, 2nd ed. Springer, 2009.
  18. [18]Sutton, R. S., and Barto, A. G. Reinforcement Learning: An Introduction, 2nd ed. MIT Press, 2018.
  19. [19]Heuer, R. J. Psychology of Intelligence Analysis. Center for the Study of Intelligence, Central Intelligence Agency, 1999. (Analysis of Competing Hypotheses.)
  20. [20]Gelman, A., Carlin, J. B., Stern, H. S., Dunson, D. B., Vehtari, A., and Rubin, D. B. Bayesian Data Analysis, 3rd ed. CRC Press, 2013.
  21. [21]Pearl, J. Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. Morgan Kaufmann, 1988.
  22. [22]Kochenderfer, M. J. Decision Making Under Uncertainty: Theory and Application. MIT Press, 2015.
  23. [23]Shafer, G. A Mathematical Theory of Evidence. Princeton University Press, 1976.
  24. [24]Reid, D. B. “An algorithm for tracking multiple targets.” IEEE Transactions on Automatic Control 24(6), 1979.
  25. [25]Blackman, S. S. “Multiple hypothesis tracking for multiple target tracking.” IEEE Aerospace and Electronic Systems Magazine 19(1), 2004.
  26. [26]Bar-Shalom, Y., Li, X. R., and Kirubarajan, T. Estimation with Applications to Tracking and Navigation. Wiley, 2001.
  27. [27]Kaelbling, L. P., Littman, M. L., and Cassandra, A. R. “Planning and acting in partially observable stochastic domains.” Artificial Intelligence 101(1–2), 1998.
  28. [28]Metropolis, N., and Ulam, S. “The Monte Carlo Method.” Journal of the American Statistical Association 44(247), 1949.
  29. [29]Kroese, D. P., Brereton, T., Taimre, T., and Botev, Z. I. “Why the Monte Carlo method is so important today.” WIREs Computational Statistics 6(6), 2014.
  30. [30]Keeney, R. L., and Raiffa, H. Decisions with Multiple Objectives: Preferences and Value Tradeoffs. Cambridge University Press, 1993.
  31. [31]Hwang, C.-L., and Yoon, K. Multiple Attribute Decision Making: Methods and Applications. Springer, 1981.
  32. [32]Saltelli, A., Ratto, M., Andres, T., Campolongo, F., Cariboni, J., Gatelli, D., Saisana, M., and Tarantola, S. Global Sensitivity Analysis: The Primer. Wiley, 2008.
  33. [33]Browne, C. B., et al. “A Survey of Monte Carlo Tree Search Methods.” IEEE Transactions on Computational Intelligence and AI in Games 4(1), 2012.
  34. [34]Bertsimas, D., and Tsitsiklis, J. N. Introduction to Linear Optimization. Athena Scientific, 1997.

Back to PLOT · Open Select on the desk · Launch demonstration