Skip to content

Latest commit

 

History

7 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

polybolos-denied-comms-c2

Status & recognition (factual)

OASW(SO/LIC) Accelerator Event - July 2026 (GoColosseum)
Submission status: Selected. Per the portal, Selected means the submission was found technically meritorious and is under evaluation/consideration. 1BCT / 82nd Airborne - Operation Epic Fury challenge (GoColosseum)
Submission status: Submitted. Industry challenge from 1st Brigade Combat Team (1BCT), 82nd Airborne Division (CENTCOM AOR, Operation Epic Fury) for rapidly deployable commercial solutions in contested tactical environments. Polybolos portal submission: Advanced Tactical UAS, Edge Compute, Network, Inferencing Solutions. AFRL engagement - April 2026
COMMAND HOTL materials were provided to Air Force Research Laboratory contacts at their request:

  • Col Christopher Rondeau (AFRL/RQ): after receiving the package, requested permission to share it with additional colleagues while building out this portfolio; permission granted (portfolio review / distribution interest).
  • Isaac Weintraub, PhD (Control Science Center, Air Warfare Directorate / RA): detailed technical Q&A on risk awareness, weaponeering, kinematics, and coordination. He wrote that the exchange helped him understand "the state of the art" and what can be gained through future partnerships, and indicated he would convey SBIR topic materials and/or partnering. That is attributed scientific and portfolio dialogue. Technology maturity
    Command HOTL is assessed at TRL 5 (lab / SITL / controlled demo / Lattice developer sandbox). Decision-C2 / human-on-the-loop authority lineage. Lattice
    Sandbox / interoperability evidence (including documented scale publish-ingest work) supports Lattice-edge integration feasibility. Not a production Lattice mesh claim. Independent of Anduril; samples are not Anduril products. Inquiries: mark.brown@polybolos.org
    CAGE: 1AVY9 · UEI: RUSHH9B2UQV3 · Polybolos Institute

Problem

Every C2 platform in the world fails the moment communications are denied.

Cloud-first architectures? Jammed. Network-dependent fusion? Silent. Latency-critical systems? Dead.

The Air Force has no solution for contested denial-of-comms environments.

Until now.

Solution

Deterministic, edge-native C2 fusion that operates without external connectivity.

No cloud dependency. No network latency. No black-box ML. Just math.

Performance

8,601 measurements across 5 combat scenarios

Scenario Measurements Latency Throughput Status
Straight-Line Crossing 1,328 3.91 µs 255,814 Hz ✓ Confirmed
Loitering Target 2,647 3.57 µs 280,000 Hz ✓ Confirmed
Evasive Maneuver 1,983 3.51 µs 284,895 Hz ✓ Confirmed
Fast Intercept 877 3.52 µs 283,880 Hz ✓ Confirmed
Low-Altitude Target 1,766 3.48 µs 287,365 Hz ✓ Confirmed

Aggregate: 3.60 µs/frame | 277,884 Hz | 5/5 tracks confirmed

Validation

  • ✓ Latency < 50ms (real-time budget): PASS (3.60 µs)
  • ✓ Throughput >= 20 Hz (edge-capable): PASS (277,884 Hz)
  • ✓ Determinism (bit-for-bit reproducible): PASS
  • ✓ Track confirmation (lifecycle): PASS (5/5 confirmed)

What Makes This Work in Denied Comms

  1. Edge-Native Architecture - All processing happens on-device. No cloud round-trips.
  2. Deterministic Math - Bayesian networks, not ML. Reproducible. Auditable. Reliable.
  3. Sub-Millisecond Latency - 3.6 µs per decision cycle. Fast enough for real-time authority.
  4. Full Audit Trail - Every fusion update logged. Every decision explainable.
  5. No External Dependencies - Works with local sensor streams only.

Architecture

Bayesian Networks

Transparent, acyclic decision propagation following Pearl (1988) framework. Every belief update is auditable and reproducible.

Kalman Filtering

  • Extended Kalman Filter (EKF) - Nonlinear state estimation
  • Unscented Kalman Filter (UKF) - High-nonlinearity variant
  • Adaptive noise covariance learning from residuals

Data Association

Hungarian algorithm for optimal track-to-measurement assignment. Mahalanobis distance gating prevents spurious associations.

Multi-Sensor Fusion

  • Radar (range, bearing, velocity, RCS)
  • RF (RSSI, direction finding, emitter ID)
  • Optical (pixel coordinates, confidence, thermal signature)

Dempster-Shafer combination rules resolve sensor contradictions.

Track Lifecycle

  • Tentative - New track, needs confirmation
  • Confirmed - 3+ consecutive measurements, fully tracked
  • Coasted - Lost measurement, predicting with last-known state
  • Deleted - Too many missed detections, track removed

Build

cmake -S . -B build
cmake --build build --config Release
ctest --test-dir build -C Release --output-on-failure
./build/benchmarks/Release/fusion_benchmark_big

Tests

19 unit tests, all passing:

  • Kalman filter initialization, prediction, update
  • Mahalanobis distance gating and data association
  • Track lifecycle (confirmation, coasting, deletion)
  • Scenario generation and ground-truth interpolation
  • Determinism verification (bit-for-bit reproducible across runs)
  • Integration test (full fusion loop with synthetic multi-sensor data)

Run:

ctest --test-dir build -C Release --output-on-failure

Benchmarks

Run the full scenario benchmark:

./build/benchmarks/Release/fusion_benchmark_big

Output shows latency, throughput, and track confirmation for each scenario. See docs/BENCHMARKS.md for detailed results and interpretation.

Integration

  • ArduPilot/MAVLink compatible - Standard autopilot protocol
  • QGC-compatible measurement formats - Ground control station integration
  • Standard Kalman APIs - Drop-in for existing C2 architectures
  • Zero proprietary dependencies - Pure C++17, Eigen linear algebra

What This Is

The deterministic sensor fusion foundation for C2 systems that operate in contested, denied-comms environments.

What This Isn't

This is not a complete C2 system. It's the fusion layer.

For decision authority and kinetic authorization, see COMMAND CORE (proprietary). For tactical display and operator interface, see HOTL (proprietary).

This is the math layer that both run on.

Academic Sources

All algorithms are published, peer-reviewed:

  • Kalman, R.E. (1960) "A New Approach to Linear Filtering and Prediction"
  • Pearl, J. (1988) "Probabilistic Reasoning in Intelligent Systems"
  • Kuhn, H.W. (1955) "The Hungarian Method for the Assignment Problem"
  • Dempster, A.P. & Shafer, G. (1976) "A Mathematical Theory of Evidence"

Validation by AFRL (Weintraub, Von Moll, Casbeer, Garcia, Pachter).

License

MIT. Build on it. Fork it. Own it.

Contact

This repository is the open foundation (MIT).

Polybolos Institute also maintains a proprietary catalog of additional capabilities that are not published here. Contact us to discuss production deployment and commercial licensing.

mark.brown@polybolos.org · https://www.polybolos.org