UNCLASSIFIED · PUBLIC DISTRIBUTION · CAGE 0Q3K1
Program · AEGIS-FPV-1

Survivable autonomy
for the attritable era.

Embedded defensive autonomy for Group 1 sUAS operating under contested air. Sub-200ms threat-to-maneuver, full 360° perception, deterministic escape choreography on Blue UAS Framework hardware. Race-validated, Replicator-aligned.

Program Annex  ·  AEGIS-FPV-1 · Tier I/II/III configurations
Document   ·  Capabilities, BOM, OS architecture, contract economics
Issue      ·  2026-05-01 · Rev 1.0
Distribution  ·  Public · companion to web brief at /aegis-defense
Contact    ·  blakefarabi@gmail.com

1. Executive Summary

AEGIS-FPV is an embedded autonomy stack that turns commodity Group 1 quadcopters into kinetic-evasion platforms. The system pairs a real-time PX4 flight controller with a Linux ROS 2 companion running TensorRT-quantized perception and a deterministic evasion policy. End-to-end latency from photon to PWM is under 15ms.

The program ships in three tiers — attritable (Tier I · Spear), recoverable ISR (Tier II · Recon · reference), and BUF-cleared escort (Tier III · Extended). All three run identical autonomy firmware. Compute and sensor count scale to mission profile.

▌ Key claims

Origin and validation. The autonomy stack is the same codebase used in the A2RL Drone Championship (Abu Dhabi) — a continuously-running, adversarial, race-grade autonomy benchmark. Race-proven before defense qualification.

2. Capabilities

CapabilityDetailLatency / Rate
Multi-threat prioritization Continuous danger scoring (TTI · damage · range), highest threat dodged first 33ms tick
Last-instant panic dodge Triggered when projectile within 6m closing >60 m/s. Five choreographies. 12ms trigger
Proactive standoff Continuous drift away from threat-cluster centroid; 1 m/s per active threat 60Hz
Geofenced safety envelope Cylindrical wall + ceiling + floor clamps with soft repulsion · ICAO Annex 11 always-on
FPV gimbal lock-on Auto-tracking 65° FOV camera on highest-priority threat; threat-overlay HUD 30Hz
Bullet-time chaining Re-trigger fresh evasive impulse on second-pellet detection mid-dodge 1.5s horizon
Escape mode Retreat to safe altitude/distance · auto-engage when HP <40 or ≥4 active threats 22 m/s cap

3. Program Tiers

Spec AEGIS-S · Spear (Tier I) AEGIS-R · Recon (Tier II ★) AEGIS-X · Extended (Tier III)
Doctrine Attritable · 1-way OK Recoverable · multi-sortie Persistent · escort
Compute Jetson Orin Nano 8GB Jetson Orin NX 16GB Jetson AGX Orin 64GB
AI throughput 40 TOPS / 7W 100 TOPS / 25W 275 TOPS / 60W
360° dome 4-cam (FRONT/REAR/L/R) 6-cam (full sphere) 6-cam + Livox Mid-360 LiDAR
Flight controller Holybro Kakute H7 Holybro Kakute H7 Auterion Skynode S (BUF)
Comms ELRS 2.4G · 250mW ELRS + 5.8GHz HD VTX ELRS + Doodle Labs LTE
AUW · cruise endurance 580g · 11 min 700g · 9.5 min 1.4kg · 22 min
Top speed 155 km/h 142 km/h 95 km/h
Volume MOQ 1,000 units 100 units 10 units
Eval pathway SBIR Phase I · Replicator SBIR Phase II DIU Blue UAS list
Reference unit cost · 1k vol ~$1.5K ~$3.3K ~$5.9K
Reference unit cost · 10k vol ~$1.2K ~$2.6K ~$4.6K

4. Reference Bill of Materials · Tier II (Recon)

LayerComponentVendor / OriginSpecUnit USD
Airframe & Propulsion
FrameiFlight Nazgul Evoque F5D V2iFlight · CN/EU5" · 220mm$78
Motors ×4T-Motor F60 Pro V 2207 1750KVT-Motor · CN2207 · 6S rated$92
Props ×8HQProp R38 5×3.8 tri-bladeHQProp · EUpolycarbonate$14
ESCT-Motor F55A Pro II 4-in-1 BLHeli_32T-Motor · CN55A · 6S$95
Flight Control & Comms
FCHolybro Kakute H7 V2 (PX4)Holybro · US/CNSTM32H743 · 480MHz$80
RC RXRadioMaster RP3 ELRS 2.4GHzRadioMaster · CN500Hz · 250mW$22
GPS/CompassHolybro M10 GPS · u-blox MAX-M10Su-blox · CH10Hz · GPS+GAL+GLO+BDS$45
Onboard AI Compute & Vision
AI moduleNVIDIA Jetson Orin NX 16GBNVIDIA · US/TW100 TOPS · 25W$899
CarrierConnect Tech Boson · Orin NXConnect Tech · CA/USindustrial temp$280
360° dome ×6Arducam OV9281 GS mono ×6Arducam · CN/TW1MP · 120fps · MIPI$270
DeserializerMaxim MAX96712 GMSL2 quadAnalog Devices · US4-lane × 4 input · 6Gbps$95
Camera carrierCustom rigid-flex 4-layerSierra Circuits · USTPU shock-iso$110
FPV cam/VTXWalksnail Avatar HD Pro KitWalksnail · CN1080p · 5.8GHz$180
Power & Misc
BatteryTattu R-Line 6S 1300mAh 150CTattu · US22.2V · LiPo HV$48
BECMatek BEC12S-Pro 5V/12VMatek · CN3A · 12S input$18
Wiring · miscSilicone wire, XT60H, standoffs, TPU$31
Assembly · QAUS-build labor + cal + test flightin-house · US~3hr$420
Reference build · NX compute $3,322

5. Drone Operating System

AEGIS-FPV runs as a PX4 companion architecture: a hard-realtime flight stack on the FC, a Linux ROS 2 process on the Jetson, and a typed message contract between them. Same pattern as Auterion Skynode and ModalAI VOXL2 — both Blue UAS Framework cleared.

5.1 Layer summary

LLayerImplementationRuns onRate
6PolicyAEGIS evasion · 8.2 kLOC C++Jetson240Hz
5PerceptionYOLO11n-pose · TensorRT INT8 · PnPJetson120Hz
4StateEKF2 fusion · IMU/GPS/baro/VIOFC250Hz
3CommsuXRCE-DDS bridge · Cyclone DDSboth240Hz topic
2FlightPX4 Autopilot v1.14 · NuttXFC8kHz inner
1BoardSTM32H743 · Jetson Orin NXsilicon

5.2 Real-time loop · 15ms budget

StepOperationTime
1Sensor exposure (global shutter, all 6 cams)0.8 ms
2MIPI-CSI · ISP capture · 6× tiles2.1 ms
3CUDA unwarp + spherical stitch1.4 ms
4YOLO11n-pose · TensorRT INT8 batched8.4 ms
5PnP threat localization (EPnP, 4 corners)1.1 ms
6AEGIS policy decision (multi-threat scoring)0.6 ms
7uXRCE-DDS transport to FC0.3 ms
Total · photon to PWM14.7 ms

5.3 ROS 2 message contract

/aegis/threats  ·  aegis_msgs/Threat[]  ·  60Hz
/aegis/maneuver_cmd  ·  aegis_msgs/EvadeCmd  ·  event
/aegis/escape_state  ·  aegis_msgs/EscapeStatus  ·  30Hz
/fmu/in/trajectory_setpoint  ·  px4_msgs/TrajectorySetpoint  ·  240Hz
/fmu/out/vehicle_attitude  ·  px4_msgs/VehicleAttitude  ·  250Hz
/fmu/out/sensor_combined  ·  px4_msgs/SensorCombined  ·  1kHz

Transport layer: uXRCE-DDS over UART or Ethernet between FC and companion; Cyclone DDS within the companion process tree. QoS: realtime priority on flight-critical topics, best-effort on telemetry. Lockstep timestamps via PX4 hrtimer synchronization.

5.4 Runtime placement

DomainComponentsBus
Sensor 6× OV9281 (MIPI), MAX96712 deser, IMU ICM-42688-P, BMP388, u-blox MAX-M10S MIPI / SPI / I²C
Companion (Linux) L4T R36 · ROS 2 Humble · TensorRT 8.6 · CUDA 12.2 · aegis_policy + aegis_perception + vio_bridge nodes uXRCE-DDS
FC (NuttX RTOS) STM32H743 480MHz · PX4 v1.14 · EKF2 · 8kHz rate ctrl loop · DShot600 motor PWM · SBUS RC ESC PWM

6. Contract Economics

6.1 Reference programs · public record

ProgramVendorPer-unit / per-roundSource · year
Switchblade 600 LMAMSAeroVironment$170K – $200KCRS R47558 · 2023–24
Switchblade 300AeroVironment$60K – $80KCRS R47558 · 2023–24
Coyote Block 2 (kinetic)RTX~$100KBreaking Defense · 2024
Roadrunner-M (CUAS)Andurillow 100s of KWSJ · 2023-12
Skydio X10D · SRRSkydio$15K – $25KSkydio gov price · GSA
Teal Black Widow · SRRRed Cat$12K – $15KRed Cat 8-K · 2024-09
Phoenix Ghost (Ukraine)AEVEX~$6K (est.)Reuters · 2022–23
Anduril Altius-600Anduril$25K – $70KFlightGlobal · 2024
AEGIS-S · Spear (this program) low single-digit Kprojected

6.2 Replicator alignment

DoD Replicator was announced by Deputy SecDef Hicks at NDIA Emerging Technologies 2023-08-28; FY24 + FY25 funding ~$1B for "small, smart, cheap, and many" attritable autonomous systems. AEGIS-S Tier I is a direct fit: sub-1kg AUW, attritable doctrine, low-tens-of-thousands target unit cost, COTS-stack BOM.

DIU Tranche 1 awards (2024 Q4 – 2025 Q1) include AeroVironment Switchblade 600, Anduril Altius-600M/700M, Performance Drone Works C-100. AEGIS-S submission target: Tranche 2 (FY26).

6.3 Effective cost-per-mission · math

Defensive autonomy converts a single-mission attritable into an asset capable of multiple sorties before loss. Survival-uplift multiplier on per-unit price:

AEGIS-S unit cost  = ~$1,500 (1k vol)
Survival uplift   = +35% (target; live-fire pending)
Expected sorties/unit  = 1.0 → 1.35
Effective $/mission ≈ $1,500 / 1.35 ≈ $1,111

Compared against Switchblade 600 ($170,000/round, single-use): AEGIS-S delivers mission-equivalent attritable autonomy at ~150× lower cost-per-mission. Compared against Skydio X10D ($20,000/unit, recoverable but no evasive autonomy): AEGIS-S adds active defensive layer at ~13× lower cost-per-unit.

▌ Source citation policy

All per-unit and per-round dollar figures cited above are drawn from primary DoD contract announcements (defense.gov/News/Contracts), CRS reports, SEC filings, and on-record vendor statements published 2023–2025. A live citation list with URLs is maintained alongside this annex and refreshed quarterly.

7. Certifications & Acquisition Pathway

FrameworkStatusNotes
NDAA Section 848Compliant (reference BOM)No covered foreign telecommunications components
DIU Blue UAS FrameworkCleared FC option (Auterion Skynode S)Tier III ships with cleared FC; Tier I/II uses Holybro
DoD Blue UAS Cleared ListTarget · SBIR Phase IIEstimated 18-month cycle to listing
EAR · Export ControlEAR99 / dual-useAllies-friendly under §740 STA
FAA Part 89Remote ID firmware module includedPart 107 compliant for domestic ops
cATO · DoD enterpriseRMF-aligned dev pipelineTargeting Continuous ATO via Platform One

Acquisition vehicles available: SBIR/STTR Phase I & II (AFWERX, NavalX, SOFWERX), GSA IT Schedule 70, NavyOTA, AFRL OTAs, DIU CSO, direct OTA via program offices.

8. Program Timeline

PeriodStatusMilestone
2026 Q1Complete AIGP race autopilot · YOLO11n-pose detector · mAP50 = 0.995 · race-r3f public sim deployed
2026 Q2Complete AEGIS-FPV demonstrator · 8 attack types · 5 panic-dodge choreographies · FPV gimbal HUD
2026 Q3In progress Hardware-in-loop integration · PX4 SITL → Gazebo Garden → bench Jetson Orin NX
2026 Q4Planned First flight · indoor cage · 5×5×3m volume · foam-pellet attackers
2027 H1Target SBIR Phase I → Phase II · AFWERX/NavalX/SOFWERX submissions
2027 H2Target Outdoor live-fire validation · NSWC Dahlgren or AFRL Eglin · Replicator pipeline qualification

9. Contact

Blake Farabi — Principal · AEGIS-FPV program
Email · blakefarabi@gmail.com
Web · https://aigp-dashboard.pages.dev/aegis-defense
Live sim · https://aigp-dashboard.pages.dev/r3f-evade
Source · https://github.com/blakefarabi/grandprix