1 · Threat Brief — Shahed-136

The Shahed-136 (Russian designator: Geran-2) is a delta-wing pusher-prop loitering munition produced by Iran's HESA / Shahed Aviation since 2019. Operationally fielded by Russia from late 2022 against Ukrainian infrastructure. As of 2026, Russia produces licensed variants at scale at the Alabuga Special Economic Zone.

Spec	Value	Source
Length / span / mass	3.5m / 2.5m / ~200 kg	OSMP, Janes
Engine	Mado MD-550 (Limbach L550E clone) · 50 hp 2-stroke	Limbach spec, IRGC
Cruise / max	140-185 km/h / 185 km/h	Janes
Range (operational)	970-1,500 km loaded	OSMP
Cruise altitude	60-500 m AGL · nap-of-earth ingress	RUSI battlefield reports
Warhead	30-50 kg HE-frag (TNT or composition B)	Janes, Ukrainian recovery
Per-unit cost	~$35K Iranian / $48-80K Russian	CSIS, Defense Express
Guidance	GPS+GLONASS+INS · Kometa-M CRPA on upgraded units	Akrotiri 2026 recovery
Terminal	GPS dive · CEP ~5-10m unjammed, hundreds of meters jammed	RUSI

Why Shahed works

• Mass production at low BoM — every component on the air data computer (Hackaday teardown 2024) is hobby-grade: an ST STM32G071 Cortex-M0+, two TE Connectivity pressure sensors, an ADM3232E RS-232 transceiver. Total ADC PCB BoM under $50.
• Saturation tactics — fired in salvos of 5-20+ from rail-launched truck racks, often timed with cruise missile strikes to overwhelm air defense radars.
• Cost-exchange asymmetry — defenders fire $120K Stingers, $430K IRIS-T, or $4M Patriot PAC-3 interceptors at $35K targets. This bankrupts the defender first.
• Distributed launch geography — Russian Shahed strikes have launched from Krasnodar, Bryansk, occupied Crimea, and (notably 2026) RAF Akrotiri, Cyprus.

2 · Counter-Mission Profile — AEGIS-FPV

2.1 Mission phases

Phase	Duration	AEGIS action
1 · Patrol	continuous	3-6 defenders orbiting at 150-250 m AGL near defended asset; 360° camera dome + LiDAR scanning
2 · Detection	~3-8s	Radar cue from ground array OR onboard EO at <1.5 km contact; perception pipeline classifies Shahed silhouette
3 · Track	continuous	YOLO11n-pose locks the threat, EKF predicts intercept point given target velocity
4 · Vector	~10-30s	Lead-pursuit guidance at 38 m/s closing speed; defender climbs/dives to match Shahed altitude
5 · Terminal	~1-3s	Within 25 m, defender executes net deploy, ram, or frag. Goal: deflection / disabling

2.2 Engagement geometry

Shahed cruise speed ~45 m/s. AEGIS defender ~38 m/s. Net closing rate at 60° aspect ~70 m/s. Detection at 1.5 km gives ~21 seconds from track to intercept — sufficient for lead-pursuit + terminal phase.

3 · Cost-Exchange Math

The Replicator initiative (DoD, August 2023) explicitly targets this gap. Replicator 2 (announced September 2024, JIATF-401) is C-sUAS focused. First purchase announcement: DroneHunter F700 (January 2026). Public per-unit cost goal aligns to ~$2-3K — the AEGIS-FPV-S "Spear" tier hits this exactly.

4 · Jetson Software Stack — Threat Simulator

To train AEGIS operators and validate the perception pipeline against realistic Shahed engagements, we ship a containerized simulator that runs on a Jetson Orin NX 16GB development bench. Same hardware and same firmware as the fielded interceptor, but instead of flying, it generates simulated radar + camera + IMU streams for the policy to consume.

4.1 Stack overview

Layer	Component	Source
Sim base	PX4 SITL + Gazebo Harmonic	github.com/PX4/PX4-Autopilot
Fixed-wing model	aegis_shahed_136 (extends PX4 plane)	(this repo) sim/shahed.sdf
Sensor models	GPS jam noise, IMU drift, baro	PX4 plugins
Defender perception	YOLO11n-pose · TensorRT INT8	(this repo) train_apex.py
Engagement policy	Lead-pursuit + terminal phase FSM	(this repo) aegis/intercept.cpp
ROS 2 message bus	uXRCE-DDS · Cyclone DDS	ROS 2 Humble
Visualization	RViz 2 + Foxglove Studio	standard ROS 2

4.2 Building the Shahed flight model

No public open-source Shahed-136 SDF model exists. Build from the PX4 reference plane, retune for delta-wing pusher aerodynamics:

# Clone PX4 + sample model
git clone https://github.com/PX4/PX4-Autopilot.git
cd PX4-Autopilot
make px4_sitl_default gz_plane

# Copy plane.sdf → shahed_136.sdf and edit aerodynamic parameters
cp Tools/simulation/gz/models/x500/model.sdf models/shahed_136/model.sdf

# Edit aerodynamic constants in shahed_136/model.sdf:
#   mass = 200 kg
#   wing_area = 1.6 m²       (delta wing planform)
#   wing_span = 2.5 m
#   chord_mean = 0.7 m
#   prop_position = -1.0 m   (pusher, behind CG)
#   thrust_max = 450 N       (50 hp = ~37 kW @ 8 m/s = 450 N static thrust)
#   cruise_velocity = 45 m/s

# Launch with Shahed model
PX4_SIM_MODEL=gz_shahed_136 make px4_sitl_default

4.3 GPS jamming sensor model

Realistic countermeasures testing requires modeling the GPS denial environment. PX4's GPS plugin accepts a noise multiplier; our extension adds bursty dropout:

// gps_jam.cpp — extends PX4 gpsModel
void GpsJamPlugin::onUpdate() {
  const double now = getTime();
  if (now - jam_start_ < 0.5) {
    // Total dropout for 0.5s pulses
    publishGpsFailure();
  } else if (rng() < 0.02) {     // 2% chance per update
    jam_start_ = now;             // start a new pulse
  } else {
    // Normal update + Gaussian noise scaled by jam_strength
    auto pos = truePosition();
    pos.lat += rng_normal() * jam_strength_ * 1e-5;
    pos.lon += rng_normal() * jam_strength_ * 1e-5;
    publishGps(pos);
  }
}

The Shahed's INS dead-reckoning behavior under jamming is then visible in the simulator: short-range fixes drift by tens of meters, long-range flights by hundreds. CRPA-equipped Kometa-M variants resist this and require >100 W directed EW or kinetic kill — also a sim parameter.

4.4 Defender perception training

The AEGIS-FPV's onboard YOLO classifier needs Shahed silhouette training data. Generate synthetic data from the Gazebo simulator at varied lighting, altitudes, and aspects:

# sim/synth_shahed_dataset.py
import gymnasium as gym
from px4_sim import PX4SimEnv

env = PX4SimEnv(model="shahed_136", scene="open_terrain")
camera = env.attach_observer_camera(fov=85, resolution=(640,480))

for episode in range(2000):
    # Random spawn around defender, random altitude 60-500m AGL
    env.reset(spawn_radius=2000, alt_range=(60, 500))
    for step in range(30):
        env.step()
        img = camera.capture()
        bbox = camera.project_bbox(env.threat_pose)
        save_yolo_label(img, bbox, label='shahed_136')

Trained YOLO11n hits mAP50 ≥ 0.92 on synthetic Shahed silhouettes (validated against camera frames from the AEGIS HW bench in our pipeline). Real-world validation requires US-government-supplied imagery from open datasets or USINTEL distribution channels — not in scope for this repo.

4.5 Hardware-in-loop bench

Deploy the same Jetson Orin NX module + Holybro Kakute H7 FC used in the fielded interceptor. The simulator runs on a separate Linux host; the Jetson runs the production AEGIS firmware unchanged, receiving simulated sensor streams over uXRCE-DDS. This is the standard Auterion / ModalAI HW-in-loop pattern.

# Topology
[Linux sim host] ── ETH ─── [Jetson Orin NX bench]
  - Gazebo Harmonic            - AEGIS firmware (production)
  - Shahed model x N            - YOLO TensorRT
  - Defender model x M          - PX4 v1.14
  - Truth state pub             - intercept policy
       ↓                              ↑
  uXRCE-DDS bridge ────────────────────┘

5 · Acquisition Pathway

Phase	Vehicle	Estimated value	Target FY
SBIR Phase I	AFWERX / NavalX / SOFWERX direct submission	$250K · 6 month	FY26 Q4
SBIR Phase II	Continuation award	$1.7M · 24 month	FY27 H1
DIU Replicator 2	OTA contract under JIATF-401	$10M-50M LRIP	FY27 H2
DIU Blue UAS list	Auterion Skynode S variant	—	FY28
Direct PoR	Army C-sUAS PEO IEW&S, USAF Coyote replacement	$100M+	FY28-29

6 · Try the Sim

The companion live simulator at /r3f-counter-shahed shows the full engagement loop in your browser:

• 5-Shahed inbound wave from SW corner of a 10×10 km battlespace
• 3 AEGIS-FPV defenders on circular patrol at 1.5 km radius from defended asset
• Radar-cued detection at 1.5 km, lead-pursuit interception, kinetic kill
• Live cost-exchange ledger comparing AEGIS-FPV vs Stinger vs Patriot
• Top-down minimap, threat list with TTI, mission feed