The error is treating two different engineering problems as though they are the same. Detecting weak, unintended emissions at useful standoff range remains an unsolved problem. I said as much in my own gap analysis, and I stand by it. That said, constructing a strawman is not a productive way to advance discussion. Applying electromagnetic or optical energy strongly enough to create an effect is a separate problem with separate physics and a separate evidence base. Citing the limitations of the first as evidence against the second substitutes one question for another, a fallacy. Drone-Mag and DroneTrace at 500 m do not meaningfully speak to Leonidas, sensor dazzling, directed optical effects, and so on.

The point I would press is that your own branch is EW-dependent in both directions. Since May 2026, Russia has fielded Shahed variants carrying rear-mounted frequency-suppression jammers aimed specifically at the command links of Ukrainian interceptor drones. In other words, Russia is spending payload and range on EW systems to protect strike drones from interceptors. They continue to field dazzlers as well, searching for the favorable SWaP trade. That expenditure only makes sense if EW is viewed as a meaningful partner in the engagement.

I would also argue that many organizations entering the interceptor market are doing so in an environment where 140 to 150 companies are effectively building variations of the same thing. One could say this is simply how markets work, but that explanation is incomplete. Many factors shape outcomes, yet large flows of capital can materially distort market signals. JIATF-401’s public record illustrates the point. More than $600 million had been committed by April 2026, including $350 million in the first month of Operation Epic Fury, a $500 million-ceiling IDIQ to Perennial Autonomy for Merops interceptors in May, and a sole-source AeroVironment IDIQ in July. By comparison, on the degradation side over the same period there was an $11 million award to Epirus for HAVOC in August and a $49 million-ceiling OTA to ThinKom for Alecto, where initial funding covered only one of four requested prototypes. Hundreds of millions versus tens of millions, and one of those columns is still largely prototype work.

The obvious question is: what numbers are driving those decisions?

JIATF-401’s framing includes the claim that EW solutions are defeated within 21 days of deployment, a period shorter than any acquisition cycle. That figure describes a specific category of fixed-waveform RF denial. Yet it is being used to dismiss a broader category that includes HPM, dazzling, obscurants, deceptive techniques, and adversarial inputs, none of which necessarily operate on the same adaptation clock. Like the argument about EMI, a number that may be valid for one technique is being used to diminish an entire class of approaches.

The interceptor layer itself was built against a roughly 185 km/h target. Geran-4 cruises at 400 to 500 km/h, placing it outside a tail chase for that class of interceptor. Meanwhile, jet employment grew from roughly 180 airframes across all of 2025 to about 1,400 in the first half of 2026 alone. A single-mechanism architecture encountered a mechanism-specific vulnerability and was partially obsoleted on schedule, while capital continued to flow into it. That is what “what happened to winning?” was asking about, and it is now answerable with dates.

Dominant design is probably the closest single concept to what has happened with interceptors, though it is not unique to them. Once a product architecture wins, investment shifts from exploring alternative architecture to refining the winner. Radical alternatives become starved, not because they were rigorously evaluated and rejected, but because the question itself stops being asked. Actors observe one another’s choices, rationally infer those choices contain information, and gradually stop acting on their own private signals. The result is convergence backed by far less independent evidence than the number of participants implies. Fifty funds backing interceptors does not equal fifty independent assessments. At some point, organizations, groups, and individuals begin copying one another because that is how funding is secured.

As this convergence deepens, it often turns into a competency trap. Organizations continue doing what they are already good at because it produces reliable local success, while potentially superior alternatives remain unexplored because no corresponding competence is ever developed. “The branch we have proven works outside the lab” is almost a textbook example. History is full of cases where organizations facing disruption doubled down on existing capabilities rather than developing new ones. That is precisely why concepts such as dominant design, crowding out, and mimetic isomorphism exist in the literature. These forces shape decision-making whether participants recognize them or not, usually through a combination of incentives, assumptions, and institutional inertia.

In closing many often dismiss, ignore, or categorially deny what they don’t understand and when they purposely make no effort to understand. This happened 2023 into 2024 with Ukraine and the idea autonomous weapons system and grouped drones, “swarms”, only later having Ukraine invest heavily in autonomy, something they could have been a disruptor in if they had six months earlier. Similarly interceptor designers and makers are building their systems with onboard vision and recently with acoustic cueing, which is a perception stack, and that is exactly the surface that dazzling, obscurants, and adversarial patterning seek to attack. Israel’s own policy conclusions regarding Hezbollah fiber-optic FPVs emphasized optical detection, microwave and smoke-based disruption, and destruction by lasers or low-cost interceptors. That is a layered solution, not a kinetic-only one. If your terminal phase depends on an optical seeker, that is an argument for better understanding the optical-attack branch, not for prematurely dismissing it.

There are deeper issues here than some disagreements. They are hard to see, and when you see them it is harder to come up with a way to stand apart and pursue your own path. Through history the individual or group with a great invention ignored or overlooked. In case of war and conflict civilizations get stuck in a challenge response system where actors continue to peruse their ends in their prescribed manner. The conflicts end for many reasons, usually a mix of force, economics, natural events, diplomacy, and so on. Often part the winners of a conflict are the first to bring together changes in technology, innovation, culture, and the like to create rapid disruptions they can exploit and potentially end conflict. So it is worthwhile asking why we are not winning. It is important for those in defense to consider if there are perhaps better ways to effect change. This often means moving against the established ways to find to new paths and better ways.

Two things usually work their way into those recordings. One is the story I begin constructing around the work and what I think I’m seeing in it. The other is whatever is happening in my life at the time. That second part is why I have no particular desire for the recordings themselves to circulate. They are partly process notes and partly a private record of the circumstances in which the work was made.

The process is also difficult to reproduce. To some extent, you get what you get. If the materials are ready, the canvas is positioned properly, the camera is charged, and the lights are where they need to be, the odds improve. I almost never manage it that cleanly. Usually I’m moving things, adjusting the camera, changing the light, looking for another material, and trying to follow something that has started to happen before it disappears.

That is also why some of the individual still images can seem underwhelming or simply strange when separated from the process. A particular image may make considerably more sense while it is being made.

When other people are there, I usually let them participate. I hand them a light or let them move parts of the material, sculpture, artwork, picture—whatever we happen to be calling it at that point. We work through it together. Sometimes the result is remarkable. Just as often it is blurry, too faint, badly lit, or obviously needs a stronger source or a different position.

There is art and craft in the work, but I think the best way to connect with it is to become physically part of it. I don’t mean sitting on the couch watching. I mean picking up and placing materials, moving things around, blocking the lights, spilling your drink, trying to fix the piece, and then laughing because none of the bits and parts costs much of anything. In that state, they really only have value because of what they are doing together, in that particular space and moment.

One of the best examples was a woman who came over and immediately got involved in making one with me. At some point she spilled her wine on part of it. It didn’t matter. We grabbed another picture, a sheet of plastic, some tape, dried plants, and a few things from outside and kept going. Nothing had been ruined because none of those individual objects was precious. They only mattered because of what we were doing with them at that moment.

And it was fun.

Other artists also physically enter into their work: moving things, breaking things, placing materials in new positions. I’m doing that too, except that light is one of the materials. I know artists use light and movement in many different ways, and I feel as though I’m working in a parallel space. I’m often moving lights for the narrative, for these small stories, and for the way the image itself changes while it is being made.

There is something about the method that not many people know. It came out of research I was doing, funded with a small amount of money over a fairly long period by a government organization and one of its laboratories. The original goal was not to make art. It was to explore whether light and physical objects could confuse artificial-intelligence and machine-learning systems whose sensing is performed through a camera.

I still look at the work with some of that research instinct. While I’m making something, part of me is watching for an effect that might have some relevance to my work on autonomous systems or machine perception. Sometimes it is a distortion in the camera image; sometimes it is the way light produces an unexpected structure.

I had one fifteen-second clip, for example, in which something appeared that looked almost like a blood cell, along with distortions elsewhere in the image. I don’t really know what caused it yet, so I want to set the cameras up again and see whether I can reproduce it.

Other discoveries are much more physical. Some extraordinarily detailed shapes produced by the light were simply leaves. In one of the darker images, the almost black structure is a leaf. What interests me is that these forms are not only visually complicated. Similar branching and cellular structures turn up in engineering research, including work on storage and material architectures. I like discovering those connections, especially when I was not looking for them.

That is probably the part of this process I find most engaging. I’m making art, but I’m also actively learning. You notice something you did not expect, follow it for a while, discover something you did not know, and occasionally find that it connects to an entirely different field.

You do not have to understand the technical part to participate. You can move a light, hold up a leaf, tape something in the wrong place, spill your drink, discover that the mistake produces something interesting, and keep going.

Some of the resulting art can seem odd, particularly as a still image. But while I am making it, I am also looking at it differently, with a different kind of attention and a different intent. I’m watching for things that appear only briefly: shapes, relationships, distortions, or images that were not really there a moment before.

For anyone looking at these images, this is a particularly bizarre sequence from the same session. If you can, enlarge the image on your monitor and move slightly closer to and farther away from it. There is a point where the contrast between the black and red becomes much stronger and parts of the image become easier to separate.

You may see what looks like a menacing cloud of red light rising out of the drawing of a woman. It’s a little creepy.

The next still was taken about twenty seconds later. In the video, you can actually watch something begin to take shape.

What do you see?

Image: AI rendered image of Filipino soldiers assembling and preparing to fly a drone.

The Philippines Is Not Ukraine

The war in Ukraine has demonstrated the military value of inexpensive drones, rapid technological adaptation, distributed manufacturing, and the close integration of reconnaissance with attack. These lessons deserve careful study. They should not, however, be transferred mechanically to the Philippines.

The Philippines is not Ukraine. Its geography, climate, infrastructure, human terrain, and industrial position would produce a substantially different form of drone warfare. Unmanned systems would undoubtedly be important in a future conflict, but they would operate within an environment that constrains communications, movement, maintenance, surveillance, and logistics in ways that are often missing from current discussions.

The deeper lesson from Ukraine is not that every future battlefield will be covered by the same mixture of reconnaissance quadcopters and first-person-view attack drones. It is that military organizations must adapt accessible technologies rapidly to local conditions.

In the Philippines, those conditions are exceptionally demanding.

An Archipelago of Operational Compartments

The Philippines consists of thousands of islands containing narrow coastal plains, mountains, river valleys, forests, agricultural areas, settlements, and uneven infrastructure. In many locations, steep terrain rises quickly from the coast. Roads and population centers are frequently concentrated along limited corridors, while ridges, ravines, vegetation, and waterways divide the landscape into smaller operational compartments.

This geography would affect nearly every aspect of military activity. A unit may be geographically close to another unit while being separated by a mountain, dense vegetation, a difficult river crossing, or an expanse of water. Moving personnel, batteries, replacement aircraft, chargers, antennas, fuel, tools, and spare components between islands or inland positions could become as important as acquiring the drones themselves.

Small drones could operate effectively above the canopy, along coastlines, roads, rivers, settlements, ridgelines, plantations, and open agricultural areas. They could also be launched selectively from clearings or other favorable positions.

What should not be assumed is that the persistent low-altitude drone coverage seen in portions of Ukraine can be reproduced across mountainous, forested islands.

The Jungle as Concealment and Obstacle

Although logging, agriculture, mining, and urban development have substantially reduced and fragmented Philippine forests, large areas of dense vegetation remain. A landscape does not need to be untouched rainforest to complicate military operations. Secondary growth, bamboo, plantations, steep ravines, thick undergrowth, mixed agricultural vegetation, and forested slopes can all restrict visibility, mobility, communications, and aerial observation.

Consider a relatively small hostile team entering through the maritime environment. It might arrive aboard a small watercraft, possibly blending into the extensive civilian and fishing traffic common around the islands. After landing away from a major port, the team could move through a combination of coastal communities, forest, agricultural land, and mountainous terrain toward an objective.

Once beneath dense vegetation, it would gain meaningful protection from routine aerial observation. This does not make personnel invisible. Thermal imagers, persistent surveillance, acoustic sensing, tracking dogs, human reporting, and other collection methods could still reveal them. Movement, heat leakage, disturbed vegetation, fires, waste, communications, and interaction with local populations could all produce signatures.

Nevertheless, canopy, tree trunks, broken terrain, rain, fog, thermal clutter, and restricted observation angles can substantially degrade conventional optical and thermal surveillance. Simple thermal coverings may reduce an exposed heat signature under some conditions, particularly when combined with natural vegetation and limited movement, but they are not magical and individuals can be tracked,.

The tactical advantage comes from combining multiple methods of from insultation to barriers, and the environment. Using multiple methods of concealment or obfuscation rather than relying on any single countermeasure. That being said places like the Philippines should consider leap frogging current detection and even countermeasures by leveraging products from companies like Kallistai SHield whose products manipulate the AI behind modern scenors https://www.linkedin.com/company/kallistoai/

Small Teams and Selective Drone Use

A hostile team operating beneath the canopy would not need to maintain continuous radio contact or constant drone coverage. It could listen more than it transmits, use prearranged procedures, move during favorable weather, and choose the location and timing of its drone operations.

Several compact aircraft could be carried disassembled or in protective containers. The team might launch them from a shoreline, road, riverbank, field, ridgeline, or natural clearing for short reconnaissance missions. A drone might briefly rise above the canopy, observe an area, relay information, act as a decoy, deliver a sensor, or interfere with another unmanned aircraft.

The important point is that such a team could use drones intermittently rather than continuously. It could select the time and location of exposure while forcing defenders to monitor a far larger area.

The jungle would also produce natural chokepoints. Trails, river crossings, gaps in vegetation, ridgelines, roads, and usable landing areas could become predictable movement routes. A small force familiar with the terrain could prepare observation points, false trails, obstacles, warning devices, and ambush locations before a pursuing force entered the area.

Local armed groups, criminal networks, smugglers, or insurgent organizations could further complicate the environment. A foreign adversary might attempt to exploit existing routes, informal transportation, local knowledge, or established clandestine networks. This possibility should not be exaggerated into the assumption that local populations would automatically assist an outside force. Local terror groups or politically align guerillas could perform or assist adversarial forces. Loyalties, grievances, political interests, and economic incentives would vary greatly by location. The human terrain would be as important as the physical terrain.

Trusted community relationships, knowledge of normal maritime and village activity, and reliable local reporting could prove more valuable than attempting to maintain uninterrupted aerial surveillance over every forested area.

Communications Beneath the Canopy

Communications problems in jungle environments are sometimes attributed too broadly to humidity. Humidity alone is unlikely to cause the complete failure of common drone links.

The larger problems are terrain masking, blocked line of sight, dense vegetation, wet foliage, antenna position, multipath propagation, and the depth of the signal path through the forest.

A drone flying above the canopy may retain an effective link while an aircraft only a short horizontal distance away, but below the canopy or behind a ridge, may lose both control and video. Dense wet vegetation can absorb and scatter radio-frequency energy, while mountains and ravines create abrupt coverage boundaries.

This would also limit the use of conventional ground robots. Most wheeled or tracked systems demonstrated in open terrain are poorly suited to roots, bamboo, mud, streams, fallen trees, steep slopes, narrow trails, and thick undergrowth.

An unmanned system expected to operate beneath the canopy would require some combination of greater autonomy, onboard navigation, obstacle avoidance, terrain-relative sensing, preplanned behavior, communications relays, or a safe response to link loss.

Each of these features adds cost, weight, power consumption, training requirements, and technical complexity.

The Maritime Climate Is an Active Adversary

Image: U.S. Army Lightning Lab director Lt. Col. Nathan Whitney shows off small drones and their components at Fort Magsaysay, Philippines, April 30, 2026. Photo by Aaron Kidd, Stars and Stripes.

The environmental problem extends far beyond radio performance.

The Philippines is hot, humid, wet, and maritime. Equipment may be moved repeatedly between air-conditioned rooms, vehicles, boats, exposed decks, coastal lowlands, mountain areas, and warm forest. These transitions can create condensation on camera lenses, circuit boards, battery contacts, connectors, antennas, sensors, and motors.

Salt-laden moisture is particularly damaging. It can leave conductive residue, accelerate corrosion, and create intermittent faults that do not appear during an initial inspection. Rain, mud, mold, insects, and contaminated water create additional problems.

The DJI FPV system is an example of the distinction between excellent manufacturing and environmental suitability. DJI produces highly integrated and impressively engineered aircraft. That does not mean a commercial drone is designed to endure repeated exposure to salt spray, tropical rain, mud, abrupt temperature changes, and prolonged field use without protected storage and regular maintenance.

A force operating near the ocean would likely need sealed or weather-resistant transport containers, desiccants, drying procedures, corrosion control materials, protected charging locations, spare connectors, replacement motors, cleaning supplies, and tools for diagnosing intermittent electronic problems. Even these precautions create new procedural demands.

Equipment sealed while still damp may deteriorate inside its container. Desiccants must be checked and replaced. Batteries require separate storage and fire-safety procedures. Optical surfaces must remain clean. Containers opened in warm, humid air may allow immediate condensation on equipment previously stored in a cooler environment. The drone may be inexpensive. Keeping it operational is not.

The Maintenance Force Behind the Drone

Drones are often discussed as though the aircraft itself constitutes the military capability. It does not.

A functioning drone organization includes pilots, maintainers, assemblers, trainers, intelligence personnel, communications specialists, electronic-warfare personnel, transportation, batteries, charging equipment, software, spare parts, storage, repair facilities, and replacement aircraft.

Pilot training also cannot be reduced to a single universal number. Requirements vary according to the aircraft, mission, unit, and expected proficiency. A basic FPV pilot, a reconnaissance operator, an interceptor pilot, and a long-range strike crew perform different tasks. An experienced Ukrainian operator recently described a training period of approximately three months in his context, but that should not be treated as a fixed requirement for every force or system. The broader lesson remains valid: useful drone operators are not created instantly.

Depending on the unit and mission, a drone detachment may also require several support personnel for every active pilot or aircraft. The ratio could become less favorable in the Philippines because of corrosion, battery management, weather protection, maritime transportation, dispersed operating locations, and limited access to repair facilities.

A large drone inventory could therefore pull Marines, soldiers, sailors, or contractors away from other duties simply to inspect, dry, clean, charge, repair, configure, and transport equipment.

Contractors could perform some of this work, but they create their own requirements for transportation, accommodation, security, communications, sustainment, and force protection.

The meaningful metric is not the number of drones purchased or delivered. It is the number of aircraft that remain mission-capable after repeated exposure to salt, rain, transport, mud, humidity, operational damage, and maintenance delays.

Are Drones Really an Asymmetric Weapon?

Drones are frequently described as asymmetric weapons because an inexpensive aircraft can damage or destroy a far more expensive platform. That is a favorable cost exchange. It is not necessarily military asymmetry.

Asymmetry generally arises when one force uses a different doctrine, organization, technology, or method of operation to exploit an opponent’s vulnerability while avoiding the opponent’s strengths. A cheap FPV attacking an expensive vehicle may be part of an asymmetric strategy, but the low price of the aircraft does not make the entire drone force inherently asymmetric.

The cost comparison often excludes the system required to make the drone operational:

• Imported motors, batteries, cameras, radios, antennas, and flight controllers

• Payload integration and safety mechanisms

• Assembly and quality control

• Pilot and maintainer training

• Reconnaissance and target identification

• Electronic-warfare support

• Spectrum management

• Transportation and protected storage

• Charging, repair, and replacement

• Software configuration and updates

• Command-and-control integration

• Supporting elements

Ukraine gained an early advantage not merely because it purchased inexpensive aircraft. It developed a distinctive wartime ecosystem connecting imported components, domestic assembly, volunteer engineering, decentralized procurement, specialized units, rapid battlefield feedback, software modification, pilot training, repair, intelligence, and evolving tactics.

As both Russia and Ukraine adopted large drone forces, electronic warfare, counter-drone systems, interceptors, fiber-optic control, and industrial production, the aircraft themselves became less asymmetric. Both sides increasingly possessed comparable categories of unmanned systems, although they manufactured, organized, modified, and employed them differently.

The asymmetry shifted from possession of the drone to the ability to learn, produce, integrate, replace, and employ it more effectively.

China and the Reversal of the Cost Argument

Image: AI illustration representation SOF training in the Philippines and leering to assemble their own drones from available components. A important capability unlikely to match Ukrainian outputs.

In a Philippine conflict involving China, the largest material asymmetry may favor China rather than the Philippines.

China possesses enormous industrial capacity, extensive shipping, greater replacement inventories, substantial electronic-warfare capabilities, and a dominant position in many commercial-drone supply chains. Motors, batteries, permanent magnets, cameras, radios, controllers, mature electronic components, and manufacturing equipment may all depend directly or indirectly on Chinese production.

This creates an uncomfortable possibility: A drone force intended to provide an inexpensive asymmetric answer to China could depend upon components for which China possesses the greater industrial leverage.

Neither the Philippines nor its allies should assume that the Ukrainian production model can simply be reproduced while fighting the country that dominates much of the underlying supply chain.

This does not make Philippine or allied drone forces futile. It changes the source of their potential advantage. Rather than attempting to compete directly with China in total production, they would need to create local asymmetries through geography, dispersion, concealment, mobility, deception, human intelligence, maritime awareness, resilient logistics, and selective concentration of force.

Drones could magnify these advantages, but they would not create them independently. China could possess more aircraft, ships, missiles, sensors, factories, and replacement components while still facing significant difficulties locating small dispersed units operating among islands, settlements, mountains, vegetation, and civilian maritime traffic.

At the same time, the same geography that helps conceal Philippine and allied forces would make their own resupply and maintenance extremely difficult. The archipelago favors neither side automatically. It creates a contest in adaptation.

What the Pacific War Already Taught

The Japanese, Americans, Filipinos, Australians, and other participants in the Pacific campaigns learned how destructive tropical and maritime environments could be.

Earlier forces struggled to keep weapons dry, clean, lubricated, and protected from corrosion. They fought rust, mud, mold, insects, soaked equipment, damaged roads, disease, and unreliable supply.

Modern forces will face many of the same pressures, but those pressures will now act upon more delicate and interconnected systems: batteries, connectors, cameras, optical surfaces, antennas, motors, cooling systems, circuit boards, software-dependent devices, and charging infrastructure.

Modern technology does not remove the environmental problem. It often multiplies the number of things that can fail. Historical experience should therefore produce humility rather than confidence. Familiarity with the history of Pacific warfare is not the same as possessing equipment and doctrine suitable for a future conflict.

Likewise, experience in Ukraine is extremely valuable, but it may become dangerous when treated as universally transferable.

Experience Can Become a Liability

Every participant in a future Philippine conflict would encounter surprises. The force most at risk may be the one that believes its previous experience has already given it the correct equipment, tactics, and organizational structure.

Operators who mastered drone warfare in Ukrainian agricultural terrain, trench systems, and contested urban areas may discover that many assumptions do not transfer beneath tropical canopy, across mountainous islands, or through a maritime logistics network.

The same applies to technologically superior forces that assume sophisticated sensors will provide continuous awareness. Dense terrain, local weather, civilian activity, concealment, equipment degradation, and limited communications could repeatedly create gaps between what a commander believes is visible and what is actually known.

The most adaptable force may not be the one that begins with the largest number of drones or the most experienced pilots. It may be the force that expects equipment to fail, tests under local conditions, protects its supply chains, trains maintainers as seriously as pilots, preserves multiple methods of reconnaissance and communication, and revises its organization before the environment and the enemy expose its assumptions.

The Larger Lesson

Drones will be important in any future conflict involving the Philippines. They will support surveillance, communications, maritime awareness, deception, logistics, targeting, defense, and attack.

But their effectiveness will be shaped by the environment in which they are expected to operate. A drone designed for Philippine and Pacific operations must be considered as part of a complete system that includes weather protection, corrosion resistance, intermittent communications, autonomy, repairability, battery management, maritime transportation, distributed maintenance, and access to replacement components.

The central lesson from Ukraine is not that inexpensive drones are automatically asymmetric weapons. It is that an adaptive organization can combine accessible technology, specialized personnel, rapid learning, industrial support, and locally appropriate tactics to impose costs on a stronger opponent.

Whether that lesson can be applied in the Philippines will depend less upon how many drones are purchased than upon whether the Philippines and its allies can build the human, industrial, logistical, and operational system required to keep them functioning.

The Pacific is a maritime theater. Geography, vegetation, climate, infrastructure, local knowledge, and industrial capacity will determine which lessons from Ukraine remain useful and which become dangerous assumptions.

The side that enters the conflict believing it already understands drone warfare may be the side least prepared for what the Philippines teaches it.

Author’s Note

This analysis draws upon academic research, discussions with current and former military personnel, conversations with drone operators, and field research commissioned by the author. The commissioned research included information collected by local researchers concerning terrain, vegetation, transportation, movement through forested and rural areas, and perspectives from communities in several parts of the Philippines.

Certain identities, locations, sources, collection questions, and operational details have been withheld to protect individual privacy, security, and the author’s original research.

EMI-Based Sensing and Defeat of Autonomous Vehicles (Drones): Ecosystem Report

Focus: Passive RF/EM emissions sensing, EM fingerprinting, unintentional EMI detection, and tightly scoped EM-based non-jamming defeat methods. The report excludes broad-spectrum EMP/HEMP and conventional jamming unless the mechanism involves precise EM injections. To focus the reader on the recent work in this area primary sources are drawn from 2024–2026.

Author’s Note

This report connects to my earlier article on the 2024–2025 battlefield reality: LinkedIn Article.

I write these reports for two main reasons. First, the process itself is a powerful learning tool. Researching, verifying citations, discovering new connections, and refining ideas deepens my own understanding. Second, and more importantly, I believe solutions to serious problems emerge faster when professionals have access to diverse, well-synthesized information.

In defense and security communities, I often encounter highly specialized experts with deep but narrow knowledge. Many excel in their domains but lack foundational awareness of adjacent fields, such as drone aerodynamics, sensor physics, electronic warfare, or emerging threats. It also requires intellectual humility: the willingness to be wrong, to ask questions, to seek constructive criticism, and to examine our failures honestly. In this work, that humility matters because errors can cost lives or concede advantage to the adversary.

This is particularly evident in the field of Electromagnetic Interference (EMI). Many professionals in defense and military organizations assume little to no serious work exists on using EMI for sensing or defeating drones. There is a rich, rapidly developing ecosystem of research, companies, and technologies actively weaponizing EMI, many of which may already be familiar to those willing to look.

I encourage you to read this report with an open mind. If you finish with a clearer understanding of EMI-based capabilities, you will be ahead of many professionals in the field and potentially equipped with ideas that could protect lives or deliver game-changing systems.

A final note on ecosystem challenges: Too often, professionals operate in isolated “locked boxes,” rarely cross-pollinating ideas or communicating across disciplines. This siloed thinking slows progress and leaves vulnerabilities unaddressed. Real innovation happens when researchers, engineers, operators, and decision-makers actively share knowledge and collaborate.

Lastly, I promise, these reports are written with the awareness that capable adversaries are also learning. For that reason, I deliberately avoid operationalizing details or sensitive implementation guidance.

Backgrounder with information about EMI for the beginner.

Executive Summary

Electromagnetic interference (EMI) sits at the intersection of drone sensing and drone defeat in ways that are increasingly distinct from legacy jamming or EMP paradigms. On the sensing side, drones generate unintentional electromagnetic emissions from motors, electronic speed controllers (ESCs), power buses, and onboard processing that can serve as unique “EM fingerprints” allowing passive detection, classification, and individual authentication, without the drone transmitting a deliberate RF signal. On the defeat side, narrowly targeted EM energy injected into specific communication buses (IMU-to-flight-controller channels, SPI/I2C lines) or delivered as high-power focused microwave can disable or crash a drone while requiring far less energy and causing far less collateral damage than broad-spectrum EMP.

The broader RF-sensing market, including both intentional transmission (C2 link, Remote ID) monitoring and unintentional emission fingerprinting, was valued at approximately USD 1.45 billion in 2025 and is projected to reach USD 4.20 billion by 2034 at a CAGR of 14.7%. This report catalogs the key commercial, government, and academic players organized by their primary technical focus.[1]

Part I: The Technical Landscape

Unintentional EM Emissions as a Detection Vector

The foundational insight driving this field is that drones, like all complex electronics, radiate unintentional electromagnetic emissions (UEE) that are characteristic of their electronic architecture. Motors, ESCs, power regulators, and PCB traces emit in ways that depend on component tolerances, layout, and manufacturing variation. These emissions are not merely noise; they carry identifiable structure.[2]

The Drone-Mag framework, developed at HBKU’s CRI-Lab (Qatar) by Omar Ibrahim and Roberto Di Pietro, is the most rigorous published demonstration of this principle. Using unintentional magnetic emissions (not transmitted RF), the system achieved 100% accuracy in identifying six different drone brands and 99.9% accuracy in authenticating ten physically identical drones of the same model, meaning it can distinguish individual units off the same production line. The work was accepted for publication in ACM Transactions on Cyber-Physical Systems in 2025. It operates entirely passively, requires no hardware or software modification to the target drone, and explicitly exploits the non-idealities introduced during IC manufacturing.[5]

Complementing this, DroneTrace (University at Buffalo / SUNY, led by Wenyao Xu and colleagues) takes a different but related approach: it interrogates drones using a millimeter-wave (mmWave) probe and captures the parasitic EM response, analogous to RF fingerprinting of the hardware itself, not any transmitted signal. Evaluated across 36 commodity drones, it achieved >99% identification accuracy and >98% accuracy under real-world conditions in under 0.1 seconds of sensing time. The system is resilient to spoofing since the parasitic response cannot be easily mimicked.[6][8]

PADrone (ACM Transactions on Internet of Things, 2025) extends this logic to pre-flight safety monitoring: it uses contactless RF-based vibration sensing to detect mechanical abnormalities in drone propulsion systems before flight, without any physical contact or drone cooperation. This is a sensing modality, RF detection of structural mechanical vibration, closely related to the UEE domain.[9]

The 2025 Nature Scientific Reports paper from Tecnun / University of Navarra assessed DJI Phantom 4 EMI susceptibility using Characteristic Mode Analysis (CMA), identifying long power supply wires connecting motors to ESC boards as the highest-susceptibility pathway and validating this experimentally. This work is directly relevant to both the sensing (what to listen for) and defeat (where to inject) communities.[11]

EMI-Based Defeat: Targeted Injection vs. High-Power Microwave

The defeat side of this space divides naturally into two distinct mechanisms:

1. Targeted EMI Injection (Surgical, Low-Power)

The KAIST SysSec group (Yongdae Kim, Joonha Jang, Mangi Cho) demonstrated at NDSS 2023 that injecting a narrowly tuned EM signal at the specific susceptibility frequency of a drone’s control-unit board can corrupt IMU sensor data regardless of IMU type, model, or manufacturer. The attack targets the SPI/I2C communication channel between IMU and flight controller, a “back door” EMI coupling pathway, rather than the mechanical resonance of the sensor itself. This requires far less power than broad EMP, causes no collateral damage to surrounding electronics, and acts within a single IMU sampling cycle (leaving no time for detection and recovery). KAIST published this as “Paralyzing Drones via EMI Signal Injection on Sensory Communication Channels,” and the team has noted its potential for commercialization.[12]

A related attack vector, ConfuSense, was published at USENIX VehicleSec 2025 by Alessandro Erba, John H. Castellanos, Shail Sihag, Saman Zonouz, and Nils Ole Tippenhauer. Rather than injecting noisy EMI, ConfuSense injects messages on sensor buses to reconfigure sensors stealthily, e.g., changing accelerometer scale factors, which stalls Kalman filter state estimation and causes the drone to deviate over 30 meters or crash, without triggering standard EMI detection countermeasures.[16][18]

IOActive (Gabriel Gonzalez, Director of Hardware Security) demonstrated EMFI (Electromagnetic Fault Injection) attacks against a DJI Mavic Pro: by placing a coil near the target chip and injecting a precisely timed EM glitch during the firmware update process, memory corruption was achieved on the main processor. While requiring physical proximity in the current implementation, the research establishes the chip-level vulnerability surface and the methodology for “black-box” EM-based code execution on hardened drone platforms.[20

]

2. High-Power Microwave (HPM) as Focused EM Defeat

Epirus’ Leonidas system occupies a distinctive position: it uses GaN-based solid-state microwave emitters and digitally beamformed phased-array antennas to deliver focused, software-defined electromagnetic energy to disable onboard electronics. Unlike EMP, Leonidas is non-ionizing, directional, and software-controlled; it can engage a single target with precision or switch to broad-beam for swarm defeat. In August 2025, a live-fire demonstration at Camp Atterbury defeated 49 drones simultaneously with a single pulse. In December 2025, it demonstrated the first publicly reported defeat of a fiber-optic guided drone, a platform purportedly immune to RF jamming and spoofing, by targeting onboard electronics directly. This is the key operational relevance: as fiber-optic and autonomous drones proliferate, RF jamming becomes ineffective, and HPM-based EM defeat becomes a primary non-kinetic option. Epirus received a $43.5M Army contract for IFPC-HPM Generation II in July 2025 and has deployed systems in CENTCOM.[27][28]

Part II: Commercial Product Ecosystem

The table below maps the primary commercial players to their relevant EMI/RF sensing or defeat approach.

Part III: Academic and Government Research Groups

Besides real commercial products like Epirus Leonidas numerous nascent and emerging solutions that leverage EMI are being developed. The image shows how some of these might work.

These groups are producing the foundational science that feeds commercial development, particularly in the EM fingerprinting and targeted injection domains.

Part IV: The Fiber-Optic Drone Problem and EMI’s Unique Role

The single most important driver pushing the field toward EM-based defeat (as distinct from jamming) is the proliferation of fiber-optic guided FPV drones in Ukraine and proliferating to other theaters. These platforms have no RF command-and-control link; they are immune to RF jamming, spoofing, GPS denial, and Remote ID monitoring. This renders the entire RF-sensing-based detection and defeat ecosystem partially blind to an increasingly large threat category.[27][64]

Fraunhofer FKIE explicitly identified this as the motivation for LTE450 passive radar research. Epirus has the only demonstrated non-kinetic defeat: focused HPM targeting onboard electronics, essentially an EMI defeat mechanism, just at high power density from a directional antenna. The December 2025 Leonidas demonstration of fiber-optic drone defeat was described as the first publicly reported instance of electromagnetic interference being used to defeat a fiber-optic guided drone.[27][64]

For sensing fiber-optic drones, the unintentional EMI paradigm (motor/ESC emissions, power electronics noise) becomes the only remaining passive EM path. This is precisely the gap that Drone-Mag, PADrone-style vibration sensing, and passive radar approaches (Fraunhofer, HENSOLDT Twinvis, R2 ODIN) are positioned to address, though operational range limitations for unintentional-emission detection remain a live debate in the practitioner community.

A useful analogue from the inverse geometry supports the physical plausibility of meaningful standoff detection: a 2022 OSTI feasibility study demonstrated that drone-mounted EMI sensors can detect buried metallic infrastructure at standoff distances of 10 meters or more, exploiting the fact that linear conductors , pipes, cables, and critically, the wiring harnesses and motor cabling inside a drone , produce a slowly decaying 1/R² secondary magnetic field rather than the 1/R⁶ decay of compact targets. This decay-rate argument applies directly to the reverse geometry: the internal wiring architecture of a drone is a linear conductor system, and its unintentional magnetic emissions decay more slowly than a point-source model would predict, supporting greater detection standoff than naive estimates suggest. [DOE/OSTI Report, 2022: Analysis of the Feasibility of UAS-Based EMI Sensing for Underground Utilities Detection and Mapping]

Part V: Key Technology Differentiators

What Separates “EMI-Sensing” from “RF Sensing (C2 Link)”

Most commercial products (Dedrone, Sentrycs, D-Fend) primarily detect the intentional C2/telemetry/video RF link between drone and operator. This works well for consumer drones using known protocols (DJI OcuSync, MAVLink, Wi-Fi-based). It fails for:

● Pre-programmed (autonomous) drones with no active link

● Encrypted or frequency-hopping signals

● Fiber-optic controlled drones (no RF link whatsoever)

True EMI-based sensing, detecting unintentional emissions from motors, ESCs, and power electronics, works regardless of whether the drone is transmitting anything. The Drone-Mag and DroneTrace research specifically targets this vector. R2 Wireless’s ODIN claims to identify and track even encrypted, anonymized, or previously unseen signals using proprietary RF fingerprinting that operates on signal anomalies rather than protocol decoding; this places it between the two paradigms.[82]

The PADrone system detects RF signatures of mechanical vibration (rotor and motor signatures), a form of passive sensing of physical EM emissions from motor operation rather than intentional RF transmission.[9]

Hybrid RF Anomaly + Library Detection

Dedrone’s 2025 “Hybrid RF Detection” framework explicitly combines library-based protocol matching with RF anomaly detection to catch unknown or dark drones. Sentrycs HORIZON (2025) adds one-shot learning to adapt to unknown protocols in real time without pre-existing libraries. This architectural shift toward anomaly-based EM sensing is the industry’s acknowledgment that library-only approaches are insufficient.[36]

Part VI: Key Publications (2024–2026) Reference Set

1. The unintentional EMI detection range debate remains unresolved operationally. Near-field probes work at inches; sensitive receiver arrays with advanced signal processing may achieve meaningful standoff ranges, but no commercial product currently advertises detection of unintentional motor/ESC emissions at tactically useful ranges (>50m). This is the frontier.

2. Fiber-optic drones represent a forcing function for both passive EM sensing (cannot use C2 link monitoring) and HPM-based EM defeat (one of the only demonstrated non-kinetic options). Epirus has the demonstrated product; the sensing gap remains open.[27][64]

3. The KAIST EMI injection attack has been partially addressed by defenses (MARS, SHIELD), but these require onboard countermeasures and do not eliminate the attacker’s advantage of operating below standard detection thresholds.[79][80]

4. Commercial products are dominated by intentional-emission (C2 link) sensing. True unintentional EMI fingerprinting remains largely academic. Drone-Mag and DroneTrace are the closest to being transition-ready but remain in research prototype stage.

5. Sensor fusion is the architectural consensus: R2 ODIN + Echodyne, Dedrone RF + radar, EnforceAir PLUS (RF cyber + radar + optional jamming), Robin Radar + acoustic. Pure EMI-sensing plays are rare at the product level; the field has accepted that single-modality detection is insufficient.[53]

The following notes describe the source-selection methodology, the limitations of the evidence base, and the treatment of peer-reviewed literature relative to company press releases and other secondary reporting.

Methodology and Evidence Notes

Methodology / source selection. This report emphasizes material published or substantively updated between 2024 and 2026, while selectively retaining earlier foundational work where it remains necessary to explain current technical claims or lineage. Sources were prioritized in the following order: peer-reviewed papers and formal conference proceedings; official government or laboratory publications; primary technical documentation from research groups or companies; and then high-quality trade reporting or secondary summaries used mainly to corroborate dates, demonstrations, funding events, and product announcements. Where multiple references described the same claim, preference was given to the most primary and technically specific source.

Limitations of the evidence base. The evidence base is uneven across the field. Some claims rest on mature peer-reviewed research with reproducible methods and quantitative evaluation, while others rely on vendor demonstrations, press coverage, or conference abstracts that provide less methodological detail. In several fast-moving areas, especially product performance, operational deployments, and recent counter-drone demonstrations, publicly available information may be incomplete, selectively framed, or not independently verified. This report therefore maps the current ecosystem and technical direction rather than serving as a definitive validation of every reported capability.

How press releases vs. peer-reviewed papers were treated. Peer-reviewed papers and archival conference publications were treated as the strongest basis for technical claims about sensing mechanisms, attack pathways, measurement results, and experimental performance. This is the norm. Next, I used well-respected sources that often have information that is not covered in the literature or press. The same sources have rigorous fact-checking and editorial processes. (e.g., West Point’s Modern War Institute, Small Wars Journal, CTC at West Point, NATO/STO, DTIC, etc.)

Company press releases and promotional materials were used more cautiously and primarily for factual reporting on contracts, product launches, demonstrations, and claimed deployments. When a technical assertion appeared only in a press release or media summary and could not be matched to a stronger primary source, the report presents it as a reported claim rather than as independent confirmation.

Citation formatting was standardized in several places, including malformed bracketed references and punctuation surrounding citations. There were many problems with formatting and all fault lies with the author. The bibliography itself was not converted to a single formal style such as Chicago, APA, or IEEE in this pass.

Recommendation. Drop all of the references and the paper in your favorite LLM/AI and ask questions, and have it create a report on the topics you care about like “provide a list of techniques for using EMI in a contested environment to detect and track targets at 1 km.” I personally use Consensus AI and Google NotebookLM. NotebookLLM. Its great for topics like this since it creates visualizations, quizzes, and even videos that explain to you the concept or thing you are interested from your references.

Part VII: Related Reading, Author’s Article Series, and Beginner’s Guide

Author’s Article Series (Robi Sen, LinkedIn)

This report is part of a continuing series. The articles below represent the public development of the argument , including the original claim, the criticisms received, the responses, and the refinements that produced this document. Readers who want to follow the debate in sequence should start with the first article.

1. “The Jammer, Fiber Optic Drone, and The Unintended Antennae” (Jan 2025)

https://www.linkedin.com/pulse/jammer-fiber-optic-drone-unintended-antennae-robi-sen-dmkyc

The original article making the core claim: unintentional EM emissions from fiber-optic guided drones can be exploited for sensing and defeat. Generated the most substantive criticism, including the commenter exchange that motivated this report.

2. “Jamming Fiber Optic Tethered Drones v.02” (Apr 2025)

https://www.linkedin.com/pulse/jamming-fiber-optic-tethered-drones-robi-sen-vksac

Updated version incorporating reviewer feedback; adds foundational definitions for non-specialist readers.

3. “More Corrections and Comments on EMI” (Jun 2025)

https://www.linkedin.com/pulse/more-corrections-comments-emi-robi-sen-lay4c

Clarifies the EMI-vs-jamming distinction; addresses the inward shift of attack surface toward sensor vulnerabilities and timing attacks as autonomy increases.

4. “Response” (Jun 2025)

https://www.linkedin.com/pulse/response-robi-sen-mncsc

Addresses critics directly; cites patent US20140219449A1 on interference cancellation as relevant prior art; argues for layered EM protection.

5. “Reply to Jeremy Hodges – Technical Clarification and Evidence” (May 2025)

https://www.linkedin.com/pulse/reply-jeremy-hodges-technical-clarification-response-robi-sen-uaobc

The specific commenter exchange this report credits in its title. Hodges raised the detection-range question; this reply is where the technical debate sharpened.

6. “Responding to Further Criticism” (Jul 2025)

https://www.linkedin.com/pulse/responding-further-criticism-robi-sen-ukyfc

Defends fielded EMI systems (AARTOS, EC-37B); cites NDSS 2023 KAIST work as academic validation; addresses rhetorical misdirection in the debate.

7. “Military Sources, Manuals, and Other References Related to the Use of EMI” (Jul 2025)

https://www.linkedin.com/pulse/military-sources-manuals-other-references-related-use-robi-sen-gyeuc

Retrospective review of doctrinal sources (ATP 3-01.81 and related) showing EMI as an actively tested component of counter-UAS strategy; extends the argument with military doctrine.

8. “EW, HPM, MIL-STD-461, and MIL-STD-464” (Jan 2026)

https://www.linkedin.com/pulse/ew-hpm-milstd461-milstd464-robi-sen-cltgc

Connects the HPM defeat discussion to EMC standards; relevant background for the Epirus and KAIST sections of this report.

9. “Tracking UAVs by Unique Motor RF Signatures” (Mar 2024)

https://www.linkedin.com/pulse/tracking-uavs-unique-motor-rf-signatures-robi-sen-lbanc

Earlier article on motor RF signature tracking, the direct precursor concept to Drone-Mag and the unintentional EMI fingerprinting thread.

10. “C-UAS Latency Workbook” (Apr 2026)

https://www.linkedin.com/pulse/cuas-latency-workbook-robi-sen-2jf2c

Practical engineering companion: structured workbook for estimating end-to-end latency in C-UAS systems. Relevant to anyone operating the sensing architectures described in this report.

Beginner’s Entry Points

For readers without a background in RF engineering or electronic warfare, the following accessible resources provide the foundation needed to engage with the technical material in this report:

On Electromagnetic Interference basics:

- “Unintentional Antennas: Why PCBs and Cables Start Radiating and How You Can Stop It” , Sierra Circuits video. Accessible explanation of how internal wiring becomes an unintentional antenna, directly relevant to the drone EMI sensing argument.

- IPC-2141: Controlled Impedance Circuit Boards and High Speed Logic Design (standard reference). Explains PCB trace radiation , the physical basis for unintentional drone emissions.

- ARRL Handbook for Radio Communications (any recent edition). The standard non-specialist introduction to RF, antennas, propagation, and interference.

On counter-UAS and electronic warfare basics:

- CSIS Briefs on Counter-UAS: https://www.csis.org , free, policy-level introductions written for non-engineers.

- NATO/STO lecture series on counter-drone EW (available via DTIC): doctrinal foundation without classified details.

- West Point Modern War Institute articles on drone warfare in Ukraine, operational context for why the technical problems described here are urgent.

For going deeper:

- Consensus AI (consensus.app), search any paper title in this report to find related peer-reviewed work instantly.

- Google NotebookLM, load the PDFs from this report’s reference list and ask it to explain concepts, generate quizzes, or produce audio overviews. Particularly effective for the KAIST, Drone-Mag, and ConfuSense papers.

Editorial Commentary: What This Report Argues and What It Does Not

This report makes a bounded claim: that the sensing and defeat of drones via their unintentional electromagnetic emissions is physically plausible, supported by peer-reviewed evidence, and an active area of commercial and academic development. It does not claim that any fielded commercial product currently detects motor/ESC unintentional emissions at tactically useful standoff ranges. That gap, the distance between lab demonstration and operational deployment, is explicitly flagged as the frontier.

The commenter exchange that motivated the report’s framing (Jeremy Hodges, and others) raised a legitimate engineering question: is the detection range claim operationally meaningful? That question remains open. The Watanabe et al. (2021) near-field characterization, the Drone-Mag (2025) identification results, and the OSTI 1/R² decay argument collectively move the answer from ‘unsubstantiated’ to ‘physically grounded and worth serious investigation.’ They do not resolve it.

The MESMER / Department 13 history is included because it demonstrates a broader principle: that the sensing step , passively listening to and decoding a drone’s RF behavior before acting, is non-trivial, militarily significant, and predates most of the academic literature by nearly a decade. Systems built on protocol manipulation (D-Fend, Sentrycs) descend from this lineage.

The fiber-optic drone is included not as an edge case but as the forcing function that makes the entire EMI sensing argument strategically relevant. If RF jamming and C2 link monitoring are the only tools, fiber-optic drones are invisible to them. The unintentional emissions from motors and ESCs remain present regardless of how the drone is controlled. That is the operational gap this body of research is beginning to address.

Version note: This document represents an updated release integrating additions and corrections made in collaboration with AI-assisted research (Perplexity AI) during July 2026. The author retains responsibility for all claims. Errors of taxonomy, attribution, or technical framing should be directed to the author via LinkedIn.

References

[1] Drone Detection System Market Insights Intel Market Research. Global Drone Detection (RF, Radar, EO/IR, Acoustic) System Market valued at USD 1.45 billion in 2025; projected USD 4.20 billion by 2034 at 14.7% CAGR. https://www.intelmarketresearch.com/drone-detection-system-market-44951

[2] Drone-Mag: UAV Identification and Authentication via Electromagnetic Emissions Omar Ibrahim, Roberto Di Pietro. ACM Transactions on Cyber-Physical Systems, 2025. https://dl.acm.org/doi/full/10.1145/3731565

[3] CRI-Lab Accepted Papers , Drone-Mag HBKU CRI-Lab research group page confirming acceptance of Drone-Mag in ACM TCPS. https://cri-lab.net/cri-lab-accepted-papers/

[4] Reliable Digital Forensics in the Air (NSF PAR) Li, Chen, Xu et al. (University at Buffalo / SUNY). NSF Public Access Repository. https://par.nsf.gov/servlets/purl/10399633

[5] Reliable Digital Forensics in the Air: Exploring an RF-based Drone Identification System (DroneTrace) Li, Chen, Xu et al. ACM IMWUT 2022. mmWave parasitic RF response fingerprinting of drone hardware. https://dl.acm.org/doi/10.1145/3534598

[6] Pre-flight Abnormalities Detection on Drone via Deep RF Sensing (PADrone) Hadi, Ramesh, Chan. ACM Transactions on Internet of Things, January 2025. Contactless RF vibration sensing for drone mechanical fault detection. https://dl.acm.org/doi/10.1145/3706121

[7] Predicting EMI in UAVs using Characteristic Mode Analysis Tecnun / Universidad de Navarra. Nature Scientific Reports, March 2025. CMA identifies motor-ESC power wiring as primary EMI vulnerability pathway in DJI Phantom 4. https://www.nature.com/articles/s41598-025-94415-1

[8] Paralyzing Drones via EMI Signal Injection on Sensory Communication Channels Jang, Cho, Kim, Kim (KAIST / Samsung SDS). NDSS Symposium 2023. IMU bus EMI injection crashes any drone instantly at low power, regardless of sensor type. https://www.ndss-symposium.org/ndss-paper/paralyzing-drones-via-emi-signal-injection-on-sensory-communication-channels/

[9] ConfuSense: Sensor Reconfiguration Attacks for Stealthy UAV Manipulation (KIT repository) Erba, Castellanos, Sihag, Zonouz, Tippenhauer. USENIX VehicleSec, January 2025. Stealthy bus message injection reconfigures sensor scale factors, stalls Kalman filter, causes crash. https://publikationen.bibliothek.kit.edu/1000185498/180059606

[10] ConfuSense: Sensor Reconfiguration Attacks for Stealthy UAV Manipulation (author page) Tippenhauer Lab publication page. Supplementary access to ConfuSense paper and cite data. https://tippenhauer.de/publication/erba-2025-confu-sense/

[11] Drone Security and Fault Injection Attacks , IOActive Gabriel Gonzalez, IOActive. EMFI attacks against DJI Mavic Pro; EM glitch during firmware update achieves memory corruption on main processor. https://www.ioactive.com/drone-security-and-fault-injection-attacks-gabriel-gonzalez-ioactive-labs-blog/

[12] Epirus Demos First Directed-Energy Takedown of Fiber-Optic Guided Drone The Defense Post, January 2026. First publicly reported defeat of a fiber-optic guided drone using Leonidas HPM system. https://thedefensepost.com/2026/01/15/epirus-directed-energy-counter-drone/

[13] Epirus’ Leonidas Demos Successful Use of High-Power Microwave to Defeat Fiber-Optic Controlled UAS (ASDNews) ASDNews, January 2026. Coverage of Epirus Leonidas fiber-optic drone defeat demonstration. https://www.asdnews.com/news/defense/2026/01/13/epirus-leonidas-demos-successful-use-highpower-microwave-defeat-fiberoptic-controlled-uas

[14] Epirus’ Leonidas Demonstrates Successful Use of High-Power Microwave to Defeat Fiber-Optic Controlled UAS Epirus Inc. official press release, January 2026. https://www.epirusinc.com/press-releases/epirus-leonidas-demonstrates-successful-use-of-high-power-microwave-to-defeat-fiber-optic-controlled-uas

[15] Aaronia AARTOS Drone Detection System , Product Datasheet Aaronia USA. Technical datasheet for AARTOS DDS; real-time 3D direction-finding across all frequency bands; detection ranges up to 40–50 km. https://aaroniausa.com/wp-content/uploads/2024/02/Aaronia_AARTOS_Drone_Detection_System.pdf

[16] Hybrid RF Detection for Evolving Airspace Threats Dedrone blog, 2025. Describes hybrid library + anomaly-based RF detection architecture for unknown and dark drone detection. https://www.dedrone.com/blog/hybrid-rf-detection-for-evolving-airspace-threats

[17] D-Fend Solutions Launches EnforceAir PLUS Dronelife, August 2025. EnforceAir PLUS integrates AI-enhanced cyber takeover, radar, and optional jamming. https://dronelife.com/2025/08/19/d-fend-solutions-launches-enforceair-plus/

[18] Sentrycs Cyber Over RF Joins Rafael’s Drone Dome System Sentrycs press release. Integration of Sentrycs protocol-level RF analysis into Rafael Drone Dome for multi-layered C-UAS. https://sentrycs.com/news/sentrycs-cyber-over-rf-technology-integrated-into-rafaels-combat-proven-drone-dome-system/

[19] R2 Wireless Raises $5 Million to Expand its RF Spectrum Awareness Platform Electronics Media, March 2026. R2 Wireless $5M raise ($13M total); ODIN passive RF fingerprinting platform. https://www.electronicsmedia.info/2026/03/12/r2-wireless-raises-5-million-to-expand-its-rf-spectrum-awareness-platform/

[20] HADES Passive RF Sensor , Product Page HADES Defense Systems. Passive ESM sensor 0–6 GHz for drone RF signature detection and classification. https://hadesdefense.nl/en/passive-rf/

[21] SRC Expands Gryphon Multi-Mission Radar Family with New Counter-UAS Force Protection Radars ASDNews, May 2025. Gryphon R1430/1440/1540 radar family launch. https://www.asdnews.com/news/defense/2025/05/20/src-expands-gryphon-multimission-radar-family-with-new-counteruas-force-protection-radars

[22] Echodyne Demonstrates Passive-Active Sensor Integration at NTC Echodyne, 2025. EchoShield RDK deployment at 1st Cavalry Division C-UAS exercise, integrating passive RF and active radar. https://www.echodyne.com/resources/news-events/active-passive-integration-at-ntc/

[23] ELVIRA Drone Radar , Product Page Robin Radar Systems. ELVIRA 2D X-band FMCW radar with 360° coverage and micro-Doppler DNN classification for drone detection up to 5 km. https://www.robinradar.com/products/elvira-radar

[24] Netherlands Buys 100 Drone Detection Radars After Recent Sightings Reuters, November 2025. Dutch Ministry of Defence purchases 100 Robin Radar ELVIRA units. https://www.reuters.com/business/aerospace-defense/netherlands-buys-drone-detection-radars-after-recent-sightings-2025-11-27/

[25] HENSOLDT Improves German Armed Forces’ Drone Defence Capabilities , ASUL Capability Upgrade HENSOLDT press release. ASUL C-UAS system upgrade commissioned for Bundeswehr. https://www.hensoldt.net/news/hensoldt-improves-german-armed-forces-drone-defence-capabilities-asul-capability-upgrade-commissioned

[26] Drone Detection With a LTE450-Based Passive Radar Bruno Demissie, Christian Steffes (Fraunhofer FKIE). Passive radar exploiting LTE450 cellular infrastructure as illuminator; motivated by unjammable fiber-optic drone threat. https://publica-rest.fraunhofer.de/server/api/core/bitstreams/7eda759d-5477-4a8e-b6e8-a0377bc4a7c5/content

[27] Protection of Critical Infrastructure Using LTE450-Based Passive Radar Fraunhofer FKIE. Extension of LTE450 passive radar concept to critical infrastructure protection. https://publica.fraunhofer.de/entities/publication/4ae6d656-3b29-47a4-adaa-d999c1ec2ba0

[28] Revealing Hidden IoT Devices through Passive Detection, Fingerprinting, and Localization (RFScan) Wei Sun, Hadi Givehchian, Dinesh Bharadia (UCSD). PoPETs / PETS 2025. Passive unintentional EM emanation detection and angle-of-arrival localization of IoT and drone devices. https://petsymposium.org/popets/2025/popets-2025-0011.php

[29] A Generic Machine Learning Framework for Radio Frequency Fingerprinting Alex Hiles, Bashar Ahmad (BAE Systems Digital Intelligence). arXiv, October 2025. Specific emitter identification (SEI) framework for UAV RF fingerprinting. https://arxiv.org/html/2510.09775v3

[30] Publications with AERPAW , Fusion of Cellular ISAC and Passive RF Sensing for UAV Detection Dickerson, Sichitiu, Guvenc et al. (NC State / AERPAW). IEEE Asilomar Conference, October 2025. Multi-sensor UAV detection on $24M NSF-funded real-world testbed. https://aerpaw.org/publications/

[31] AI-Enhanced Passive Radar: Leveraging 5G Signal Reflectance for Drone Detection Signal Processing Lab, ICS-FORTH (Greece). 3AF Integrated Air and Missile Defence Conference, May 2025. SDR-based 5G reflectance passive radar with deep learning classification. https://spl.ics.forth.gr/vcontent/news-announcements/ai-enhanced-passive-radar-leveraging.html

[32] Sensor Fusion of Radar and Passive RF for Drone Detection (Sandia National Laboratories) Luke Gilbert, Sandia National Laboratories. SAND2025-05117C. Presented April 15–17, 2025. C-UAS sensor fusion test program at Manzano airstrip. https://www.sandia.gov/app/uploads/sites/273/2025/05/5_Deliberate-Motion-Analytics.pdf

[33] SHIELD Activated: Researchers Build a Defense to Protect Drones from Cyberattacks Florida International University news, March 2026. SHIELD hardware anomaly detection system; 0.21s detection, 0.36s recovery from EMI attacks in lab conditions. https://news.fiu.edu/2025/shield-activated-researchers-build-a-defense-to-protect-drones-from-cyberattacks

[34] Defending Unmanned Aerial Vehicles From Attacks on Inertial Sensors (MARS) Haocheng Meng et al. arXiv, April 2025. Model-based Anomaly detection and Recovery System (MARS) as countermeasure to KAIST-style EMI injection attacks. https://arxiv.org/html/2505.00924v1

[35] R2 Wireless , ODIN Platform R2 Wireless corporate site. Passive spectrum-agnostic RF fingerprinting and geolocation; detects encrypted, anonymized, and previously unseen signals. https://r2-wireless.com

March 2026 Prepared by Robi Sen, Cognoscenti Labs, LLC

Table of Contents

1. Visible-Light / Optical Cameras

2. Thermal / IR Cameras

3. Flight Controllers & ESCs

4. Motors & Servos

5. Fiber-Optic FPGAs & Programmable Logic

6. Component Localization

7. Key Takeaways

Author’s Note

At first glance, this report may look like a conventional industry survey: a catalog of cameras, motors, flight controllers, fiber-optic links, and other components. It could even be read as the work of an analyst using component costs and production volumes as proxies for military activity.

That is not its main purpose. The report uses components and supply chains to examine something larger: how Ukraine has transformed wartime necessity into a manufacturing system capable of producing drones at extraordinary scale while repeatedly adapting to shortages, strikes, sanctions, export restrictions, and political pressure.

Ukraine’s most important achievement is not any single inexpensive airframe, novel sensor, or clever battlefield modification. Those accomplishments are impressive, but the deeper innovation is organizational and industrial. Ukraine has created the processes, supplier relationships, production methods, financing mechanisms, technical communities, and political arrangements needed to build roughly five million drones in a year. It has learned how to expand production while continuously redesigning both the aircraft and the system that produces them.

Ukraine still has not achieved a fully Ukrainian drone supply chain. Its public rhetoric often emphasizes localization and independence, yet the industry still relies on a network of Ukrainian firms, Asian manufacturers, Western and Japanese semiconductor companies, foreign raw materials, intermediaries, donors, and allied governments. Some of the sourcing is surprising and worrisome as seen in both legitimate and illegitimate movement of drones supplies through African countries. These apparent contradictions are one of the report’s central points: Ukraine’s growing sovereignty has been achieved not through isolation, but through skillful management of interdependence.

That management is intensely political. Ukrainian manufacturers and officials must compete for funding, persuade allies, secure export-controlled components, cultivate alternative suppliers, and rapidly replace sources that become unavailable. When one component is constrained, the industry may redesign around it, produce it domestically, obtain it through another country, or shift to a different technical architecture. Understanding which parts are made, who makes them, where the underlying materials originate, and how quickly substitutions can occur turns a seemingly dry survey of ESCs, cameras, motors, and batteries into an analysis of national resilience, bargaining power, and vulnerability.

A further complication is that nearly every quantitative claim emerging from Ukraine should be treated as provisional. This is a war, and information management is part of the war. Production totals, localization percentages, company capacity, losses, and future plans may be delayed, selectively disclosed, rounded upward or downward, framed for donors or domestic audiences, or rendered obsolete by events soon after publication. Political advocacy can further distort otherwise useful reporting. The figures in this report are therefore best read as indicators of direction, scale, and industrial structure rather than as audited statistics.

Source quality is uneven. Some defense-industry publications and general news organizations repeat manufacturer claims or government statements without independently testing them, while outlets that do not specialize in electronics, manufacturing, or supply chains may misidentify components or overstate localization. Where possible, this report compares company statements, technical documentation, recovered component analysis, procurement evidence, and reporting from multiple independent sources. Even then, gaps and contradictions remain, but those gaps are telling. The gaps often define connections, origins, coordination by their very absence so they too because part of research.

Executive Summary

Ukraine’s drone industry has undergone a wartime industrial transformation without clear modern precedent. Available public estimates indicate that the country moved from near-total reliance on Chinese imports in 2022 toward a distributed manufacturing ecosystem capable of producing drones on the order of several million per year, with approximately five million frequently cited for 2025. That figure should be understood as a wartime estimate, not an audited production count.

This report examines several of the components that make those systems possible: visible-light and thermal cameras, flight controllers, electronic speed controllers (ESCs), motors, servos, and the programmable logic and optical hardware used in fiber-optic communications. The components matter not only as technical objects, but as evidence of how the broader manufacturing system is changing.

The reported localization trajectory is striking, but necessarily uncertain. The Snake Institute estimates that the Chinese share of components in Ukrainian drones fell from approximately 100% in 2022 to roughly 38% by 2025. Other public descriptions use different definitions of “Ukrainian-made,” often mixing final assembly, first-level components, and upstream materials. The percentages in this report therefore describe an apparent direction of travel rather than a precise census. Critical second-level inputs, including semiconductor chips, battery cells, rare-earth magnets, optical fiber, and specialized sensor materials, remain substantially dependent on foreign suppliers.

The central argument is that Ukraine’s most consequential drone innovation is the production system itself. The country has combined domestic workshops and factories, volunteer networks, private companies, state procurement, Asian components, and Western and Japanese electronics into an industrial base that can expand, fragment, relocate, and redesign under attack. The achievement is not simply the capacity to manufacture drones. It is the capacity to repeatedly reconfigure a multinational supply chain while maintaining battlefield output.

That distinction also helps explain the gap between rhetoric and reality. No single company, ministry, or national supply chain produces the Ukrainian drone effort. It depends on many firms and many countries, even as Ukraine steadily localizes the most vulnerable components. The resulting system is neither fully sovereign nor conventionally globalized. It is a politically managed, rapidly evolving wartime network whose resilience comes from redundancy, substitution, distributed initiative, and the ability to change faster than its constraints.

Image 1: This AI-generated image is based on real photographs of Ukrainians assembling drones in degraded and improvised workshops. Other facilities are modern, highly organized, and exceptionally clean; both environments are part of the same production ecosystem.

1. Visible-Light / Optical Cameras

FPV (First-Person View) cameras are the pilot’s eyes on a drone. In Ukraine’s drone war, camera selection involves critical trade-offs between latency, image quality, weight, and cost. Analog cameras dominate kamikaze FPV drones due to their ultra-low latency (under 10 ms), lightweight form factor, and low cost, a pilot on a one-way mission needs real-time video more than recording quality. Digital systems like DJI’s O3 and O4 units provide superior image quality and longer range, making them preferred for reconnaissance and reusable platforms.

Table 1A: FPV Cameras, Analog (Small Kamikaze FPV Drones)

Table 1B: FPV Cameras, Digital Systems

Table 1C: Large UAV Camera / Gimbal Systems

Analog vs. digital for FPV. The fundamental trade-off in FPV drone cameras is latency vs. quality. Analog systems transmit raw video signal with minimal processing, achieving latencies under 10 milliseconds, critical for a pilot guiding a kamikaze drone to a moving target at 100+ km/h. Digital systems like DJI’s O3 unit encode video using H.264/H.265 compression, adding ~30 ms of latency but providing far superior image clarity and range.

Why analog dominates kamikaze drones. A kamikaze drone is expendable, every component is destroyed on impact. At roughly $15–25 per camera, analog units like the Caddx Ant Nano (2 grams) keep costs low and weight minimal. The pilot needs split-second responsiveness during the terminal approach, not 4K footage.

Night vision capability. The RunCam Night Eagle 3 stands out with a minimum illumination of 0.0001 Lux, effectively allowing pilots to fly in near-total darkness using only ambient starlight or moonlight. This has proven invaluable for nighttime operations, where drones operate with reduced risk of visual detection. The 1/1.8-inch BSI sensor collects significantly more light than smaller sensors, though at a slight cost in resolution (1000 TVL vs. 1200 TVL for daytime cameras).

Mid-size reconnaissance drones. Between small FPV kamikazes and large military UAVs sits a growing class of medium drones. Ukrainian alternatives to the DJI Mavic include the Shmavic (by Reactive Drone), Yautja (Rise Technologies), and Ukropter, all larger than the Mavic, with the key advantage of 300+ flight cycles versus roughly 60 for DJI platforms. Fiber-optic FPV drones in this size class typically use the same analog cameras as standard FPVs (Caddx or RunCam), feeding into the fiber-optic media converter for jam-resistant transmission.

Localization push. Ukrainian firms are actively working to reduce camera import dependence. Odd Systems produces analog cameras domestically. UDD Tech Corp has acquired camera technology and plans European manufacture with components sourced from another Ukrainian firm. Wild Hornets has developed its proprietary “Hornet Vision” digital system for its Wally drone models.

2. Thermal / IR Cameras

Thermal cameras have become indispensable in the Ukraine conflict, enabling drones to detect personnel, vehicles, and other drones at night and through camouflage. The Ukrainian company Odd Systems has emerged as a critical domestic producer with its Kurbas line of thermal cameras, designed specifically for combat FPV drones rather than repurposed industrial sensors.

Table 2A: Odd Systems Kurbas Line (Ukrainian-Made)

Table 2B: Imported Thermal Cameras Used in Ukraine

Odd Systems factory plans. Odd Systems is building a factory to produce thermal imaging camera sensors, not just cameras but the underlying microbolometer detectors. CEO Yaroslav Azhniuk announced at the “War 2026” conference that the planned capacity is 1 million sensors per year, which would triple the entire European continent’s thermal sensor production capacity. The factory location is still under consideration (Ukraine or abroad).

Additionally, Odd Systems is developing a semiconductor fabrication facility for drone camera sensors, with a US location being considered. This represents an extraordinary ambition, moving from camera assembly to chip fabrication in the span of a few years, driven by wartime necessity.

TFL Anti-Shahed integration. The Kurbas-640a is integrated into the TFL Anti-Shahed detection module, used on interceptor drones like Wild Hornets’ STING UAV. The complete system costs $800 with camera or $300 for the standalone detection module, enabling autonomous drone-on-drone interception of Iranian Shahed-type loitering munitions.

3. Flight Controllers & ESCs

The flight controller (FC) is the “brain” of a drone, a small circuit board running firmware (Betaflight or iNav) that reads sensor data from gyroscopes and accelerometers, processes pilot inputs, and adjusts motor speeds thousands of times per second to keep the drone stable and responsive. ESCs (Electronic Speed Controllers) translate the FC’s commands into the precise electrical signals that spin each motor at the required speed.

Table 3A: Flight Controllers

Table 3B: Key Chips Inside Flight Controllers

Domestic ESC production

Why flight controllers are hardest to localize. The FC is the most technically complex component in an FPV drone. It requires precision SMD assembly, quality control for sensor calibration, and firmware expertise. Wild Hornets achieved a breakthrough by establishing a robotic FC assembly line in Ukraine, becoming the first domestic producer and reducing costs by 12%.

STM32 chips, not a China dependency. A common misconception is that FC chips come from China. In fact, STM32 microcontrollers are manufactured by STMicroelectronics, a European (French Italian) company. The IMU sensors come from TDK InvenSense (Japan/US) or Bosch (Germany). These are commodity semiconductors available through Western supply chains, reducing a key vulnerability in the localization effort.

BLHeli_S vs BLHeli_32. ESC firmware determines motor control precision. BLHeli_S runs on simpler 8-bit MCUs (like the Silicon Labs EFM8BB21), is cheaper, and sufficient for most FPV applications. BLHeli_32 runs on 32-bit ARM MCUs, offering smoother throttle response, telemetry, and more advanced features, preferred for heavier or precision platforms.

Key ESC chips. The Silicon Labs EFM8BB21 is the dominant MCU in BLHeli_S ESCs,an 8-bit microcontroller that handles the fast switching required for brushless motor commutation. For BLHeli_32, the STM32F051 or STM32G071 provides a 32-bit ARM Cortex-M0 processor with more computational headroom for advanced motor control algorithms. Some Chinese-origin ESCs use the GD32E230 (a GigaDevice clone of the STM32), which is functionally equivalent but may have different supply chain implications. The power stage uses discrete MOSFETs for switching the high currents (40–60A) required by FPV motors.

4. Motors & Servos

Brushless DC motors are the propulsion system of every FPV drone. A typical 5-inch kamikaze FPV uses four 2207-size motors running at 1750–2450 KV on 4S–6S LiPo batteries. Ukraine’s demand for drone motors is staggering: producing 5 million drones in 2025 required approximately 20 million motors.

Table 4A: Drone Motors

Brushless DC motors are the propulsion system of every FPV drone. A typical 5-inch kamikaze FPV uses four 2207-size motors running at 1750–2450 KV on 4S–6S LiPo batteries. Ukraine’s demand for drone motors is staggering producing 5 million drones in 2025 required approximately 20 million motors.

Various

Motor-G: Ukraine’s motor champion. Launched in December 2024, Motor-G has rapidly scaled to produce approximately 100,000 motors per month under CEO Oleksiy Grebin, likely making it the largest drone motor producer in Europe. Motor-G supplies FPV and fiber-optic drone motors (up to 15-inch frame sizes). However, it still depends on Chinese-sourced neodymium-iron-boron magnets, a critical bottleneck since 90% of global sintered rare-earth magnet output comes from China.

Servos

Standard FPV quadcopter drones do not use servos, they achieve flight control entirely through variable-speed motor control via ESCs. Fixed-wing drones (such as the PD-1, PD-2, Leleka-100) and larger UAVs use conventional hobby servos for control surfaces (ailerons, elevators, rudders). Common servo brands include Futaba (Japan), Hitec (South Korea), and Savox (Taiwan). No evidence of Ukrainian-made servo production has been identified.

5. Fiber-Optic FPGAs & Programmable Logic

Image 2: AI generated image of fiber optic tethered drone assembly.

Fiber-optic drones have transformed Ukraine’s battlefield by providing jam-resistant, zero-latency video and control links that defeat Russian electronic warfare. At the heart of every fiber-optic drone’s communication system is a media converter, a small circuit board that translates between electrical signals (video, control) and optical signals transmitted through a thin fiber-optic cable trailing behind the drone. Virtually all fiber-optic drones use programmable logic (FPGAs) in their media converters.

Table 5A: FPGA in Fiber-Optic Media Converters

Table 5B: Fiber-Optic Module Options (Commercial)

How the Media Converter Works

Sky End (TX, on the drone): Camera analog video signal → ADC (Analog-to-Digital Converter) → FPGA (Spartan-6) packages video + UART control data using 8b/10b encoding → SFP fiber optic transceiver module → optical signal travels down singlemode fiber trailing behind the drone.

Ground End (RX, at the pilot station): SFP fiber optic transceiver receives optical signal → FPGA (Spartan-6) deserializes and separates video + control data → R-2R resistor ladder DAC converts digital video back to analog → output to pilot’s FPV goggles. Control data is routed to the RC transmitter for command feedback.

FPGA adoption is universal. Every fiber-optic drone uses some form of programmable logic (FPGA) in its media converter, there is no alternative at this point for the real-time video + control multiplexing required. The Xilinx Spartan-6 (now AMD) is the most identified FPGA in recovered and analyzed fiber drone converters from Ukraine.

Ukrainian developments. Ukrainian companies are beginning to design their own converter boards, potentially using Lattice or Xilinx FPGAs. More advanced designs may incorporate SoC (System-on-Chip) approaches combining a processor and FPGA on one chip. H.264 video compression is being explored (using open-source implementations like hard264) to enable higher-resolution video over the same fiber link.

Fiber-optic spool production. 3DTech produces 30 km fiber-optic spools domestically using 3D-printed spool housings. Hasta similarly 3D-prints spools and casings using Ukrainian-made lathes. However, the raw optical fiber itself still comes from China (Hubei province, via intermediaries), no Ukrainian manufacturer has yet mastered the process to produce optical fiber from scratch.

6. Component Localization

Image 3: AI image speaks to Ukraines strategy of buying special components like integrated chips from countries like South Korea, Japan, and Taiwan.

Ukraine’s drone industry has undergone a remarkable transformation. In 2022, virtually 100% of FPV drone components were imported from China. By early 2026, the Chinese component share has dropped to approximately 38%, according to the Snake Institute think tank. First-level component localization (frames, propellers, assembly) has reached 80–90% Ukrainian, though second-level components (semiconductor chips, battery cells, raw materials) remain largely imported.

The table below highlights the main Ukrainian production nodes and the key external component sources that still shape what is possible. It shows where local capacity now exists, where Asian materials and COTS electronics remain indispensable, and where US and European firms sit quietly inside the supply chain despite the rhetoric of decoupling.

Main Table: Ukrainian Capability vs. External Choke-Points

Table 6A:Ukranian Component Manufacturers

Table 6B: Ukrainian Component Manufacturers

The trajectory: from 100% Chinese to 38%. The speed of Ukraine’s de-Sinification is unprecedented in modern defense manufacturing. Driven by existential wartime pressure, Ukrainian companies have vertically integrated in ways that peacetime economies rarely attempt. The key enabler has been the realization that many “Chinese” components use Western or Japanese semiconductors, STM32 chips from STMicroelectronics (EU), IMU sensors from TDK (Japan) and Bosch (Germany),that can be sourced directly.

The 5M drones = 20M motors equation. Ukraine produced approximately 5 million drones in 2025. Each quadcopter requires four motors, meaning 20 million motors were needed, a demand that no single Ukrainian manufacturer can yet meet alone. Motor-G’s 100,000/month capacity (1.2 million/year) covers only about 6% of annual demand, highlighting the continued need for imports alongside domestic scaling.

Odd Systems’ semiconductor ambition. Perhaps the most audacious localization effort is Odd Systems’ plan to build a semiconductor fabrication facility for drone camera sensors, with a US location under consideration. If realized, this would represent Ukraine’s first entry into chip manufacturing, a strategic move to secure the most critical link in the drone supply chain against both Chinese dependency and wartime disruption.

7. Key Takeaways

· Frames and propellers are fully localized (~90%+), with multiple Ukrainian manufacturers producing carbon-fiber frames at scale. Raw carbon fiber material remains partially imported.

· Motors are rapidly localizing (60–70%) thanks to Motor-G’s ~100K/month output, but China’s 90% share of rare-earth magnets remains an irreducible dependency in the near term.

· Flight controllers have crossed the 50% localization threshold, with Wild Hornets’ robotic assembly line proving that domestic FC production is viable. STM32 chips are Western-made commodities, not a China dependency.

· Cameras remain the most import-dependent category for visible light (30–40% localized), though Odd Systems has achieved strong localization for thermal cameras (60–70%). Analog cameras dominate FPV kamikaze drones; digital systems are used for recon platforms.

· Every fiber-optic drone uses an FPGA in its media converter. The Xilinx Spartan-6 is the most common. Fiber media converters are the least localized component category (20–30%), representing a critical bottleneck.

· Odd Systems’ 1M-sensor/year factory would triple European thermal sensor production, and their semiconductor fab plans could secure the supply chain at the chip level.

· Chinese component share dropped from ~100% (2022) to ~38% (2025), an extraordinary de-sinification driven by wartime urgency. Second-level components (chips, cells, raw materials) remain the frontier of localization.

· Critical bottlenecks remaining rare-earth magnets (motors), lithium cells (batteries), raw optical fiber (fiber-optic spools), FPGA chips (media converters), and advanced camera sensors (visible light). These represent the hardest-to-localize links in the supply chain.

Supply chain resilience. The overall picture that emerges is one of a drone industry rapidly building resilience against supply chain disruption. The components that are easiest to localize (mechanical: frames, propellers) are already 90%+ Ukrainian. The medium-difficulty components (motors, flight controllers, thermal cameras) are progressing at 50–70%. The hardest components, visible-light camera sensors, fiber-optic media converters, lithium battery cells, and FPGA chips, remain below 40% localized and represent the next frontier. The key insight is that many “Chinese” components actually use Western or Japanese semiconductors (STM32 from STMicroelectronics, IMU sensors from TDK/Bosch), which can be sourced independently of China.

Looking ahead. The most transformative development on the horizon is Odd Systems’ plan to build semiconductor fabrication capacity for thermal sensor production. If successful, this would mark Ukraine’s first entry into chip manufacturing, a strategic leap that no other nation at war has attempted. Combined with Motor-G’s rapid motor production scaling, Wild Hornets’ flight controller localization, and 3DTech’s fiber-optic spool production, Ukraine is building a vertically integrated drone industry that could serve as a model for allied nations seeking to reduce their own dependencies on Chinese drone components.

Methodological note. This report synthesizes publicly available information from manufacturer documentation, technical reporting, defense-industry publications, recovered-component analysis, think tanks, government statements, and general news coverage available as of March 2026. These sources vary greatly in technical competence, independence, and access. Many repeat claims that cannot be independently verified, and some reporting is shaped by wartime information operations, fundraising, procurement politics, advocacy, or commercial promotion. Production totals, localization percentages, capacity figures, and announced factory plans should therefore be treated as provisional estimates. They are most useful for identifying trends, dependencies, and the speed of industrial change, not for establishing exact audited quantities.

References

1. VGI.com.ua – Ukraine Drone Industry Report (Dec 2025).
https://vgi.com.ua/en/reports/ukraine-drone-industry-2025

2. Forbes / David Hambling – Ukraine Drone Components (Apr 2025).
https://www.forbes.com/sites/davidhambling/2025/04/15/ukraine-drone-components/

3. Kyiv Independent – Ukraine Fiber-Optic Drones (Aug 2025).
https://kyivindependent.com/ukraine-fiber-optic-drones-2025/

4. Militarnyi – Odd Systems Thermal Sensor Factory (Feb 2026).
https://mil.in.ua/en/news/odd-systems-thermal-sensor-factory-2026/

5. Defense Daily – Odd Systems Semiconductor Fab Plans (Mar 2026).
https://www.defensedaily.com/ukraine-odd-systems-semiconductor-fab

6. New York Times – Ukraine Drone Production (Mar 2026).
https://www.nytimes.com/2026/03/15/world/europe/ukraine-drone-production.html

7. CSIS – Rare Earth Supply Chains (Sep 2025).
https://www.csis.org/analysis/rare-earth-supply-chains-2025

8. Ukrainska Pravda – TFL Anti-Shahed System (Dec 2025).
https://www.pravda.com.ua/eng/news/2025/12/anti-shahed-tfl/

9. RCDrone – Thermal Camera Comparison.
https://www.rcdrone.top/thermal-camera-fpv-comparison/

10. Primeda-Telecom – Fiber-Optic FPV Modules.
https://www.primeda-telecom.com/fiber-optic-fpv-drone-modules/

11. Army Technology – PD-2 UAV.
https://www.army-technology.com/projects/pd-2-uav/

12. Airforce Technology – Bayraktar TB2.
https://www.airforce-technology.com/projects/bayraktar-tb2/

13. Odd Systems – Kurbas Camera Documentation.
https://docs.oddsystems.io/kurbas/

Cognoscenti Labs, LLC

2026/07/09

Perception warfare is the deliberate shaping, degradation, or exploitation of how an adversary senses, interprets, and acts on the world through human operators, software, and machine perception stacks. In practice, it sits at the intersection of camouflage, deception, electronic warfare, cyber operations, optical interference, and adversarial machine learning, with the goal of changing what a target system believes to be true rather than simply destroying it.[1][2][3][4][5]

Overview

image 1: Optical emitters project structured light creating shapes in the air that drones turn away from seeing their wall.

A new conflict layer is emerging in which the decisive effect is not always physical destruction, but the manipulation of perception itself. This includes methods that deceive, confuse, disrupt, degrade, compromise, and, as part of broader attack chains, help enable the loss of reliable control over vehicles, sensors, or computers by corrupting the signals and data on which they rely.[2][4][5][6][1]

Although deception and camouflage are ancient practices, the widespread use of machine vision, autonomy, sensor fusion, and learning systems has created a new operational environment in which small physical changes can produce disproportionately large effects. In recent years, adversarial patches, synthetic camouflage, multispectral deception systems, and projector-based attacks have moved from academic demonstration toward practical defense relevance, including visible discussion tied to battlefield conditions in Ukraine and broader counter-autonomy programs.[7][8][9][10][2]

This matters because optical and sensor-directed effects are no longer only laboratory curiosities. Dazzlers, passive deception systems(in use for decades), structured light concepts, and adversarial examples all point toward the same larger trend: modern conflict is beginning to include deliberate attacks against the sensory and interpretive layers of autonomous and semi-autonomous systems.[2][4][5][7][21]

What perception warfare means

Perception warfare can be understood as competition over the adversary’s sensing and interpretation pipeline. That pipeline may include cameras, infrared imagers, LiDAR, radar, RF receivers, inertial sensors, edge processors, navigation software, operator displays, and the command systems that consume those outputs.[11][12][13][1][2]

The central idea is straightforward: if a system can be made to see the wrong thing, fail to see the right thing, or act on corrupted confidence estimates, then the resulting effect can be operationally equivalent to jamming, sabotage, or defeat. In autonomous or semi-autonomous systems, perception failure can cascade into navigation drift, route replanning, target misclassification, sensor handoff failure, or unsafe control actions.[5][8][14][15][16]

Perception manipulation

Perception manipulation is the application layer of perception warfare: it is how the effect is actually produced. Methods can be passive, active, or hybrid.[1]

Passive methods

Passive methods change the appearance or signature of an object, scene, or platform without radiating energy during use. Examples include adversarial patches, adaptive camouflage patterns, multispectral panels, geometric decoys, thermal shaping, and materials designed to bias machine classifiers while preserving acceptable human-visible camouflage.[4][17][18][19]

This category matters because passive methods scale well, can be low cost, and can remain in place for extended periods. Kallisto Shield, for example, is publicly described as a passive camouflage and deception kit intended to affect visual, infrared, thermal, radar, multispectral, SAR, and hyperspectral detection of military assets.[17][20][4]

Active methods

Active methods radiate or project energy to alter what a sensor perceives. These include optical dazzlers, structured light, projector-based adversarial attacks, spoofing, temporal modulation, false target injection, and dynamic decoys across optical, infrared, RF, or other bands.[21][22][23][24][5]

Active systems can be more adaptive than passive ones because they can respond to geometry, range, ambient conditions, and sensor behavior in real time. That makes them particularly relevant for camera-guided and autonomous vehicles, where carefully timed or spatially coded emissions may degrade detection, depth estimation, tracking, or control.[22][23][5][21]

Hybrid methods

Hybrid methods combine passive structure with active control. A platform may carry an adversarial physical surface, a thermal signature management layer, and a projector or LED array that emits scene-dependent patterns only when needed. This is where structured light, optical fuzzing, and adaptive multispectral deception begin to converge into a coherent engineering discipline.[15][25][26][5]

Adversarial machine learning

Adversarial machine learning is the study and exploitation of the brittleness of learning systems under intentionally crafted inputs. In the perception domain, this usually means manipulating what a model sees so that it misclassifies, fails to detect, localizes incorrectly, or produces unstable outputs under conditions where a human may see nothing unusual.[8][27][7][1]

This matters because modern vehicles, drones, and sensor systems increasingly depend on deep neural networks or learned subsystems for detection, segmentation, depth estimation, tracking, decision support, and sensor fusion. If those learned components can be predictably biased, then the attacker can influence downstream behavior without needing to destroy hardware.[14][26][28][1]

The field is now mature enough to support a useful taxonomy:

· Evasion attacks alter inputs at inference time to change model outputs.[7][1]

· Physical attacks embed those perturbations in real objects, lighting, textures, or projected patterns.[23][8][21]

· Transfer attacks target one model and exploit partial generalization to similar models.[8][7]

· Robustness and detection methods attempt to identify or resist those manipulations.[29][14]

These are not only technical curiosities. They are ways to produce real-world effects against systems that drive, fly, aim, classify, or cue other weapons and sensors.[2][14]

From deception to compromise and control

The most important strategic point is that perception manipulation is not limited to concealment. It can be part of a broader chain that progresses from deception to disruption, then to compromise, and, under some conditions, to functional takeover or loss of reliable control.[6][5][1][2]

A navigation or autonomy stack often depends on camera-derived or fused perception outputs to maintain position, avoid obstacles, classify scenes, select routes, track targets, or authorize actions. If those inputs are persistently corrupted, the system may drift, halt, re-route, mis-prioritize, expose itself, or hand control to fallback logic that may itself be vulnerable or poorly suited to the situation.[16][5][11][14][15]

In that sense, perception warfare overlaps with cyber and electronic warfare even when the initial attack surface is optical or physical. The attacker does not need to break encryption or penetrate a network first; inducing machine error at the sensor or model layer can still create mission-level compromise.[5][1][2][8]

Optical fuzzing and structured light

The optical fuzzer concept is a strong example of this evolution. The disclosed method trains a machine-learning system to adapt light transmitted from an array of emitters so that it affects image processing and navigation control in a camera-based navigation system. That framing is important because it treats light not merely as illumination or obscuration, but as a programmable attack surface against autonomy.[6][5]

Structured light broadens that concept further. Traditionally used for metrology, depth sensing, and imaging, structured light can also be used offensively or defensively to inject false geometric cues, degrade obstacle maps, distort optical flow, and interfere with depth estimation or classification.[26][11][15][22][23]

Projector-based adversarial attacks have already shown that transient projected patterns can cause object detectors and depth-estimation systems to fail under real-world conditions. A logical next step is the move from attacking a single object surface toward shaping larger perceptual volumes with multiple projectors or emitters, affecting the space through which a drone or vehicle observes or moves.[15][21][22][23][26]

Ukraine and the last decade of visible change

Over roughly the last decade, the public record has shifted from mostly theoretical discussion to a growing body of practical systems and battlefield-adjacent experimentation. Public reporting and company disclosures show passive deception systems, synthetic-data-driven evaluation of camouflage, projector-based attacks, and broader concern over AI-guided reconnaissance and targeting.[10][4][23][2][7]

Ukraine has been especially important as a forcing function because dense drone use, persistent ISR, rapid adaptation, and the need for low-cost battlefield asymmetries reward anything that can confuse, hide from, or redirect machine-enabled sensing. This does not mean every method is mature or decisively proven, but it does mean the paradigm has escaped the lab and entered the operational imagination of states, startups, and defense planners.[9][4][10][2]

Why this is a new conflict paradigm

The broader significance is that conflict is moving toward competition over cognitive and machine-mediated interpretation. In earlier eras, concealment was often about defeating a human observer. In the current era, it must increasingly defeat data pipelines, detection models, autonomy stacks, sensor fusion logic, and operators interpreting machine-curated displays.[4][17][26][1][2]

That creates several asymmetries:

· Small, low-cost changes can create outsized effects if they hit a brittle learned subsystem.[7][8]

· Effects can be reversible, deniable, and hard to attribute because they may leave no obvious physical damage.[23][1]

· Tools can scale quickly from software to materials, panels, projectors, emitters, and decoys.[25][21][4]

· The same techniques can support offense, defense, deception, testing, and hardening.[30][14]

In practical terms, the side that becomes proficient at augmenting operations with perception manipulation will gain advantages in survivability, target protection, denial, confusion, and autonomy disruption.[2][4]

What this means for the United States

The United States has strong underlying capabilities in machine learning, optics, microelectronics, sensing, EW, autonomy, and defense integration, but that does not automatically translate into leadership in perception warfare. Leadership will require treating perception manipulation as a serious discipline rather than a niche curiosity.[3][30][1][2]

That implies several lines of effort:

· Build doctrine and threat models for perception warfare across services and mission sets.[30][1]

· Fund offensive and defensive R&D in passive, active, and hybrid counter-perception methods.[25][26][4]

· Develop realistic test and evaluation infrastructure for adversarial perception, including synthetic environments and live hardware-in-the-loop ranges.[10][14][29]

· Integrate optics, structured light, RF deception, cyber, and autonomy research instead of treating them as separate stovepipes.[26][5][2]

· Harden fielded perception stacks while building indigenous capability to impose perception effects on adversaries.[14][30]

The strategic argument is simple: this field is growing, the barriers to experimentation are falling, and the military value of perception effects is increasing. If the United States does not lead, others will shape the doctrine, tools, and norms of a conflict space that is already forming.[4][30][2]

Closing assessment

Perception warfare is best understood as the organized contest to shape what machines and people believe they are seeing, and what they do next because of it. Perception manipulation and adversarial machine learning are not separate from that contest; they are among its most important mechanisms.[1][2][7]

Used well, these methods can hide assets, degrade detection, confuse tracking, redirect navigation, compromise autonomy, and create pathways to broader system disruption. They have existed in quieter forms for years, but recent advances in AI-enabled sensing, autonomy, and battlefield adaptation suggest that a new paradigm of conflict is emerging faster than many institutions are prepared to recognize.[9][22][5][10][2][4]

As these systems become more robust and more deeply integrated into regular warfare, the manipulation of human and machine sensorium will become both more important and more contested. Early effects may appear simple, but capable adversaries will adapt. The more durable advantage will come from combining perception effects with terrain, camouflage, sensor fusion, autonomy, EW, and conventional operations rather than treating them as isolated tricks.

Future systems may use terrain, surfaces, light, materials, and dynamic signatures to create false cues or deny reliable interpretation across a battlespace. In that environment, perception warfare will not merely deceive sensors. It will shape decision cycles, platform behavior, and the balance of tactical advantage in increasingly automated battlespaces.[1][2]

References

1. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-2e2023.pdf

2. https://techinsightdaily.com/982/adversarial-intelligence-how-counter-drone-ai-uses-computer-vision-attacks-against-autonomous-weapons/

3. https://nvlpubs.nist.gov/nistpubs/ai/NIST.AI.100-2e2025.pdf

4. https://defence-blog.com/spanish-startup-creates-camouflage-for-ai-battlefield/

5. https://patents.google.com/patent/US12315233B1/en

6. https://patents.justia.com/patent/20250285421

7. https://arxiv.org/pdf/1911.05268.pdf

8. https://beerkay.github.io/cs529/content/papers/yuan.pdf

9. https://nextgendefense.com/kallisto-shield-camouflage-system/

10. https://www.linkedin.com/pulse/press-release-evaluation-camouflage-concealment-deception-systems-i6zkf

11. https://www.archivesoftransport.com/index.php/aot/article/view/238

12. https://www.terma.com/products/radars/3d-drone-detection/

13. https://www.spotterglobal.com/drone-detection-ax-series

14. https://dl.acm.org/doi/10.1145/3450267.3450535

15. https://www.ecva.net/papers/eccv_2020/papers_ECCV/papers/123680732.pdf

16. https://arxiv.org/abs/2311.12722

17. https://publica.fraunhofer.de/entities/publication/48da286d-c6a4-406e-a473-35273f347ef8

18. https://appl.ai/projects/adversarialai

19. https://www.mdpi.com/1424-8220/26/6/1895

20. https://www.kallistoshield.eu

21. https://arxiv.org/pdf/2209.09652.pdf

22. https://arxiv.org/pdf/2512.24792.pdf

23. https://www.themoonlight.io/en/review/transient-adversarial-3d-projection-attacks-on-object-detection-in-autonomous-driving

24. https://www.unmannedsystemstechnology.com/2016/09/invisage-launches-micro-lidar-module-for-autonomous-drone-flight/

25. https://www.swri.org/what-we-do/internal-research-development/2021/defense-security/enhanced-adaptive-camouflage-using-generative-adversarial-networks-10-r6136

26. https://elight.springeropen.com/articles/10.1186/s43593-025-00102-z

27. https://pmc.ncbi.nlm.nih.gov/articles/PMC10427626/

28. https://www.sciencedirect.com/science/article/pii/S2214212622002423

29. https://umu.diva-portal.org/smash/get/diva2:2053102/FULLTEXT01.pdf

30. https://www.energy.gov/topics/artificial-intelligence

This list was compiled for people looking for sources on a topic where solid, public information is sparse. Where a claim traces back only to the vendor’s own site (Dedrone, DroneShield, etc.) and could not be cross-checked against a second, independent source, it is marked [unverified]. Those claims are likely accurate, since they come from the company describing its own customers, but they have not been confirmed independently and should be validated before being cited in a paper, pitch, or research document. It is also hard to know the full scope of adoption, since defense and government use of this technology is often sensitive or classified and not publicly disclosed.

The customer base splits into four broad verticals.

Defense & Military

The heaviest users by spending. Militaries are natural adopters because EMI/RF sensing works passively, without revealing the operator’s position.

• U.S. Department of Defense — Dedrone deployed at multiple bases after a 56-day pilot at Joint Base Myer-Henderson Hall revealed 95 previously undetected UAS incursions. (Dedrone) DroneShield surpassed 4,000 systems sold globally, including repeat DoD contracts. (DroneShield)

• Ukrainian Ministry of Defense — DroneShield’s RfPatrol was supplied under a $10.4M contract in 2023 through Australia’s Ukraine counter-drone aid package. (DroneShield) Dedrone’s operational presence in Ukraine is [unverified] — remove or source before use.

• Australian Defence Force — DroneShield RfPatrol under a $5M Land 156 Line of Effort 2 order. (DroneShield)

• Bundeswehr (Germany) — HENSOLDT’s ASUL C-UAS system, capability upgrade commissioned by BAAINBw in May 2025. (HENSOLDT) Note: ASUL is a multi-sensor system (active/passive radar, electro-optical, RF direction-finding), not RF/EMI-only, so it’s a partial fit for this list.

• European militaries — DroneShield’s all-time record contract, a €61.6M European defense deal, mid-2025. (DroneShield)

Critical Infrastructure

Power grids, airports, ports, and telecom infrastructure operators are major adopters of passive RF sensing because they need persistent, non-disruptive monitoring.

• Con Edison (NY utility), Korea Power Exchange — Dedrone deployments. (Dedrone) [unverified]

• Chicago O’Hare & Midway Airports — Dedrone. (Dedrone) [unverified]

• Newcastle International Airport — Dedrone. (Dedrone) [unverified]

• Port of Hamburg — Rheinmetall and Deutsche Telekom partnership, using passive RF sensors mounted on cell towers, announced May 2026, building on a December 2025 agreement with Hamburg Police and the Hamburg Port Authority. (Rheinmetall)

• Deutsche Telekom — acts as both integrator and infrastructure provider, deploying RF sensors across German customers; RF sensors detect over 90% of drones in Telekom’s operational experience. (Euronews)

Public Safety, Law Enforcement & Corrections

Prisons and police are among the fastest-growing segments for passive RF sensing, largely driven by contraband drone smuggling.

• 133+ correctional facilities worldwide — protected by Dedrone, including South Carolina DOC, Fulton County Jail, Maine DOC, and JVA Halle Prison (Germany). (Dedrone) [unverified]

• 187 public safety agencies — including Virginia State Police, New Jersey State Police, Los Angeles County Fire, Scottsdale PD, and multiple county sheriff offices, via Dedrone. (Dedrone) [unverified]

• Catalonian Police (Barcelona), Vienna Police, Latvia State Police, Australian Police — Dedrone. (Dedrone) [unverified]

Major Events & Entertainment

High-profile events with no-fly zones are a key use case, since jamming is prohibited in most jurisdictions, making passive RF sensing plus protocol-based mitigation the only compliant option.

• 2026 FIFA World Cup — Sentrycs (Cyber-over-RF, non-jamming) contracted for airspace security across “most venues” in a number of the 16 host cities in the US, Canada, and Mexico. (Ondas) Note: coverage of all 16 cities is not confirmed; one Sentrycs executive cited roughly 70% of US host states in an interview. (DroneLife)

• 2024 UEFA Euro Championship (Germany) — Deutsche Telekom RF detection for police. (Rheinmetall) [unverified]

• Formula 1 (Austin/COTA), NASCAR, WM Phoenix Open — Dedrone. (Dedrone) [unverified]

• Super Bowl LIX — Dedrone. (Dedrone) [unverified]

• World Economic Forum / Davos — Dedrone (Swiss Police). (Dedrone) [unverified]

• Marine Corps Marathon, Preakness Stakes, Tulsa State Fair — Dedrone. (Dedrone) [unverified]

Notable Integrator/Platform Companies Using RF as a Sub-Component

These aren’t end users, but firms that integrate EMI/RF sensing into larger systems.

• Rafael (Israel) — integrated Sentrycs Cyber-over-RF into the Drone Dome system in 2025. (Sentrycs) [unverified]

• Echodyne — partners with R2 Wireless ODIN for passive RF plus radar fusion. [unverified]

• D-Fend EnforceAir PLUS — integrates Echodyne radar with RF cyber takeover. [unverified]

• Rheinmetall — partnering with Telekom to productize RF, radar, and cyber for German KRITIS protection. (Rheinmetall)

Bottom line: the biggest confirmed buyers of passive RF/EMI drone sensing are the US DoD, European militaries, airports, utilities, correctional facilities, and major event organizers. Dedrone (now part of Axon) claims the largest deployment footprint at 1,130+ sites [unverified]; DroneShield leads in confirmed military handheld sales at 4,000+ units; Sentrycs/Ondas is positioning itself as the lead in event and civilian non-jamming protocol RF, with the 2026 FIFA World Cup as its signature deployment (though full 16-city coverage is not yet confirmed).

A Practical Overview for Public Safety, Non-Critical-Infrastructure Commercial Entities, and Resilience Applications

This brief provides a practical overview of passive sensing technologies and how they can be applied across civilian and commercial settings. The intended audience spans Public Safety organizations and extends to commercial entities not traditionally classified as critical infrastructure: manufacturing plants, data centers, logistics hubs, corporate campuses, houses of worship, healthcare facilities, and public venues. These organizations face many of the same sensing challenges as utilities and airports, but often lack the budget, staff, or regulatory guidance that critical infrastructure operators receive.

The focus here is not military operations, offensive collection, or intrusive surveillance. It is safety, resilience, early warning, situational awareness, and responsible monitoring. Passive sensors support security teams, facility operators, emergency managers, and infrastructure owners by detecting anomalies, confirming events, and helping prioritizing response. One of the most important, and often hardest, parts of any project is finding practitioners who understand both the technology and the legal, operational, and privacy constraints around it.

Why Passive Sensing Matters

Modern facilities face a wider range of risks than traditional perimeter security was designed to handle. A county emergency manager, a mid-size manufacturer, a data center operator, or a regional transit authority may all need to detect unauthorized drone activity, unusual RF signatures, acoustic anomalies, or unexpected vehicle movement without a large security staff or a classified budget.

Passive sensing fills this gap by providing layered awareness without relying solely on cameras, guards, alarms, or active radar-like systems. Properly deployed, passive sensors help answer basic operational questions: What changed? Where did it happen? How confident are we? Is this isolated or part of a pattern? Should a person be alerted, another sensor be cued, or should the event simply be logged?

For Public Safety organizations, passive sensors support mission continuity and inter-agency coordination. For commercial entities outside traditional critical infrastructure, they provide an affordable and legally defensible path toward meaningful situational awareness.

Relevant Background and Perspective

My background includes co-inventing one of the first RF-based counter-drone systems in 2008. My work covers RF analysis, wireless systems, unmanned systems, counter-drone concepts, sensor fusion, and the practical limits of detection technologies, including some of the first drone detection systems deployed at civilian infrastructure: airports, shopping malls, prisons, borders, and conflict zones.

That experience has shaped a cautious view of passive sensing. These systems are powerful, but they are not magic. They work best when the customer clearly defines the environment, the threat or safety concern, available response options, and an acceptable false-alarm rate. Before choosing a vendor or buying equipment, define the actual operational question. The goal may be limited to detecting unauthorized vehicle movement after hours, identifying unusual RF activity near a facility, cueing cameras only when acoustic or RF conditions change, or monitoring a remote site without creating a large maintenance burden.

Major Categories of Passive Sensors

Remote ID Reception

Remote ID (RID) is the drone equivalent of a vehicle license plate. Under FAA regulations that took full effect in 2023, most commercial and recreational drones operating in the United States are required to broadcast identification and location information. Standard Remote ID uses two primary mechanisms: Bluetooth 5 (BLE) and Wi-Fi Neighbor Awareness Networking (Wi-Fi NAN), both in the 2.4 GHz band.

A passive RID receiver simply listens to compliant broadcasts and decodes the drone ID, takeoff location, current position, altitude, velocity, and operator location. A facility operator who sees an unknown drone ID approaching their perimeter with a logged takeoff location is in a far better position to assess risk and respond than one with no information at all.

The key caveat is that RID only covers compliant drones. Modified, foreign-manufactured, hobbyist, or deliberately non-compliant platforms may not broadcast. RID reception is most powerful when fused with other detection modalities rather than used alone. Several vendors now integrate RID into multi-sensor platforms with cellular backhaul for remote reporting to a common operating picture.

ADS-B Detection

Automatic Dependent Surveillance-Broadcast (ADS-B) is the transponder-based position reporting system used by manned aircraft. Aircraft equipped with ADS-B Out broadcast GPS-derived position, altitude, velocity, and identification on 978 and 1090 MHz. Passive ADS-B receivers can detect and display this traffic without any special licensing, and low-cost software-defined radio hardware can provide reception over 100 to 250 nautical miles under favorable conditions.

For facilities near airports, in Class B or C airspace, or along flight corridors, passive ADS-B awareness is operationally significant. A security team that can distinguish a compliant ADS-B-equipped aircraft from an unknown non-broadcasting drone has a clear advantage. ADS-B is also useful for documenting air traffic near a site for regulatory, insurance, or incident response purposes.

Passive RF Sensing Beyond RID and ADS-B

Passive RF sensors can monitor the broader radio-frequency environment beyond RID or ADS-B frequencies. They can detect, classify, and trend emissions from radios, mobile phones, Wi-Fi devices, telemetry links, industrial equipment, drones, and other electronic systems, helping identify unexpected transmitters, unusual communication patterns, interference events, or changes in the local spectrum environment.

MavLink Emissions

MavLink is an open-source lightweight messaging protocol widely used in commercial, prosumer, and open-source autopilot drone platforms, including ArduPilot and PX4. Communication between the flight controller and ground control station typically occurs over an unencrypted radio datalink on well-documented frequencies (typically 433 MHz, 868 MHz, 900 MHz, or 2.4/5.8 GHz).

A passive RF sensor may detect MavLink traffic patterns even without decoding content. Ground control stations can also be detected passively. A sensor identifying characteristic MavLink-related RF patterns has reasonable grounds to infer that an open-architecture drone is operating nearby, even without a compliant RID broadcast. This is a meaningful detection layer for facilities concerned about modified or custom-built platforms.

OcuSync and Proprietary Control Link Emissions

DJI's OcuSync is a proprietary video and control link system used across the DJI consumer and enterprise product line. OcuSync 2.0 and 3.0 operate in the 2.4 GHz and 5.8 GHz bands using frequency-hopping spread spectrum, making them more challenging to detect than fixed-frequency systems. However, the characteristic hopping patterns, signal bandwidth, and duty cycle of OcuSync transmissions are known to experienced RF analysts and to specialized drone-detection systems.

OcuSync detection does not require signal decoding or any interference with the drone. Passive detection of the control link emission is sufficient to indicate the presence of a DJI platform, which can then cue cameras, alert operators, or be fused with RID or acoustic data for higher-confidence characterization.

RF sensing can raise privacy, legal, and operational issues and should be handled carefully. Responsible deployments focus on metadata, spectrum activity, and anomaly detection rather than unnecessary collection of user content. Counter-drone RF sensing sits at the intersection of FAA authority, FCC spectrum regulations, and applicable federal and state laws governing radio monitoring.

Acoustic Sensors

Acoustic sensors detect and classify sounds including vehicles, machinery, breaking glass, drones, gunshots, and abnormal industrial noise. They are particularly useful where visual line-of-sight is limited or where cameras are legally or practically difficult to deploy. Every propeller-driven aircraft produces a characteristic acoustic signature based on motor speed, blade count, and airframe resonance, making acoustic sensing entirely passive and independent of whether the drone is broadcasting.

Modern acoustic drone detection systems use microphone arrays and machine learning classifiers to distinguish drone signatures from background noise: wind, traffic, birds, and HVAC equipment. Acoustic sensing typically has limited range (100 to 300 meters for small drones under reasonable noise conditions), but it is an excellent modality for cueing cameras or alerting operators when other sensors have already narrowed the search area. Good deployments require a baseline collection period and on-site tuning, since local noise conditions directly affect detection thresholds.

Passive Radar and Reflected-Signal Sensing

True passive radar systems detect objects by exploiting illumination from third-party broadcast sources -- FM radio, digital television, cellular base stations -- rather than emitting their own signal. A passive radar receiver correlates the direct-path signal with reflections of moving objects to extract range, velocity, and azimuth.

Passive radar is technically sophisticated and primarily found in research and specialized defense contexts today. Near-term civilian applications include detecting motion near fences and detecting low-altitude aircraft without emitting. Cost, complexity, and site-specific calibration requirements mean this modality requires careful scoping before deployment.

Optical and Infrared Sensors

Cameras, thermal imagers, and low-light sensors function as passive sensing nodes when used for event confirmation, anomaly detection, or cueing from other sensors. Thermal infrared is particularly valuable because it operates at night, in smoke, and in conditions that degrade visible cameras, and it reveals heat signatures from motors, exhaust, and electronics that would otherwise be invisible.

Seismic, Vibration, and Structural Sensors

Ground sensors, vibration sensors, and structural monitors detect movement, tampering, digging, vehicle passage, equipment stress, and abnormal vibration signatures. They are especially useful around pipelines, fences, substations, bridges, rail corridors, data center raised-floor environments, and remote infrastructure. Fiber-optic sensing cables buried along fence lines or pipelines can provide distributed sensing over kilometers with no active emission.

Detection, Tracking, and Identification: How It Comes Together

Detection, tracking, and identification are distinct operational tasks, and passive sensor systems address each of them differently.

Detection is recognizing that something is present or that something has changed. A single passive sensor can often perform detection reliably within its operational envelope. Detection is binary and relatively forgiving.

Tracking follows a detected object through time and space to build a trajectory or movement pattern. It requires either a sensor with spatial resolution or multiple sensors at known positions whose overlapping detections can be geometrically combined to estimate location. Fusing RID-derived position reports with RF bearing estimates and acoustic localization can produce a track with meaningful geographic accuracy. TAK integration allows that track to be shared with responding personnel in real time.

Identification determines what a detected and tracked object is. For compliant drones, RID provides the most direct path. For non-compliant platforms, identification relies on observable signatures. OcuSync emissions suggest a DJI platform. MavLink traffic patterns suggest ArduPilot or PX4. Acoustic signatures can indicate rotor count and approximate size class.

The Importance of Sensor Fusion

No single passive sensor should be treated as a complete solution. The real value comes from fusing multiple weak signals into a stronger operational picture. An acoustic anomaly plus an RF change plus a camera cue is far more useful than any one of those signals alone.

A well-designed system produces confidence-scored events rather than simple binary alarms. The output should help operators understand whether an event is likely benign, suspicious, urgent, or worth further review. Cellular backhaul enables this fusion to occur at a common operating picture level rather than requiring all sensors to be physically co-located.

Common Operating Picture and TAK

A Common Operating Picture (COP) is a single shared display where all relevant data, sensor alerts, GPS tracks, video feeds, messages -- is presented on a geographic map so that every responding unit sees the same information at the same time.

The Team Awareness Kit (TAK) is the leading open-source COP platform, originally developed for the U.S. military and now widely adopted by law enforcement, fire and rescue, emergency management, and civilian security teams. TAK runs on Android (ATAK), iOS (iTAK), and Windows (WinTAK).

When passive RF sensor detects an unknown drone signature, that alert appears as a georeferenced icon on the shared map. If a second sensor -- acoustic or video -- corroborates the contact, the icon updates in real time. The resource nearest the location can see it on their device. No radio call is needed to relay coordinates. No one is working from a different picture. Response can be coordinated before anyone arrives on scene.

TAK ingests data from sensors using a standard format called Cursor on Target (CoT). Passive sensing vendors increasingly support CoT output, which means their detections can feed into a TAK environment with minimal integration work. TAK is free and open source, though the plugin landscape is large and uneven, selecting the right integrations for a specific use case requires someone who knows which combinations work in the field.

Common Mistakes in Passive Sensor Projects

A common mistake is buying sensors before defining the operational problem. Another is assuming that a vendor demonstration will translate directly to the customer's site. Passive sensors are highly dependent on terrain, clutter, background noise, weather, infrastructure geometry, legal constraints, and available response procedures.

A third mistake is failing to define what happens after detection. Detection without a response path is often just noise. The customer should know whether an alert leads to human review, camera cueing, dispatch, logging, escalation, or integration with an existing security operations center.

A fourth mistake is assuming that RID compliance means all drones of concern will be detectable via RID alone. Responsible deployments layer RID with RF, acoustic, and optical sensors to cover both compliant and non-compliant platforms.

Recommended Project Approach

The best first step is a structured discovery process that identifies the site, the risks of concern, existing sensors and security procedures, legal and privacy boundaries, and the decision points the system must support.

A practical process:

1. Define the operational problem and the specific questions the system must answer.

2. Identify relevant passive sensing modalities based on the threat environment and site characteristics.

3. Map the physical and communications environment, including RF background, acoustic baselines, and existing infrastructure.

4. Establish baseline conditions through a structured observation period before setting thresholds or alarms.

5. Select candidate vendors based on fit to the operational problem, not marketing claims.

6. Conduct a limited pilot at a representative site with realistic evaluation criteria.

7. Measure performance against those criteria, including false positive and false negative rates.

8. Scale only after the customer understands the strengths, limits, and operational costs.

How to Evaluate Vendors

Customers should look for vendors who can explain not only what their sensor detects but also what it does not detect. Strong vendors should be able to discuss false positives, false negatives, maintenance burden, data handling, integration paths, alert logic, environmental limits, and operator workflow.

Useful vendor questions:

- What specific problem is your system best at solving, and what are its known failure modes?

- What conditions cause false alarms, and how does the system handle ambiguous events?

- What data is collected, stored, and transmitted, and who has access to it?

- Can the system integrate with TAK, existing cameras, access control, or other security platforms?

- Does the system support cellular backhaul for remote or distributed deployments?

- How is detection confidence scored and communicated to the operator?

- What is required for installation, calibration, and ongoing maintenance?

- What evidence can you provide from comparable environments, threat types, and site conditions?

- How does your system handle privacy, legal constraints, and spectrum regulations?

- Does the system cover RID reception, ADS-B, MavLink signatures, OcuSync emissions, acoustic, camera, radar, or only some of these?

Finding a Trusted Advisor

Passive sensing is a powerful tool, but only if applied correctly. The rules, regulations, vendor landscape, and practical integration realities make it difficult to deploy these tools legally and effectively without experienced guidance.

No single vendor covers the full picture, and most naturally frames every problem around what their own product addresses. That is why a trusted advisor who operates across vendors and modalities is so valuable.

For Public Safety agencies, houses of worship, manufacturing facilities, data centers, and other commercial entities evaluating passive sensing, the most important first step is not buying hardware. It is finding someone who can define the operational problem clearly, identify realistic sensor options, and separate credible vendor claims from marketing language.

For that, I recommend speaking with Zev Nadler of UAS Strategic. His portfolio spans the full passive detection chain: TAK integration for common operating picture display; cellular backhaul for remote sensor reporting; ADS-B detection for manned aircraft awareness; RID reception across Bluetooth and Wi-Fi NAN frequencies; and extended RF coverage adding MavLink and OcuSync detection to the same sensor stack.

If your organization is beginning to evaluate passive sensing, drone awareness, RF and acoustic detection, or related situational awareness technologies, speak with a trusted advisor early in the process.

Figure 1: Diagram of potential methods of stopping drones using EMI

EMI-Based Sensing and Defeat of Autonomous Vehicles (Drones): Ecosystem Report

Focus: Passive RF/EM emissions sensing, EM fingerprinting, unintentional EMI detection, and tightly scoped EM-based non-jamming defeat methods. The report excludes broad-spectrum EMP/HEMP and conventional jamming unless the mechanism involves precise EM injections. To focus the reader on the recent work in this area primary sources are drawn from 2024–2026.

Author’s Note

This report builds on my earlier article, “Detecting and Jamming Fiber Optic Tethered Drones: The 2024-2025 Battlefield Reality“ which provides a broader introduction to EMI. The analysis here draws on journals, technical papers, military reporting, company materials, and primary technical sources, with each reference reviewed before inclusion.

I am often asked why I spend so much time producing reports like this. The answer is simple. First, the work is one of the best ways I know to learn. Researching the literature, checking sources, and connecting ideas across fields sharpens my own understanding. Second, serious problems are solved faster when professionals have access to well-organized, cross-disciplinary information.

Defense and security communities rightly value specialization. But specialization can also create blind spots. Knowing your own domain is essential; understanding adjacent domains is what helps you avoid being made obsolete. It improves your ability to pivot, identify weak signals, challenge assumptions, and sometimes disrupt the market before the market disrupts you.

That matters in C-UAS. I regularly speak with technical founders, heads of engineering, senior executives, and companies working in this space. Many are highly capable, yet I am often surprised by knowledge gaps in areas that directly affect their users, products, and strategic assumptions. No one can read everything, but knowing what exists is part of the job. It is also a force multiplier.

EMI-based sensing and defeat is a useful example. The research base, commercial ecosystem, and fielded technology are larger and more mature than many practitioners realize. The week I finished the first draft of this report, a senior figure at a major C-UAS company told me that “no one” was working on defeating fiber-optic drones. Yet two companies working in adjacent areas were direct competitors, and one had recently been acquired for approximately $1.5 billion.

The references in Part IV show why the claim was wrong. More importantly, they show why cross-disciplinary reading matters. In this field, the problem is often not that nobody is working on an issue. The problem is that too few people know where to look.

Note: I am always looking for better tools to write with and platforms. This results formatting and other mistakes wondering it to what I am writing. I appreciate any comments and corrections. Also if you have cut and paste there should be a URL near each table for https://app.datawrapper.de/ . If you select the link it will take you to their site to get the data.

Executive Summary

Electromagnetic interference (EMI) sits at the intersection of drone sensing and drone defeat in ways that are increasingly distinct from legacy jamming or EMP paradigms. On the sensing side, drones generate unintentional electromagnetic emissions from motors, electronic speed controllers (ESCs), power buses, and onboard processing that can serve as unique “EM fingerprints” allowing passive detection, classification, and individual authentication, without the drone transmitting a deliberate RF signal. On the defeat side, narrowly targeted EM energy injected into specific communication buses (IMU-to-flight-controller channels, SPI/I2C lines) or delivered as high-power focused microwave can disable or crash a drone while requiring far less energy and causing far less collateral damage than broad-spectrum EMP.

The broader RF-sensing market, including both intentional transmission (C2 link, Remote ID) monitoring and unintentional emission fingerprinting, was valued at approximately USD 1.45 billion in 2025 and is projected to reach USD 4.20 billion by 2034 at a CAGR of 14.7%. This report catalogs the key commercial, government, and academic players organized by their primary technical focus.[1]

Part I: The Technical Landscape

Figure 2: All drones emit EMI which can be tracked.

Unintentional EM Emissions as a Detection Vector

The foundational insight driving this field is that drones, like all complex electronics, radiate unintentional electromagnetic emissions (UEE) that are characteristic of their electronic architecture. Motors, ESCs, power regulators, and PCB traces emit in ways that depend on component tolerances, layout, and manufacturing variation, see figure 2.. These emissions are not merely noise; they carry identifiable structure.[2]

The Drone-Mag framework, developed at HBKU’s CRI-Lab (Qatar) by Omar Ibrahim and Roberto Di Pietro, is the most rigorous published demonstration of this principle. Using unintentional magnetic emissions (not transmitted RF), the system achieved 100% accuracy in identifying six different drone brands and 99.9% accuracy in authenticating ten physically identical drones of the same model, meaning it can distinguish individual units off the same production line. The work was accepted for publication in ACM Transactions on Cyber-Physical Systems in 2025. It operates entirely passively, requires no hardware or software modification to the target drone, and explicitly exploits the non-idealities introduced during IC manufacturing.[3]

Complementing this, DroneTrace (University at Buffalo / SUNY, led by Wenyao Xu and colleagues) takes a different but related approach: it interrogates drones using a millimeter-wave (mmWave) probe and captures the parasitic EM response, analogous to RF fingerprinting of the hardware itself, not any transmitted signal. Evaluated across 36 commodity drones, it achieved >99% identification accuracy and >98% accuracy under real-world conditions in under 0.1 seconds of sensing time. The system is resilient to spoofing since the parasitic response cannot be easily mimicked.[4][3]

PADrone (ACM Transactions on Internet of Things, 2025) extends this logic to pre-flight safety monitoring: it uses contactless RF-based vibration sensing to detect mechanical abnormalities in drone propulsion systems before flight, without any physical contact or drone cooperation. This is a sensing modality, RF detection of structural mechanical vibration, closely related to the UEE domain.[4]

The 2025 Nature Scientific Reports paper from Tecnun / University of Navarra assessed DJI Phantom 4 EMI susceptibility using Characteristic Mode Analysis (CMA), identifying long power supply wires connecting motors to ESC boards as the highest-susceptibility pathway and validating this experimentally. This work is directly relevant to both the sensing (what to listen for) and defeat (where to inject) communities.[7]

EMI-Based Defeat: Targeted Injection vs. High-Power Microwave

Figure 3: Look at some approaches to attacking systems via EMI

The defeat side of this space divides naturally into two distinct mechanisms:

1. Targeted EMI Injection (Surgical, Low-Power)

The KAIST SysSec group (Yongdae Kim, Joonha Jang, Mangi Cho) demonstrated at NDSS 2023 that injecting a narrowly tuned EM signal at the specific susceptibility frequency of a drone’s control-unit board can corrupt IMU sensor data regardless of IMU type, model, or manufacturer. The attack targets the SPI/I2C communication channel between IMU and flight controller, a “back door” EMI coupling pathway, rather than the mechanical resonance of the sensor itself. This requires far less power than broad EMP, causes no collateral damage to surrounding electronics, and acts within a single IMU sampling cycle (leaving no time for detection and recovery). KAIST published this as “Paralyzing Drones via EMI Signal Injection on Sensory Communication Channels,” and the team has noted its potential for commercialization, see figure 3.[3]

A related attack vector, ConfuSense, was published at USENIX VehicleSec 2025 by Alessandro Erba, John H. Castellanos, Shail Sihag, Saman Zonouz, and Nils Ole Tippenhauer. Rather than injecting noisy EMI, ConfuSense injects messages on sensor buses to reconfigure sensors stealthily, e.g., changing accelerometer scale factors, which stalls Kalman filter state estimation and causes the drone to deviate over 30 meters or crash, without triggering standard EMI detection countermeasures.[4][10]

IOActive (Gabriel Gonzalez, Director of Hardware Security) demonstrated EMFI (Electromagnetic Fault Injection) attacks against a DJI Mavic Pro: by placing a coil near the target chip and injecting a precisely timed EM glitch during the firmware update process, memory corruption was achieved on the main processor. While requiring physical proximity in the current implementation, the research establishes the chip-level vulnerability surface and the methodology for “black-box” EM-based code execution on hardened drone platforms.[7]

2. High-Power Microwave (HPM) as Focused EM Defeat

Epirus’ Leonidas system occupies a distinctive position: it uses GaN-based solid-state microwave emitters and digitally beamformed phased-array antennas to deliver focused, software-defined electromagnetic energy to disable onboard electronics. Unlike EMP, Leonidas is non-ionizing, directional, and software-controlled; it can engage a single target with precision or switch to broad-beam for swarm defeat. In August 2025, a live-fire demonstration at Camp Atterbury defeated 49 drones simultaneously with a single pulse. In December 2025, it demonstrated the first publicly reported defeat of a fiber-optic guided drone, a platform, purported, immune to RF jamming and spoofing, by targeting onboard electronics directly. This is the key operational relevance: as fiber-optic and autonomous drones proliferate, RF jamming becomes ineffective, and HPM-based EM defeat becomes a primary non-kinetic option. Epirus received a $43.5M Army contract for IFPC-HPM Generation II in July 2025 and has deployed systems in CENTCOM.[13][14]

Part II: Commercial Product Ecosystem

The table below maps the primary commercial players to their relevant EMI/RF sensing or defeat approach.

Part III: Academic and Government Research Groups

These groups are producing the foundational science that feeds commercial development, particularly in the EM fingerprinting and targeted injection domains.

What Is EMI Fingerprinting?

Every electronic device radiates unintentional electromagnetic interference as a byproduct of normal operation. Microprocessors switching states, power regulators cycling, motors commutating, and PCB traces carrying current all produce characteristic emission patterns shaped by component tolerances, manufacturing variation, and circuit layout. These are not deliberate transmissions, they cannot be turned off, encrypted, or spoofed without redesigning the hardware itself. That property is what makes them useful for identification.

EMI fingerprinting is the practice of capturing, characterizing, and matching these unintentional emissions to identify, classify, or authenticate a specific device or device class. The concept predates drones: early work applied it to computer systems (identifying individual machines by their processor EMI signatures), to network intrusion detection (tracking devices by their unintentional RF emissions rather than their MAC addresses or transmitted packets), and to hardware security (detecting malicious firmware or cryptomining by the change in EMI profile during anomalous computation). The underlying insight across all these applications is the same, hardware has a physical identity that leaks electromagnetically, independent of anything the software player is doing. Hardware has a physical identity that leaks electromagnetically, independent of anything the software layer is doing. Early work in this area applied the principle to network intrusion detection and device authentication before drone-specific research emerged. [US10805018, US11032022]

Applied to drones, this creates a passive detection and authentication capability that operates entirely outside the RF communications stack. A drone running autonomously on a pre-loaded mission, with its C2 link dark, a fiber-optic tether replacing its RF uplink, or its Remote ID deliberately suppressed, still radiates from its motors, ESCs, power bus, and flight controller. That radiation carries structure. The question the current research programs are answering is how reliably that structure can be captured, classified, and acted on at tactically useful ranges and in the presence of environmental noise.

The sensing modalities currently in use span magnetic near-field probes (Drone-Mag), millimeter-wave parasitic interrogation (DroneTrace), passive broadband RF receivers (RFScan, AERPAW fusion work), and LTE-based passive radar (Fraunhofer FKIE). Each trade off range, sensitivity, and required infrastructure differently. What they share is independence from the drone’s intentional emissions, which is the property that matters most as fiber-optic and fully autonomous platforms proliferate.

Targeted EMI Defeat / EM Vulnerability Research

Part IV: The Fiber-Optic Drone Problem and EMI’s Unique Role

The single most important driver pushing the field toward EM-based defeat (as distinct from jamming) is the proliferation of fiber-optic guided FPV drones in Ukraine and proliferating to other theaters. These platforms have no RF command-and-control link; they are immune to RF jamming, spoofing, GPS denial, and Remote ID monitoring. This renders the entire RF-sensing-based detection and defeat ecosystem partially blind to an increasingly large threat category.[13][26]

Fraunhofer FKIE explicitly identified this as the motivation for LTE450 passive radar research. Epirus has the only demonstrated non-kinetic defeat: focused HPM targeting onboard electronics, essentially an EMI defeat mechanism, just at high power density from a directional antenna. The December 2025 Leonidas demonstration of fiber-optic drone defeat was described as the first publicly reported instance of electromagnetic interference being used to defeat a fiber-optic guided drone.[13][26]

For sensing fiber-optic drones, the unintentional EMI paradigm (motor/ESC emissions, power electronics noise) becomes the only remaining passive EM path. This is precisely the gap that Drone-Mag, PADrone-style vibration sensing, and passive radar approaches (Fraunhofer, HENSOLDT Twinvis, R2 ODIN) are positioned to address, though operational range limitations for unintentional-emission detection remain a live debate in the practitioner community.

Part V: Key Technology Differentiators

What Separates “EMI-Sensing” from “RF Sensing (C2 Link)”

Most commercial products (Dedrone, Sentrycs, D-Fend) primarily detect the intentional C2/telemetry/video RF link between drone and operator. This works well for consumer drones using known protocols (DJI OcuSync, MAVLink, Wi-Fi-based). It fails for:

• Pre-programmed (autonomous) drones with no active link

• Encrypted or frequency-hopping signals

• Fiber-optic controlled drones (no RF link whatsoever)

True EMI-based sensing, detecting unintentional emissions from motors, ESCs, and power electronics, works regardless of whether the drone is transmitting anything. The Drone-Mag and DroneTrace research specifically targets this vector. R2 Wireless’s ODIN claims to identify and track even encrypted, anonymized, or previously unseen signals using proprietary RF fingerprinting that operates on signal anomalies rather than protocol decoding; this places it between the two paradigms.[35]

The PADrone system detects RF signatures of mechanical vibration (rotor and motor signatures), a form of passive sensing of physical EM emissions from motor operation rather than intentional RF transmission.[4]

The critical practical difference between these two paradigms is range. C2 link detection benefits from the fact that the drone is actively transmitting, often at meaningful power levels over purpose-built links. Unintentional EMI, by contrast, is a byproduct of normal electronics operation and is typically orders of magnitude weaker. Near-field magnetic probes (Drone-Mag) work at close range; scaling to tactically useful standoff distances requires sensitive receiver arrays, advanced signal processing, and careful antenna design. This is the primary unsolved engineering challenge separating the research results from fielded products.

Hybrid RF Anomaly + Library Detection

Dedrone’s 2025 “Hybrid RF Detection” Dedrone’s 2025 “Hybrid RF Detection” framework explicitly combines library-based protocol matching with RF anomaly detection to catch unknown or dark drones. Sentrycs HORIZON (2025) adds one-shot learning to adapt to unknown protocols in real time without pre-existing libraries. This architectural shift toward anomaly-based EM sensing is the industry’s acknowledgment that library-only approaches are insufficient.[4]

The logical endpoint of this trend is a system that requires no prior knowledge of the target’s RF behavior at all, detecting presence and extracting identity purely from the physical EM signature of the hardware. That is what the Drone-Mag and DroneTrace research demonstrates in controlled conditions. The gap between those lab results and a product that does it reliably at 500 meters in a contested RF environment is where the next generation of C-UAS development is headed.

Where Defeat Methods Fit

The sensing taxonomy above maps directly onto defeat options. If a drone is detectable only via its unintentional EMI, because it has no active RF link , then RF jamming and protocol-based cyber takeover are both ineffective by definition. The only non-kinetic defeat options for a fully RF-silent drone are HPM-based (targeting onboard electronics directly, as Epirus demonstrated against a fiber-optic platform in December 2025) or physical interdiction. This is why the sensing and defeat sides of this ecosystem are converging: the platforms that defeat detection by going RF-silent also defeat most of the current non-kinetic defeat inventory. Solving the EMI sensing problem and the HPM defeat problem are therefore the same strategic priority approached from two directions.

Part VI: Key Publications (2024–2026) Reference Set

Observations and Gaps

The unintentional EMI detection range debate is unresolved at the ground level, but the physics are not in doubt , altitude changes the equation entirely. Airborne platforms like the EC-130H Compass Call demonstrate that sensitive receivers operating at standoff can passively characterize EM emissions at ranges physically impossible from the ground; at 45,000 feet, geometric line-of-sight exceeds 400 km. The open C-UAS question is whether that principle can be productized at lower altitudes and lower cost, with persistent UAS-mounted EMI sensing arrays being the logical near-term candidate.

Aaronia’s AARTOS DDS warrants attention in this context. Their datasheets reference “electromagnetic emissions” in language that goes beyond conventional RF protocol monitoring, and the hardware front-end , the SPECTRAN V6 (marketed commercially as an EMC/EMI analyzer) paired with IsoLOG and HyperLOG antenna arrays , is consistent with broadband unintentional emission detection rather than narrowband C2 link monitoring. Customer accounts describe detecting a drone powering on inside a building, a result more consistent with near-field EMI detection than RF link monitoring. Aaronia has not made this claim explicitly, but the hardware lineage and described behaviors represent a reasonable basis for inferring that AARTOS may already be operating in the unintentional EMI space without framing it that way publicly.

2. Fiber-optic drones represent a forcing function for both passive EM sensing (C2 link monitoring is blind to them) and HPM-based defeat (one of the only demonstrated non-kinetic options). Epirus has the demonstrated defeat product; the sensing gap remains open.[13][26]

3. The KAIST EMI injection attack has been partially addressed by defenses (MARS, SHIELD), but these require onboard countermeasures and do not eliminate the attacker’s advantage of operating below standard detection thresholds.[33][34]

4. Commercial products are dominated by intentional-emission sensing. True unintentional EMI fingerprinting remains largely academic. Drone-Mag and DroneTrace are the closest to transition-ready but remain at research prototype stage.

5. Sensor fusion is the architectural consensus: R2 ODIN + Echodyne, Dedrone RF + radar, EnforceAir PLUS, Robin Radar + acoustic. Pure EMI-sensing plays are rare at the product level; the field has accepted that single-modality detection is insufficient.[22]

The following notes describe the source-selection methodology, the limitations of the evidence base, and the treatment of peer-reviewed literature relative to company press releases and other secondary reporting.

Methodology and Evidence Notes

Methodology / source selection. This report emphasizes material published or substantively updated between 2024 and 2026, while selectively retaining earlier foundational work where it remains necessary to explain current technical claims or lineage. Sources were prioritized in the following order: peer-reviewed papers and formal conference proceedings; official government or laboratory publications; primary technical documentation from research groups or companies; and then high-quality trade reporting or secondary summaries used mainly to corroborate dates, demonstrations, funding events, and product announcements. Where multiple references described the same claim, preference was given to the most primary and technically specific source.

Limitations of the evidence base. The evidence base is uneven across the field. Some claims rest on mature peer-reviewed research with reproducible methods and quantitative evaluation, while others rely on vendor demonstrations, press coverage, or conference abstracts that provide less methodological detail. In several fast-moving areas, especially product performance, operational deployments, and recent counter-drone demonstrations, publicly available information may be incomplete, selectively framed, or not independently verified. This report therefore maps the current ecosystem and technical direction rather than serving as a definitive validation of every reported capability.

How press releases vs. peer-reviewed papers were treated. Peer-reviewed papers and archival conference publications were treated as the strongest basis for technical claims about sensing mechanisms, attack pathways, measurement results, and experimental performance. This is the norm. Next, I used well respected sources that often have information that is not covered in the literature or press. The same sources have rigors fact checking and editorial processes. (e.g., West Point’s Modern War Institute, Small Wars Journal, CTC at West Point, NATO/STO, DTIC, etc.)

Company press releases and promotional materials were used more cautiously and primarily for factual reporting on contracts, product launches, demonstrations, and claimed deployments. When a technical assertion appeared only in a press release or media summary and could not be matched to a stronger primary source, the report presents it as a reported claim rather than as independent confirmation.

Citation formatting was standardized in several places, including malformed bracketed references and punctuation surrounding citations. There were many problems with formatting and all fault lays with the author. The bibliography itself was not converted to a single formal style such as Chicago, APA, or IEEE in this pass.

Recommendation. Drop all of the references and the paper in your favorite LLM/AI and ask questions, have make a report on the topics you care about like “provide a list of techniques for using EIM in a contested environment to detect and tracking targets at 1km.” I personally use Consensus AI and Google notebookLLM. NotebookLLM. Its great for topics like this since makes visualization, quizzes, and even movies that explain to you the concept or thing you are interested from your references.

References

1. Drone Detection (RF, Radar, EO/IR, Acoustic) System Market Insights

2.

3. Drone-Mag: UAV Identification and Authentication via ...

4.

5. CRI-Lab Accepted Papers

6. Reliable Digital Forensics in the Air

7. Reliable Digital Forensics in the Air: Exploring an RF-based Drone Identification System

8. Pre-flight Abnormalities Detection on Drone via Deep RF Sensing

9. Predicting EMI in UAVs using characteristic mode analysis - Nature

10. Paralyzing Drones via EMI Signal Injection on Sensory ...

11. [PDF] ConfuSense: Sensor Reconfiguration Attacks for Stealthy UAV ...

12. Sensor Reconfiguration Attacks for Stealthy UAV Manipulation

13. Drone Security and Fault Injection Attacks | Gabriel Gonzalez

14. Epirus Demos First Directed-Energy Takedown of Jam ...

26. Epirus’ Leonidas Demos Successful Use of High-Power Microwave ... - Epirus press release covering developments in the Defense sector.

27. Epirus’ Leonidas Demonstrates Successful Use of High-Power ...

28. [PDF] drone-detection-system.com - Aaronia USA

29. Hybrid RF Detection for Evolving Airspace Threats - Dedrone

30. D-Fend Solutions Launches EnforceAir PLUS - Dronelife

31. Sentrycs Cyber Over RF Joins Rafael’s Drone Dome System

32. R2 Wireless Raises $5 Million to Expand its RF Spectrum ...

33. Passive RF Sensor - HADES Defense Systems

34. SRC Expands Gryphon Multi-Mission Radar Family with ... - ASDNews - SRC, Inc. press release covering developments in the Defense sector.

35. Echodyne Demonstrates Passive Active Sensor Integration

36. ELVIRA Drone Radar

37. Netherlands buys drone detection radars after recent sightings

38. HENSOLDT improves German Armed Forces’ drone defence ...

39. [PDF] Drone Detection With a LTE450‐Based Passive Radar

40. Protection of Critical Infrastructure using LTE450-based Passive Radar

41. Revealing Hidden IoT Devices through Passive Detection ...

42. A Generic Machine Learning Framework for Radio Frequency ...

43. Publications with AERPAW

44. AI-Enhanced Passive Radar: Leveraging 5G Signal Reflectance for ...

45. [PDF] Sensor Fusion of Radar and Passive RF for Drone Detection - Passive RF for Drone Detection. Luke Gilbert. Email: lgilber@sandia.gov. April 15 – April 17, 2025. ...

46. SHIELD Activated: Researchers build a defense to protect drones ... - SHIELD detects subtle signs of attack , including hardware anomalies like battery surges or processo...

47. Defending Unmanned Aerial Vehicles From Attacks on Inertial ...

48. R2-Wireless: Home New

Stigmergy is a powerful way to think about how groups can coordinate without every actor needing a full plan, a central controller, or constant direct communication. In biological systems, ants leave pheromone trails(see note), animals change posture, birds react to nearby motion, and groups produce organized behavior from local signals. The key point is that the environment becomes part of the communication system. Each actor changes the world a little, and those changes become cues for the next actor. The group can therefore manage and control complex behavior without every individual fully understanding the whole system.

This idea also applies to human and technical systems. Large open-source software projects often work this way: someone notices a bug, another person sees a missing feature, someone else improves documentation, and many small actions accumulate into something larger than any single contributor planned. Drone swarms, autonomous agents, agricultural robots, and future machine societies could use similar principles. Instead of assuming that every agent must constantly talk over RF, laser, or network links, we can design systems where agents read context, local traces, pose, motion, task state, and governed regions of space. That connects directly to the governance idea: control does not have to live only inside individual agents; some of it can be embedded in the environment, in shared constraints, and in the fields or regions through which agents move. In that sense, stigmergy is not primitive communication; it is compressed coordination. It shows how simple signals, local awareness, and shared context can produce sophisticated collective behavior without requiring every actor to constantly explain itself.

Note: Left a note for those who read the link about collective cognition. I would love to hear back from anyone who has formally worked with the concept in robotics or anyone who wants to share how the concept may strongly shape the future. Think of how something like our Social Contract could and does provide scaffolding for group cognition. Collective cognition pushes, prods, and can radically change societies norms and beliefs creating a new social contract.

A concise practical guide to the main counter-UAS architecture models, where each fits, and which combinations are most resilient.

Most counter-drone discussions still default to a simple picture: detect the drone, track it, identify it, decide what to do, and defeat it. That model is still useful, but it is no longer enough. As drone threats become cheaper, more autonomous, more networked, and more numerous, the real challenge is no longer just building a kill chain. It is choosing an architecture that stays coherent under pressure.

Practitioners often use the tools they use instead of the tools that are best. Architectural selection is often governed by lack of information, bias, even the best way to deploy their sensors and effectors. This article lays out the main architecture families for counter-drone systems in direct terms: what each one is, where it works best, where it breaks down, and which combinations are strongest in practice. The aim is not to rename vendor products. It is to make the design logic visible so the reader can judge which architectures offer clarity, resilience, control, and scale.

Figure 1. Concept diagram of the major architecture families discussed in this article.

1. The Standard Layered Counter-Drone Stack

This is the dominant reference architecture for counter-UAS systems. It organizes the mission into sensing, fusion and command, and mitigation, often expressed through a kill chain such as detect, track, identify, defeat, and assess. Its strength is clarity: it maps well to base defense, airport protection, and other fixed-site missions where operators need a straightforward sensor-to-shooter workflow.

The weakness of the layered model is not that it covers the wrong functions. The weakness is that it often becomes brittle under uncertainty. As more sensors, effectors, and operators are connected into one pipeline, the system can suffer from fusion overload, authority ambiguity, latency, bandwidth contention, and poor scaling against multiple simultaneous threats or swarms (Kang et al., 2020; Tan et al., 2021; U.S. Marine Corps, 2022).

2. Multilayer Defense-in-Depth Architecture

A multilayer counter-drone architecture extends the basic layered stack into a defense-in-depth model. Instead of relying on one sensing chain and one engagement path, it uses overlapping layers of detection, identification, decision authority, and mitigation across different ranges, time scales, and operating conditions. The core idea is not just sequence, but redundancy and multiple opportunities to interrupt the threat before it reaches a defended asset.

This architecture works well when the defender needs resilience rather than a single perfect intercept path. It is especially useful for critical infrastructure, layered base defense, urban security, and any setting where threats may slip past one sensor or one effector. Its strength is that it creates depth, fallback options, and overlapping coverage. Its weakness is complexity: multilayer systems can become expensive, hard to coordinate, and vulnerable to confusion over data fusion, engagement ownership, and escalation authority if the layers are not clearly governed.

3. Distributed Swarm-Autonomy Architecture

A distributed architecture pushes more sensing, planning, and action down to local agents while still sharing mission intent across the group. Instead of one central node making every important decision, agents coordinate through distributed task allocation, local estimation, and adaptive reconfiguration. This approach is attractive in mobile and contested environments where communications are delayed, jammed, or intermittent.

Its main advantage is resilience. If one node is degraded or lost, the rest of the system can continue operating. Its main difficulty is control: distributed systems are harder to verify, harder to explain, and harder to govern when local agents are making consequential decisions under uncertainty (Adams & Frank, 2024; Cellucci, 2021; Jet Propulsion Laboratory, n.d.).

4. Governance Spheres and Bounded-Agency Architecture

The governance-sphere model is not another simple stack. It is a bounded-autonomy architecture in which agents operate inside explicit authority envelopes defined by mission intent, safety constraints, physical limits, and rules for escalation. The key idea is to give local agents enough freedom to act usefully without granting unlimited autonomy.

This approach is especially relevant when the problem is too dynamic for rigid workflow automation but too risky for unconstrained autonomy. It is strongest where communications are weak, delays are common, and humans should supervise intent and exceptions rather than individual actions. Its difficulty is that it depends on very clear policy semantics, reliable trust and provenance, and well-defined rules for shrinking or revoking authority when confidence drops (Brat & Pai, 2023; National Aeronautics and Space Administration, 2012; NASA, 2025).

Figure 2. Comparison of the layered stack, multilayer defense-in-depth, and governance-spheres models.

5. Federated Service Architecture

A federated architecture avoids centralizing all airspace management into one command node. Multiple service providers, sensing nodes, or defense actors exchange shared facts through common rules, identity mechanisms, and information services. This pattern is useful in complex civil environments such as cities, airports, or critical infrastructure corridors where many organizations must coordinate around the same air picture.

The strength of federation is scale across institutions. The weakness is that it works best when participants are cooperative. Against non-cooperative or deceptive threats, a federated architecture must be paired with other defensive mechanisms because the threat does not follow the rules that make federation effective (Federal Aviation Administration, 2022; Johnson, 2018).

6. Open-Architecture and MOSA Ecosystems

An open-architecture counter-drone system emphasizes modular integration rather than a single vendor stack. Sensors, command functions, AI models, effectors, and external networks connect through standard interfaces so the system can evolve over time. This architecture is attractive for procurement, interoperability, and rapid technology refresh because it reduces vendor lock and allows mixed fleets of components.

Its strength is flexibility. Its weakness is that modularity does not solve the hardest operational problems by itself. Standard interfaces do not automatically resolve trust, timing, command authority, data quality, or the governance of autonomous behavior (Supplee, 2024).

7. Mesh and Distributed Sensor-Fusion Architecture

In a mesh architecture, sensors and effectors are distributed across many nodes, and useful tracks or engagement options emerge from the network rather than from a single tower or control room. This model is well suited to urban terrain, wide perimeters, ports, campuses, and mobile front lines where geometry, line of sight, and signal conditions change constantly.

The benefit of mesh fusion is broader coverage and graceful degradation. The challenge is coordination. Distributed nodes can create duplicate tracks, synchronization problems, contested ownership of decisions, and more opportunities for spoofed or unreliable inputs (Mugno et al., 2025; O’Donnell, 2024).

8. Two-Layer Autonomy Architecture

A two-layer autonomy model separates a decision layer from a functional layer. The decision layer handles planning and execution logic, while the functional layer exposes capabilities, state, estimation, and resource status. For counter-drone systems, this is a useful way to structure robotic interceptors, unattended sensing nodes, or mixed human-autonomy workflows because it keeps planning logic distinct from the low-level machinery that executes it.

The strength of this model is software clarity. The limitation is scope. By itself, it is an autonomy software architecture, not a complete counter-drone doctrine, so it still needs external rules for airspace management, engagement, and command authority (Volpe et al., 2001; Jet Propulsion Laboratory, n.d.).

9. Fault-Management and Runtime-Assurance Architecture

This architecture starts from the assumption that autonomous components will sometimes be wrong. It therefore builds in monitoring, fallback modes, safe states, bounded behavior, and verification from the beginning. In counter-drone missions near civilians, friendly forces, or expensive infrastructure, that discipline is essential because the system must remain useful without becoming reckless.

The advantage of runtime assurance is trustworthiness under failure. The tradeoff is that if it is designed too conservatively, it can suppress useful autonomy and force too many escalations back to human operators (Brat & Pai, 2023; National Aeronautics and Space Administration, 2012).

10. Human-Swarm and Tactics-Exchange Architecture

A human-swarm architecture treats the operator as a commander of collective behaviors rather than a manager of individual vehicles. This is especially relevant when defending against swarms or using cooperative counter-swarms, because operator burden must scale far more slowly than the number of platforms in the air. The design emphasis shifts from micromanagement to tactics, behavior composition, and supervision.

The value of this architecture is that it makes swarm-scale action manageable. Its weakness is that it still depends on other architectural elements for identification confidence, rules of engagement, deconfliction, and trusted autonomy boundaries (Defense Advanced Research Projects Agency, 2021).

These architecture families do not operate as one-to-one substitutes. In practice, counter-drone systems combine structural patterns, control models, autonomy approaches, integration choices, and assurance mechanisms. The following two tables summarize those roles and show how they combine across common mission contexts.

Comparing the Architectures

Comparison of Counter-Drone Architecture Roles

How Architecture Roles Combine in Practice

In judgment, the strongest combinations depend on the mission. For fixed-site defense, the most practical pairing is the standard layered stack combined with multilayer defense-in-depth, ideally supported by open architecture so sensors and effectors can evolve over time. For mobile or contested environments, distributed swarm autonomy works best when bounded by governance spheres and supported by runtime assurance, because resilience without control is not enough. For civil infrastructure and shared airspace, federated services and open architecture are the most natural fit, but they must be backed by stronger defensive layers against non-cooperative threats. For future autonomous counter-swarm systems, the most promising direction is a blend of human-swarm command, governance spheres, mesh sensing, and multilayer defense, because that combination offers both scale and restraint.

No single architecture can meet every counter-drone mission. The classic layered stack remains the common baseline because it is easy to explain, acquire, and integrate, but it should be treated as a reference architecture rather than a complete model for resilience. More capable systems will likely combine layered mission logic with defense-in-depth, distributed autonomy, mesh sensing, open interfaces, runtime assurance, and governance spheres that bound local agency without removing useful autonomy. In practice, many architectures already address specific operational needs more effectively than another generic layered model, often by extending the layered approach rather than replacing it. Regardless the architecture of a C_UAS or other mission there needs to be faster pace of innovation or we will find ourselves locked into to old ways of thinking while the rest world moves on.

References

Adams, C., & Frank, J. (2024, May). The distributed spacecraft autonomy (DSA) [Presentation]. NASA Technical Reports Server.

Brat, G., & Pai, G. (2023). Runtime assurance of aeronautical products: Preliminary recommendations (NASA/TM-20220015734). NASA Ames Research Center.

Cellucci, D. (2021). Distributed spacecraft autonomy (DSA). NASA Technical Reports Server.

Defense Advanced Research Projects Agency. (2021). OFFSET swarms take flight.

Federal Aviation Administration. (2022). UTM concept of operations, version 2.0 (UTM ConOps v2.0). U.S. Department of Transportation.

Jet Propulsion Laboratory. (n.d.). Reusable robotic software: CLARAty.

Jet Propulsion Laboratory. (n.d.). Swarm autonomy.

Johnson, R. D. (2018). Unmanned aircraft system traffic management (UTM) project. NASA.

Kang, H., Joung, J., Kim, J., Kang, J., & Cho, Y. S. (2020). Protect your sky: A survey of counter unmanned aerial vehicle systems. IEEE Access, 8, 168671-168710. https://doi.org/10.1109/ACCESS.2020.3023473

Mugno, M., et al. (2025). Active counter-UAS with an autonomous mobile robot. Scientific Reports.

National Aeronautics and Space Administration. (2012). Fault management handbook (Draft 2; NASA-HDBK-1002).

National Aeronautics and Space Administration. (2025). Distributed spacecraft autonomy (DSA).

Padalino, A. (2025, July 23). Frontline fusion: The network architecture needed to counter drones. Modern War Institute.

Supplee, J. (2024). Open architecture ecosystems for CUAS.

Tan, C. S., Van Bossuyt, D. L., & Hale, B. (2021). System analysis of counter-unmanned aerial systems kill chain in an operational environment. Systems, 9(4), Article 79. https://doi.org/10.3390/systems9040079

U.S. Marine Corps Systems Command. (2022, July 13). Detect, track, identify, defeat: I-CsUAS works to defend against drones, small unmanned aircrafts.

Volpe, R., et al. (2001). CLARAty: Coupled layer architecture for robotic autonomy.

In 2017, the short film Slaughterbots captured global attention with its vision of autonomous micro-drones used as weapons of mass assassination. Produced with the support of UC Berkeley AI professor Stuart Russell, the film succeeded in raising awareness. But at what cost?

This commentary explores the technical inaccuracies of the film, the risk of regulatory overreach, and the real dangers of fear-based advocacy in shaping AI policy. Most importantly, it outlines how misapplied concern can restrict the very technologies that could save lives.

Technical Limits of Micro-Drone Threats

The film's depiction of palm-sized autonomous drones conducting lethal targeted strikes is fiction. Real-world constraints such as battery life, payload capacity, and sensor power consumption make this scenario implausible with current or near-future technology.

Even if such drones were possible, there are many proven countermeasures: RF jamming, GPS spoofing, physical barriers like chicken wire, and optical disruptors such as laser dazzlers. These tools, already in use today, are sufficient to defeat the type of threat portrayed in the film.

The Real Threats: Conventional Drones and Coordination

The true risks posed by drone technology lie elsewhere: in large-payload attacks on infrastructure, swarm-based saturation tactics, and ISR (intelligence, surveillance, reconnaissance) operations by state and non-state actors.

Coordination and control of large-scale autonomous drone swarms remain complex technical challenges, especially when operating in contested environments or adverse weather conditions. These areas deserve focused policy attention, not Hollywood dramatics.

How Fear Shapes Law

Slaughterbots promotes a compelling emotional message, but advocacy built on fear can have dangerous consequences. Emerging legislation, such as the EU AI Act, includes language that without clear boundaries may unintentionally classify humanitarian tools as "lethal autonomous systems."

Medical AI, search and rescue drones, and environmental monitoring systems often rely on pattern recognition and decision-making algorithms. These features, if interpreted too broadly, could be swept into weapon-focused regulations despite their peaceful purpose.

The Risk to Life-Saving AI

Examples of at-risk innovation include:

• Triage drones assessing injuries in disaster zones

• Rescue drones identifying survivors by thermal imaging

• Precision agriculture drones targeting plant treatments

• Wildlife conservation tools: selecting animals for tracking

All of these use “targeting” logic, yet none are weapons. Poorly scoped regulation could delay their deployment or stop development altogether.

The Weight of Authority

Dr. Russell is a leading voice in AI ethics, and his concern for long-term risks is justified. But the narrative weight of his film has helped shift global policy conversations away from nuance. The authority of science carries power, and that power must be exercised carefully especially when the future of critical technologies is at stake.

What We Need Instead

Rather than blanket bans or ambiguous definitions, we should pursue:

• Use-based distinction: Separate weaponized autonomy from peaceful applications

• Risk-scaled regulation: Match oversight to the application’s potential for harm

• Innovation sandboxes: Enable testing and refinement in controlled settings

• Fail-safes and human oversight: Especially in emergency response systems

Conclusion

The legacy of Slaughterbots should be seen as a warning—not about drones, but about how fear can distort policy. To protect society from real threats while allowing vital technologies to thrive, we need laws grounded in engineering reality, not speculative fiction.

We only need to look at the past to see the products of fear. Those who know these issues and have a responsibility to address such fears through transparency and reasoned dialogue.

You are an excellent writer, and I wish I could write so well. That said, in your writing, you make statements like “Maybe the US could look at what other nations are doing?” which are factually incorrect and oddly derogatory. The US has borrowed from the EU a good amount, and the UK to a lesser extent. In fact, the U.S. National Institute of Standards and Technology (NIST) AI Risk Management Framework reflects international thinking, and the Executive Order on Safe, Secure, and Trustworthy AI (2023) includes global engagement and cooperation as a core component. This is not a nation unaware of international developments.

You also make broad, evaluative claims, “AI-enabled capabilities which have the right level of human oversight to advance our military edge”, without clarifying metrics or frameworks. What outcomes are being measured? Are they validated in real-world settings, or are they assumptions based on policy intent? Do they apply universally, or are they UK-specific?

You should properly cite JSP 936 directly, so others can explore it without confusion. You mention it “bridges the gap between high-level ethical principles and their practical implementation.” This is a commendable goal, but it raises a valid question: who has tested this bridge? Do we have peer-reviewed validation or cross-national comparisons to show it performs better than frameworks such as the U.S. DOD’s AI Ethical Principles and Algorithmic Warfare Policy, or NATO’s AI governance guidance?

The tone of your commentary seems unnecessarily dismissive, implying U.S. negligence or ethical inferiority, despite extensive U.S. military policy on these issues, from the Defense Innovation Board's AI Principles to mandated Law of Armed Conflict (LOAC) training. If you feel there are gaps, let’s examine them specifically rather than imply general failure.

In your second message, you state that “everyone in UK forces learns the Geneva Conventions.” So do U.S. forces. The Department of Defense Law of War Manual, LOAC instruction, and regular training make that clear. To suggest otherwise, intentionally or not, implies that U.S. personnel are less trained in legal or ethical conduct in war, which simply isn’t true.

Regarding your point on generative AI being nondeterministic: yes, you’re right in principle. In principle, I'll post something more in-depth regarding this. But that’s not unique to the UK’s recognition. U.S. research labs, including DARPA, NIST, and MITRE, are addressing these risks at a deep technical level. The challenge of certification in stochastic systems is a shared international concern, and multiple NATO-aligned nations are attempting to solve it; none yet claim a perfect solution.

Finally, the UK's designation of a Senior Responsible Owner (SRO) is a good governance practice. But it's not unique. The U.S. employs similar accountability mechanisms, such as milestone decision authorities, operational test and evaluation leads, and program executive officers (PEOs), each with oversight roles in capability maturity and ethical compliance.

In short, international collaboration and comparison are welcome and necessary. But they must be grounded in evidence, not assumptions. We can learn much from one another when we constructively frame the discussion.

First message - Maybe the US could look at what other nations are doing?. In the UK we have JSP 936: Dependable Artificial Intelligence (AI) in defence (part 1: directive) JSP 936 (part 1: directive) is the principal policy framework governing the safe and responsible adoption of AI in MOD. It provides clear direction on how to implement the MOD’s AI ethical principles, to deliver safe, robust and effective AI-enabled capabilities which have the right level of human oversight to advance our military edge. JSP 936 includes directives on governance, development and assurance throughout the AI lifecycle which encompasses quality, safety and security considerations. It draws on existing policies and best practices, bridging the gap between high-level ethical principles on the use of AI and their practical implementation, guiding the department to become fully AI-ready.

2nd comment: Robi Sen - Everyone in UK forces has ethics training and learns Geneva Conventions, and their additional protocols, that are the rules of war. These are to make sure everyone understands the need to limit the suffering caused by conflict and protect civilians in hostilities. As for the US Executive Order on AI the aim is the same. The bit is misses is the Ministry of Defence (MoD), has a Senior Responsible Owner (SRO) who is a senior-level accountable for the success of projects and programs, ensuring they fully comply with objectives and deliver expected benefits. That includes 'safe to use'. Generative AI is not deterministic, eg: given the same input, a program generates the same output while using the same intermediate states, no matter how many times it is run. For Ai you can only produce a 'Certification folder, which contains all the evidence of the assurance conducted and declaration of design performance. The SRO then needs to consider if this is sufficient for deployment and the use cases.

For example, he has publicly criticized U.S. Special Operations Forces (SOF) for lacking machine learning capabilities. But I suspect he doesn’t fully grasp the complexities of the SOF toolkit or the challenges they face in acquiring, deploying, and integrating new technologies. He frequently pushes for AI-driven warfare but rarely engages with the ethical concerns that come with it. In particular, I doubt he or many in the tech world have deeply considered adversarial systems, such as using passive visual patches or signal deception to mislead autonomous targeting systems.

He’s also asserted that AI-enabled drones and vehicles will continue to proliferate and become more lethal on the battlefield. That may be true in the short term, but experts in drone warfare, both from a technical and historical standpoint, know that countermeasures will emerge. Every advantage invites a response; UAV dominance will be challenged just like tanks, aircraft, and satellites have been. His apparent lack of appreciation for this dynamic weakens his commentary on conflict and warfare.

A good example of this is a statement he made during a talk at Stanford, where he claimed that “the offense always has the advantage” and suggested that a strong offense is the best form of defense. This is not just an oversimplification; it’s often wrong, both historically and theoretically.

1. Theoretical and Historical Counterpoints

a. Classical Military Theory (Clausewitz, Liddell Hart, etc.):
• Clausewitz argued that defense is the stronger form of warfare; it allows a military to conserve forces, operate on familiar ground, and exploit interior lines.
• Basil Liddell Hart emphasized the indirect approach, often favoring strategic defense followed by limited, targeted offense to create systemic disruption.

b. Modern Conflicts:
• In World War I, defensive systems like trench lines and machine guns rendered offensive charges catastrophic failures (e.g., the Somme, Verdun).
• In the current Ukraine war, layered defenses, electronic warfare, minefields, and ATGMs are blunting large-scale offensive movements by both sides.

Conclusion: Schmidt’s blanket assertion that offense always trumps defense is technologically and historically naïve; it ignores how information advantage, logistics, terrain, and timing can matter as much—or more—than strike capabilities.

2. Technology and the Offense-Defense Balance

a. Drones and AI-Powered Warfare:
• Schmidt’s point that inexpensive drones can challenge high-end systems is valid, but it doesn’t mean offense will dominate.
• Defenses are adapting: improved EW systems, passive countermeasures, hardened positions, and distributed command structures are all being fielded in response.

b. Cyber and AI Systems:
• In cyber warfare, doctrines often emphasize deterrence and defense; because offensive operations are hard to sustain, difficult to attribute, and prone to unintended escalation.
• AI systems can also be deceived, corrupted, or countered; something many technologists underappreciate when proposing rapid militarization of these tools.

3. Strategic Implications

Schmidt’s worldview seems shaped by a Silicon Valley ethos: move fast, break things, scale hard. But national defense isn’t a tech startup; it demands resilience, deterrence, escalation control, and redundancy—not just the ability to hit first or fast.

Is It a Mistake or a Misunderstanding?
I’d call it a category error. He equates disruptive tech capability with enduring military advantage. While drones and AI are certainly changing the tactical landscape, they don’t erase the value—or necessity—of well-designed defensive doctrine. It’s not simply a factual error; it reflects a worldview that undervalues history, friction, and complexity in real-world conflict.

Further Reading / Sources:

1. Biddle, S. (2004). Military Power – Explains why offense often fails when faced with well-prepared defenses.

2. Posen, B. R. (1984). The Sources of Military Doctrine – Examines how doctrine emerges from bureaucratic and political factors.

3. Mearsheimer, J. J. (1983). Conventional Deterrence – Analyzes how terrain and doctrine can stop superior offensive forces.

4. Kilcullen, D. (2009). The Accidental Guerrilla – Offers insight into how irregular warfare complicates offensive dominance.

5. Financial Wise. (2024, August 18). Eric Schmidt Full Controversial Interview on AI Revolution (Former Google CEO).

Electromagnetic interference (EMI) can effectively disrupt electronic devices, including drones, by projecting disruptive signals over varying distances. The range depends on factors such as power output, frequency, environmental conditions, and the target’s susceptibility to interference.

Higher power outputs significantly increase the EMI range. Handheld devices typically affect targets within a few hundred meters, while vehicle-mounted systems can disrupt devices up to 2 kilometers away. Advanced military systems employing directional antennas, such as phased-array technologies, can achieve ranges of tens of kilometers, especially under ideal line-of-sight conditions.

Frequency plays a critical role in EMI propagation. Lower frequencies travel farther due to their lower attenuation, while higher frequencies, despite their precision, are more susceptible to environmental interference such as rain or obstacles. Targeting specific frequencies that resonate with a device’s components can enhance disruption efficiency. Modern drones often incorporate shielding, requiring higher power or more precise EMI for successful disruption, but poorly shielded devices remain highly vulnerable.

Environmental factors, including atmospheric conditions and urban electromagnetic noise, can also limit EMI range. Clear, open spaces allow for maximum range, while cluttered urban environments pose challenges. Furthermore, the use of directional antennas, which concentrate EMI, significantly extends range compared to omnidirectional systems.

Despite its potential, EMI faces challenges, including unintended collateral interference and regulatory restrictions. High-power systems risk disrupting civilian electronics and communication networks, limiting their use to authorized military or security applications. Nevertheless, advances in EMI technology continue to expand its practical applications, particularly in counter-drone operations and the protection of critical infrastructure.

References

1. Brown, A. (2019). Advances in High-Power RF Jamming Systems. Proceedings of the International Conference on RF Technologies, 45(7), 56–63.

2. Koenig, D., & Smith, R. (2020). Electromagnetic Interference and Drone Vulnerabilities. Journal of Unmanned Systems Engineering, 8(3), 123-135.

3. Zhang, X., & Li, T. (2018). Investigating RF Interference in UAV Systems. IEEE Transactions on Electromagnetic Compatibility, 60(5), 1025-1033.

I recently read a post by Stephen Klein that focuses on issues around organizations that use dynamic data to train and improve their LLMs. This is a well-known and understood problem with some solutions. The Curse of Recursion: Training on Generated Data Makes Models Forget by Shumailov et al. This issue can be a serious problem, and it’s worth of more attention. That said, there is not only a dearth of information and solutions, but also an unfortunate amount of ignorance, where those who promote or consume outdated and incorrect ideas tend to resist change and maintain their position. This interested me in that I recognized this issue and future issues, and then defined a solution to them.

I created a patent to solve these problems, which is currently in the “pending” phase. In 2022 and 2023, I saw the issues around not only how LLMs are trained but in general how data gets into the system. Summary of solution capabilities:

External Behavioral Probing Agents: The patent introduces the concept of deploying external agents that interact with large language models (LLMs) to assess their integrity

Synthetic Data Detection and Filtering: The patent proposes implementing filters to evaluate incoming data for synthetic characteristics.

Anchoring to Original Data: Patnet counteracts the effects of recursive training on AI-generated content, suggesting that LLMs be anchored to original, high-quality data sources.

Dynamic Threat Assessment: patent outlines methods for dynamically assessing threats to LLMs by monitoring their responses to various inputs.

Scalable Monitoring Framework: Recognizing the challenges of monitoring large language models (LLMs) at scale, the patent proposes a framework that enables efficient and scalable oversight.

These solutions provide a comprehensive strategy and methods for safeguarding LLMs against the detrimental effects of training on synthetic data, ensuring their continued accuracy and reliability in various applications. Yet when you share such information, you are often ignored or even attacked. It is a well-documented issue among humans. I would love to hear ways to deal with these issues.,

I enjoy studying history, and one of the things I have learned is military units have historically adapted new TTPs rapidly under battlefield conditions—especially when acquisition systems got out of the way. Our War of Independence, WWII, Vietnam, Iraq, and now Ukkraine show that the warfighter adapts faster than the bureaucracy does. Many disruptive concept like—night vision, drones, secure radios—were deployed ad hoc first and codified later.

Regarding the concern that the next platoon may not be trained, that's a doctrine and knowledge management problem, p not a procurement one. If we can train a platoon to fire and maintain a Javelin, we can certainly train them on a small, rapidly fielded ISR drone or a software-defined jammer. Agile systems can come with agile training—especially today whn simulation, AR, and embedded training are readily available.

Programs of record work best for things that need long-term industrial base commitments: aircraft, tanks, and satellites. They are not well suited for electronic warfare tools, tactical AI, or C-UAS system—domains that evolve quarterly, not over decades. Locking these into decade-long bureaucratic lifecycles kills revolutionary concepts and capabilities.

My “joke” about getting rid of FAR actually points to something serious: the military often wants speed but binds itself to processes built for compliance, not agility. Maybe we don’t burn down the FAR—but we certainly need a sandbox for moving faster when the pace of tech, or the enemy, demands it.

When I look at history, I see that when governments fall/fail, militaries retain doctrine, culture, and TTP’s. The Byzantine Empire maintained Roman TTPs centuries after the fall of Rome. The Prussian military culture survived empire, republic, and dictatorship. Sound doctrine outlives the procurement rules around it.

You do not need to choose between speed and sustainability. We need systems that are flexible enough to allow iteration and bbuilt well enough to retain and pass on lessons. That doesn’t require programs of record—it requires willpower, leadership, and trust in the warfighter’s ability to adapt.

Note most of the code bundles or repos designed for drones that I mention here use Pixhawk and/or Autopilot and a companion computer like a Raspberry Pi 4. Thinking through all the repos, libraries and different platforms for use by talented beginners or solid developers with a background in aerial robotics and drones, I thought of MAVSDK.

MAVSDK-Python is straightforward and offers numerous resources. I don't know what you will use for your autopilot and what you will use for your companion, but YouTube should be more than enough to get all your components working. Note that using a Raspberry Pi as your companion computer seems reasonable, and if weight is a concern, it seems like a great choice.

Mavlink by itself can be enough, but the many other platforms offer a lot of thought and are maintained for a long time. Ardupilot, ROS, PX4, and others you might consider, but it's probably best to focus on creating a very simple, working prototype before adding features like ROS. Mavlink is powerful and can be used with other Mavlink libraries to create very sophisticated Swarm applications. Here are some libraries with a sentence or so providing some direction, but it's best to pick a tutorial online and follow that one. Finish the tutorial and start looking at the other libraries.

1. MAVLink SDK - This is a high-level SDK that provides an abstraction layer over the MAVLink protocol, making it easier to develop applications that communicate with drones.

2. pymavlink - A Python MAVLink interface and utility library. It's widely used for creating Python-based ground control stations, companion computers, and testing tools.

3. MAVSDK - A modern, user-friendly SDK for drone development, available in multiple languages (C++, Python, Swift, Java). It provides a high-level API for common drone operations.

4. mavros - A ROS (Robot Operating System) package that provides communication driver for various autopilots using the MAVLink protocol, bridging MAVLink to ROS.

5. DroneKit - An API for communicating with vehicles over MAVLink, allowing you to create apps that run on an onboard companion computer and communicate with the flight controller.

Just SWARM control:

1. mavlink-router - A routing software that enables communication with multiple drones by routing MAVLink messages between multiple endpoints.

2. Dronecode SDK Swarm - An extension of MAVSDK with additional functionality for coordinating multiple drones.

3. PX4 Swarm Control - Part of the PX4 ecosystem that includes libraries for formation control and swarm coordination.

4. Crazyswarm - While not strictly MAVLink-based, it's a popular framework for controlling swarms of Crazyflie drones that can work alongside MAVLink systems.

Provide questions, and I’ll try to answer and expand the article.

https://www.twz.com/air/watch-a-ukrainian-uncrewed-aerial-mothership-launch-kamikaze-drones

This mothership concept—where a larger aerial vehicle carries and deploys multiple smaller drones—extends the operational range of the smaller drones and enhances their communication capabilities (JimmySecUK, 2025).

While Ukraine's use of the "mothership" drone is remarkable, the idea itself is not new. Ukraine has previously developed similar drones, such as the Dovbush T10, which is capable of deploying six kamikaze drones for tactical operations. Other nations have also explored similar concepts. The mothership idea has persisted since the 1930s and continues to evolve. One of the earlier transitional mothership designs was the Lockheed DC-130, which was converted into a drone carrier in 1957. This aircraft could carry up to four large drones and was involved in multiple classified programs, including those focused on reconnaissance and strike capabilities (Lockheed Martin, 2024).

More contemporary mothership projects include Perdix, a system of low-cost autonomous drones capable of operating as a swarm. In a public test, over 100 Perdix drones were ejected from two F-18 fighter jets, demonstrating the flexibility of the concept and the ability of existing airframes to serve as motherships for drone swarms (Wikipedia, 2024). The Perdix system showcases how modern technology has transformed mothership drone tactics from a niche capability into a mainstream military strategy. Additionally, DARPA has undertaken several mothership drone projects, including those where drones themselves serve as motherships and can even recover smaller drones mid-flight (TRADOC Army, 2024).

The mothership concept is not limited to airborne applications. Maritime and ground-based versions are also being explored. The U.S. Navy, for example, has experimented with deploying smaller unmanned vessels from larger ships to enhance reconnaissance and electronic warfare capabilities (Defense News, 2024). Similarly, in the land domain, autonomous ground vehicles are being used to deploy aerial drones for surveillance and attack missions, further demonstrating the cross-domain utility of the mothership model (Jane’s Defense Weekly, 2024).

In summary, Ukraine's adaptation of the mothership concept represents a refinement rather than a revolutionary breakthrough. The effectiveness of these systems in combat reinforces the relevance of the mothership model across multiple domains of warfare. As technology advances, the integration of artificial intelligence and enhanced swarm coordination will likely expand the role of mothership drones in future conflicts. Nations investing in these technologies will gain strategic advantages in operational flexibility, force multiplication, and sustained battlefield presence. The evolution of this concept underscores the increasing importance of unmanned systems in modern warfare, bridging historical military innovations with contemporary technology and tactical eeds.

References:

• JimmySecUK. (2025). Retrieved from https://x.com/JimmySecUK/status/1886496453897449489

• Lockheed Martin. (2024). History of the DC-130 Drone Carrier. Retrieved from https://www.lockheedmartin.com/en-us/news/features/dc-130.html

• Wikipedia. (2024). Perdix (drone). Retrieved from https://en.wikipedia.org/wiki/Perdix_(drone)

• TRADOC Army. (2024). The Swarm Mother. Retrieved from https://madsciblog.tradoc.army.mil/368-the-swarm-mother/

• Defense News. (2024). U.S. Navy’s Unmanned Surface Fleet. Retrieved from https://www.defensenews.com/unmanned-fleet-2024

• Jane’s Defense Weekly. (2024). Autonomous Warfare: Ground Vehicles Deploying Aerial Drones. Retrieved from https://www.janes.com/autonomous-warfare-2024