Been thinking about if to write this article for a long time, pondering the benefits versus the potential costs of what I’m about to write. Everything that will be illustrated here is available in the public domain and most of it has been so for well over a decade now. Been battled on this as there is the not so remote possibility that some bad actors could use it as some sort of inspiration, but given the current lack of understanding among those in decision making positions it is IMO important to expose what truly are the capabilities available thanks to modern consumer technology and simple online shopping. Some links, tutorials, videos and online shopping resources have been removed from the article or not included in ti hoping to make life a little harder for those hoping to use this as a guide, but reality is that details of most that we will be discussing are just a couple of google searches away.
In the last two decades consumer technology has done impressive leaps forward. General connectivity, sensors and micro computers are just some popular examples and all this advancements are crucial to what we will be analysing with this article. Also crucial is the internet in general and the impressive amount of informations, guides, software and general resources that are shared on it.
Small rebel groups, terrorist organisations, military contractors, small military powers and even legit resistance fighting forces (see Ukraine), have been capitalising on such technological and informational improvements , and in the last decade especially we have seen a rapidly evolving “open source warfare” landscape, especially for what are commonly called “drones”.
(those in the field will bark at me for my use of the simple, generic but popular term “drone”, all kind of acronyms are actually used, but for the sake of simplicity, and for the personal pleasure of imagining some going mental over my use of the word, I plan to abuse it in this article 🙂 )
The first step will be that of explaining the very basics behind the core characteristics of the vehicles I will be writing about.
- Design type
- Engine and power source
- Guidance method
Multiple designs and vehicle types are available for the nefarious purposes we will be discussing, and while we will focus on flying platforms in this article, please know that almost everything we will discuss applies to wheeled, tracked, floating or even submerged vehicles.
Focusing on flying stuff we have two popular choices :
- Rotary (helicopter like, but more likely, multicopters)
- Fixed wing (planes basically)
Each type has different benefits which I will we try to summarise.
Among the most basic rotary designs are helicopters, which rely on a propeller to counter gravity. A more modern (and somewhat brutal) approach, possible thanks to electric engines and modern electronics, is that of multicopters which are very easy to design and rely on very simple components. Due to their diffusion and greater simplicity, we will ignore helicopter like designs in this article and focus on multicopters.
In their most popular incarnation a multicopter will have 4 engines and 4 propellers (quadcopter) rotating in opposite direction in pairs (this to counter the torque generated by the engines). Give more power to all engines to climb or lower it to descend are the basic principles, regulating the power of each engine (in pairs usually) will then allow for more complex manoeuvres as pitching and rolling.
This designs can be small, can be launched from extremely limited spaces, and are very easy to fly even for inexperience pilots. Due to their nature and the amount of energy used to counter gravity, this designs will altho have a limited range and payload.
Planes rely on lift to fly (rotary designs too, but the explanation is well out of the scope of this article), the wings, according to the forward speed, provide an upwards force that counters gravity so that the power generated by the engine is solely focused on moving the platform forward, the faster it moves, the more lift the wings generate the more gravity is countered all while getting closer to the destination. What this translates to is very good range and payload capabilities, especially compared to rotary designs. They will altho be more complex to launch as they require something akin to a small airstrip (varying according to size and weight of design) or some other launch system to get the drone in the air (smaller designs can be hand launched, but as payload and range increase, catapults of sorts become the only option other than an airstrip).
We should have learned that based on design type we can expect wildly different ranges from our DIY warfare instrument. With rotary designs ranges up to about 20 kms can be easily achieved, to get beyond that designs become more complex and progressively incur in limits brought by the energy storage method (batteries) and speed of the design. Hybrid approaches to this problem are very effective, with this design for example employing a combustion engine to power to produce the electricity to power the electric ones keeping the drone flying, thanks to this approach the vehicle can achieve over 2.5 hours (or over 200 kms in range) of flight time carrying over 20 kgs of payload.
For fixed wings designs it is instead conceivable to achieve rather impressive ranges up to multiple hundreds of kms rather easily.
With payload we intend the weight the design can carry other than that it needs to fly, such warheads, surveillance/recon equipment and eventually advanced guidance related components (more on this soon). Just as for range, rotary designs are, generally, way more limited than fixed ones, with payload capacity (today, future breakthroughs in battery technology will directly impact this) remaining in most cases well below 10 kgs.
Fixed wings designs suffer way less from payload limitations, with their payload being directly tied to the chosen power/energy source (engine/fuel), size of the design and desired range. But more on this later.
Engine and power source
Rotary designs will in most cases be limited to electric engines for propulsion and batteries as energy storage method. It is conceivable to hypothesise a rotary design similar to a helicopter with a combustion engine and fuel as energy storage method, but such designs would be more complex and out of the scope of this article. Realistically, easy to purchase or design rotary based drones will be fully electric (thing that impacts severely their range and payload), but this doesn’t mean that more committed designers can’t come up with alternative solutions.
For fixed wing designs attackers will be less limited, with simple options being once more electric engines and batteries, but combustion engines are a simple, cheap and popular choice too. According to budget even jet engines become conceivable.
The guidance methods
In their most basic incarnation this drones can simply be remotely controlled. A “pilot” will sit comfortably somewhere far from the target and guide the drone up to its final destination. Cheap and commercially available communication kits will allow our drone to be piloted from hundreds of kilometres of distance if needed with a decent enough live video feed allowing for extremely precise guidance.
This guidance type should altho be the easiest one to disrupt, as “jamming” the radio link between the pilot and the vehicle should leave the drone uncontrolled until it crashes somewhere far from its intended target. In principle, as it was up to about two decades ago, jamming such signals should be extremely easy, the radio link used to guide the drone operates on a specific radio frequency and a jammer would just need to transmit “noise” over such frequency (with a higher power than the signals to be disrupted) to break the guidance link. Well, its not so easy anymore….
In the last decades we have seen the introduction and diffusion of dozens of consumer technologies that operate via radio signals, making the EM (electro magnetic) spectrum somewhat clogged. It was basically as if concurrent signals originating from home wifi setups, cellphones, digital tv etc where “jamming” each others unwillingly.
To counter this, consumer technology has begun adopting a technique previously mostly limited to high end military systems (which had to counter enemy jamming) called “Frequency Hopping” which consists in radio transmitters and receivers constantly switching frequency in a random pattern so to minimise the chances for different signals to clash on the same frequency (as each one is used for a tiny amount of time). If a clash happens its not a big a problem as next data packet will probably go trough on a different frequency and communication link will not be much affected (for those familiar with networking, this is similar to package loss around which a network protocol can be designed so to make its effect non disrupting).
Jamming a Frequency Hopping communication link requires the jamming platform to transmit a signal more powerful than the one it want’s to jam, and has to do so over the full frequency spectrum the target communications can employ. To achieve this, the power available in the jamming platform has to be divided across multiple frequencies severely impacting its range.
Modern, consumer, RC Transmitters and Receivers employ Frequency Hopping of varying quality and complexity, and while high end RC comm gear used to be expensive, in the last decade we have seen many cheap solution becoming available, with a decent, long range (>100 KM), solution costing as little as $100 online.
All this said, modern EW (electronic warfare) solutions can, to some extent, still contrast such guidance method, both via Jamming of the signal (altho at a somewhat limited range) and more effectively by detecting the transmission source (where the pilot would be) and “taking care of it”. This because once you start transmitting a radio signal, it is extremely easy to detect such signal and pinpoint its source.
So, while modern tech has made a relatively simple guidance method as pure “remote control” relevant again in warfare scenarios, it’s practical usability is somewhat limited as the source of the signal will probably be detected (with plausibly deadly consequences), the flying platform will be detected as it is transmitting back to the control station (pilot) a data and video feed requiring a constant stream of radio transmissions, and the signal can still be jammed in many scenarios (this altho depends upon many factors, EW platforms are limited and can’t protect every potential target. Also, in some urban scenarios even pinpointing the signal source might not allow for a fast enough response)
Basic Autonomous Guidance
The diffusion of GPS like technology has since many years reached the consumer RC plane and “drone” market. Off the shelves products can be programmed to take off from a location, follow a specific and precise flight path and reach a designated target with sub meter accuracy. Many commercial quadcopters will offer such capacities with fancy and well designed apps that allow to program very complex flight plans with few taps on your smartphone or tablet.
To achieve this, commercial products rely on GPS (or similar) receivers, gyroscopes, accelerometers and digital compasses, all components that became ridiculously cheap thanks to modern smartphones. All together this sensors compose what is called a IMU (Inertial Measurement Unit) which blends the data from the different sensors and returns a precise location, speed, elevation and attitude. With this data, what is by now trivial software, can guide whichever vehicle, be it a car, a plane, a quadcopter or a futuristic stroller.
This allows for our attackers to make very few modifications to an off the shelf product (just need to trigger the warhead somehow) and obtain an autonomous vehicle that can fly, drive or in general reach a target via a more or less complex route and bring havoc to such target.
With this guidance method, the humans launching the attack can be virtually unexposed, neither those launching the attack nor the flying platforms will emit radio signals that can be detected and there is even no need for a “pilot” of sorts with the skills to fly/drive the attack platform to the target.
Those defending from such an attack have altho a rather easy way to counter it which is the jamming or spoofing of the positioning signal (GPS, Glonass etc). Due to how systems such as GPS are designed, they can be jammed very easily, they don’t employ frequency hopping and the signal is transmitted from space with very low power, so it is easy to overwhelm it and have the drone loose most of its positional awareness. Most consumer platforms in this case will simply land in panic or attempt to backtrack on their path until signal is re-acquired. Consumer grade positioning systems are also subject to what is called “Spoofing” which basically consists in tricking the platform into receiving a fake signal allowing to deny correct positioning data over wide areas (example of this in Russia / Ukraine etc link articles). Spoofing can also be used in a more complex way which would basically allow to guide the target platform to a specific position via slight adjustments to its perceived positions.
Recently the commercial world has been heading towards non GPS dependant solutions since in many situations, even for civilian use, the satellite signal isn’t reliable enough for the increasingly precise needs of those using the platforms. Inertial and optical based positioning are slowly creeping into the consumer world, and are already available for those wanting to experiment in DIY designs, but more on this later on.
Non off-the-shelf solutions
Up till now most of what I’ve wrote about can be easily achieved via simple off the shelf solutions, commercial products exists that allow attacker to setup a “suicide” drone with close to no effort simply purchasing a pre made flying platform with almost no modifications at all. Even easier if the drones are used for intelligence gathering purposes. A commercial platform can be easily “programmed” (no code required, just mouse clicks or finger taps) to take off from a location, fly a long a very precise flight path, take pictures with a high resolution camera in some specific locations, and autonomously return back “home”, all without emitting a single radio signal.
But… that is only the tip of the iceberg. Since the early 2000s the diffusion of increasingly easy to program micro controllers has made the world of “open source warfare” impressively more reachable and worryingly capable than in the past. Some core, mostly Open Source technologies are crucial to this “easy” to achieve capabilities.
The origin of modern, cheap and easy to program micro controllers, especially for the uses we are discussing, can be traced back to the late 90’s early 2000’s with the Lego Mindstorm (which itself origins from a MIT project – add links) project which had the goal of allowing kids to design and program robots with varying degrees of complexity.
A micro controller is basically a small “computer” which can be programmed and be easily interfaced with sensors, motors and actuators. At the core of the Lego Mindstorm platform there was/is a micro controller and a very simple to use programming interface which uses a simple drag and drop solution to define the robots behaviour.
The concept soon developed into more open and advanced solutions such as Arduino which employs cheap hardware, energy efficient design and an open development platform with nowadays a community of hundreds of thousands of developers contributing with knowledge, code, ideas and designs.
The initial goal of the platform was that of allowing a fast prototyping solution which modern hardware designers could use to experiment advanced features for their projects which could then be turned into commercial products (to make a simple example, in 2007 I could use Arduino to design an automated plant watering system which would use a simple humidity sensor in contact with the earth beneath a plant and command a servo motor to release water once the humidity level dropped beneath a certain threshold. Once I had something working, I could then bring the design into production with more dedicated and efficient hardware).
The versatility of the platform and the open nature of it, rapidly lead to a huge community of “thinkerers” which would use Arduino for a very wide range of DIY designs.
Thanks to similar micro controllers, coupled with once expensive and exclusive sensors becoming extremely affordable thanks to smartphones (gyroscopes, accelerometers, digital compasses, gps etc) we begun seeing the introduction of multicopters which would use a micro controller to integrate data from multiple sensors and balance the power output of a set of engines into the desired flight behaviour. At the core of the programming of similar vehicles which integrate data from sensors and use it to drive engines and actuators there is what is called a PID Controller which is a rather old concept that adapts perfectly to software implementations and grants a straightforward and intuitive way to calculate the correct input to give to engines, servos etc.
The use and correct implementation of PID Controller is what really makes it trivial to control all kind of vehicles as, independently from the movement type, given some coherent input data from attitude sensors, the PID will output correct values to pass to whichever propulsion or steering system a design employs, be it a submarine, a rover on Mars of a DIY cruise missile.
The DIY RC flight community jumped on this and by the early 2010’s we had the introduction of completely Open Source platforms such as “ArduPilot” which gradually allowed what once where extremely complex (and hence, usually gov reserved ) features to be integrated into custom flying, driving or even underwater moving designs.
You can today (and has been so since almost a decade now), build your own “drone” around ArduPilot using components of your choosing with highly customisable behaviours, all with little more than a few drag and drop operations. The Open Source nature of ArduPilot and similar solutions then allows for more crafty designers to integrate their own code and features into their creations while still taking advantage of the robust and versatile base platform.
Single Board Computers
Cheap and easy to use micro controllers as Arduino I have introduced above are at the same time very powerful but also very limited. They are powerful as per the capabilities they can bring with relative ease, but they are limited by very basic processing, limited memory and general computational power. An Arduino micro controller, or even a series of them, will have an extremely hard time at processing large amounts of data, they are perfect for flying a vehicle given basic positioning and attitude data, but would be next to useless for more advanced tasks such as realtime image analysis which has to process millions of pixels, multiple times per second.
So…. Say hello to Single Board Computers ! In the last decade we have seen the introduction and diffusion of platforms such as the Raspberry Pi which is to all affects a very small, cheap but capable, computer. Once again the Open Source design both behind the hardware and software controlling it (a full fledged, linux based, operating system in this case) lead to a wide adoption and a large and active community constantly attempting new designs, experimenting and detailing every step of their workings with the platform. Being Linux based, it is very easy to port to Raspberry Pi popular open source tools, software, frameworks and libraries which where designed for personal computers, often coming from the academic world.
Ok, but what do I need a Raspberry for ? As per the topic of the article, Raspberry and similar single boards computer (Raspberry is a popular example, Intel, NVIDIA and many other companies provide even more powerful solutions specifically optimised for computer vision) can be used to grant important additional capabilities to our “Open Source Warfare Platform”, especially capabilities that rely on precessing large amounts of data in “realtime”, they pack a lot of power in a small and energy efficient package, they are easy to program and can rely on decades of software development and researches.
A single board computer + plus decades of resources freely available online, could be used, once more with relatively little effort, to integrate into our drone something as machine learning based image analysis, giving the designs we are discussing capabilities such as discerning targets and / or geographical or architectural features (up to some extent) etc.
Similar capabilities become crucially important considering what we said earlier about ways to counter our DIY attack platform. Remote control signal can be intercepted and jammed, GPS signal can be easily jammed and even spoofed, so we need a way for our “vehicle” to navigate, at least in the final phase of its attack, with means independent from external radio signals. Cheap accelerometers and gyroscopes can provide a crude purely inertial navigation system if the GPS signal was to be not available, but the precision of such navigation would decease greatly with every kilometre travelled rapidly leading to potentially very large positioning errors, making such crude inertial navigation useless for final targeting.
But…. we can integrate the rough inertial positioning data with the result of automated image analysis, using for example peculiar buildings in a town to have our single board computer extrapolate a way more precise location just by using software (mostly already available in Open Source format), a camera which can be simply optical, infrared or even thermal and a $50 single board computer. The gps + inertial guidance just have to get our attack platform in the general area we want to target then something as image analysis can be tasked with the final, precise, targeting completely independent from external signals, so, completely resistant to the countering methods we discussed earlier.
Computer Vision / Realtime Image Analysis
Born soon after modern studies around Artificial Intelligence, Computer Vision since the 1960’s has been researched, developed and improved so to increasingly mimic what sight does for many living creatures, which can crudely be summarised as, providing spatial awareness and identifying “stuff” populating the surroundings of the creature (or software, AI, robot etc in this application). Mostly relegated into the academic, military or high end industrial worlds, computer vision begun becoming much more accessible in 1999 when Intel introduced OpenCV and made it Open Source. The library was born with the intent to provide a solid and extendable foundation for Computer Vision related applications. The open nature of the project lead to a rapid wide adoption and more importantly impressive additions to it making it now an all round solution for computer vision capable of handling even the most advanced CV related tasks.
Computer Vision is made immensely more powerful thanks to the integration of Artificial Neural Networks which are algorithms designed to loosely mimic how the human brain recognises patterns. The most common use of Neural Networks in relation to CV is commonly called “Image Classification” which uses Machine Learning models to identify objects or features in images and live video. This is achieved via a “training” process which, simply put, is achieved by providing a broad set of example images of the objects the ML Model will have to recognise to the Neural Network which will process the images and develop a more or less effective model usable for realtime applications which don’t require much computational power.
While it all might sound complex, libraries as OpenCV have hundreds if not thousands of more or less advanced and easy to follow tutorials, TensorFlow, a popular library for machine learning has a similar (and growing) ecosystem of communities, tutorials, videos etc. Just to make an example, Apple has Machine Learning training achievable via drag and drop thanks to its completely free and well documented developer tools.
DIY warfare platforms can rely on decades of open researches, libraries and frameworks to be given varying degrees of decision making capabilities. Data from sensors, as optical ones, can rather easily be integrated into a more or less centralised “decision making” process (more or crude AI). We have already looked into both the hardware (single boards computers) and some of the software/sensors combinations that can be used for this (IMU, Computer Vision etc ). To then have a our “drone” utilise that data from the various sources and make some coherent us of it, a lot of material could come from something mundane as the …… Video Games development world…
Seriously, I’m not joking ! Video games had to develop increasingly complex algorithms to control the behaviour of multiple NPCs (Non-Player Characters) within them. As computational capabilities of gaming platforms increased, so did the complexity of NPCs behaviours, be them enemy fighter planes in a realistic combat flight simulator or a pack of wolfs in a survival game.
Many sources for Artificial Intelligence related material are available other than video games, but for most uses, popular behavioural libraries used in the video game development world can provide comprehensive solutions for our needs. In general, for most of the conceivable uses suggested by this article, rather advanced “Artificial Intelligence” driven behaviours can be considered trivial to integrate.
VIO – Visual Inertial Odometry
In more recent years the huge success of commercial camera drones, the availability of cheap computational platforms, autonomous car driving related researches, advancements in Image Analysis and the general interest for robotics and automation made it so that even something previously extremely complex and unaccessible as VIO (VIO is what grants positioning information accuracy to the most advanced rovers on Mars, jut to make an example) made its way into the Open Source world with a treasure cove of material available online and even cheap integrated hardware for it.
VIO uses cameras, possibly even in a stereoscopic setup (two cameras emulating human eye separation so to extrapolate 3D informations about the scene being captured), to reconstruct a three-dimensional scene of what the drone “sees” so to correct the common problems brought by inertial movement sensors such as gyroscopes and accelerometers (which tend to loose precision rather rapidly).
What this effectively means is that modern autonomous platforms, even those now accessible to the hobbyists community, have to rely always less on GPS for positioning becoming always more completely independent from external signal sources (the ones which would be easy to negate for those defending a “target” in the context of this article). All this at the reach of everybody with an internet connection.
VIO algorithms can even be used to replicate something as Terrain Contour Matching which was previously exclusive to the most advanced cruise missiles from first world military powers. Terrain Contour Matching used to rely on the radar on board a missile to match the terrain beneath it, while in flight, against known terrain elevation data so to allow for precise positioning without the need for GPS. In 2020, you can easily access weekly updated, with almost global coverage, precise radar data coming from satellites from missions such as SENTINEL-1, data which could allow to integrate VIO with constantly updated 3D terrain maps against which to match the data from the cameras to make positioning even more precise. To all effects, a correct implementation of VIO integrated with the terrain matching would make a DIY drone more precise than current most advanced cruise missiles from a superpower such a US (this altho, with some limits, basic optical VIO requires the weather to be clear, and night flights using a purely optical solution would be somewhat limited by the view range of commercially available night vision cameras). A 2016 example that relates to this can be found at this link , in the linked research case the flying platform uses a LIDAR sensors to gather 3D informations about its surroundings and navigate up to 216 kms, with no GPS and a final positioning error of 27 meters. Nowadays, thanks to the advances in VIO, the rather expensive and energy demanding LIDAR could be replaced by cameras.
A cruder, cheaper, but still impressively effective navigation method from years ago had papers (and source code) published showing gps independent navigation based on simple “Scene matching” which used a camera on board a small drone to match what was beneath the drone with simple images taken from Google Earth and extrapolate precise positioning data. Modern, free and almost daily update decent satellite pictures would allow for far greater precision to the technique.
Another important capability that is somewhat tied to VIO is that of precisely estimating the relative distance of objects from the camera by using a stereoscopic setup (2 cameras mimicking the human eyes separation). Assisted by image analysis algorithms and made precise even at decently long rages thanks to HD and even 4K cameras, the technique can provide precise range assessment that where once reserved to complex, bulky and energy demanding laser or radar based solutions.
SDR – Software Defined Radio
Software Defined Radio uses modern processing hardware to replicate via software many components of a radio system that where previously tied to specific (and often, expensive and/or hard to find in the consumer market) hardware (ex: mixers, filters, modulators and demodulators etc). With SDR, and once more, the open source world behind it, advanced radio transmission techniques become reachable, allowing for a wide range of possible uses. Be it a secure radio network resistant (at least to some extent) to jamming efforts, very long range communications or even pretty advanced SIGINT (Signal intelligence) , SDR, the software, the cheap hardware needed and the community, make it possible.
Put simply a Mesh Network is a decentralised self-healing network where no infrastructure node is crucial for the general connectivity. Multiple routes are dynamically identified by the network and single nodes can be easily replaced maintaining a constant connectivity (up to some extent obviously).
Drone based mesh networks are being increasingly studied and employed to bring connectivity to disaster struck areas. Thanks to this, we have very detailed publications with (sort of) easy to follow steps to build your disaster relief drone network at home, most of them based on Open Source hardware and software (example-1, example-2 )
A Mesh Network is also the core of a drone swarm, once connectivity is established between the drones, their coordinated behaviour becomes sort of easy to design.
Laser based target designation
In many cases, advanced military weapons use a principle designed around a “laser designator” for final targeting. A human on the ground or a manned or unmanned flying platform will basically point a laser beam at the intended target, and such laser beam will then be used by an ammunition to know what and where to hit very precisely. The concept is sort of trivial to replicate with little effort and budget. Powerful IR based “laser” pointers can be purchased online, same goes for IR cameras. The laser beam (and actually, the IR “spot” it creates when hitting the target) can have a custom pulse frequency that a DIY weapons can precisely detect and isolate from other potential IR sources, and trivially guide themselves via what is among the most simple situations for image analysis, up to the designated target with submeter precision.
But…. There is more… Much more
Another important improvement over the past is that of size of weight, what where previously bulky, large and heavy seekers and guidance units can now be replaced by few components weighting few grams each and occupying almost irrelevant volumes, so to free mass for fuel, warheads or additional sensors.
All that we have discussed up till now is just a portion of what is actually available both to purchase or to build thanks to online resources and purely consumer grade products. According to budget and/or “craftiness” of designers, the consumer and Open Source worlds can bring many other enhancements to the attack platforms we are discussing about.
You can shop online for a Ring Laser Gyroscope which for few thousands dollars would grant your design with exponentially improved inertial based positioning awareness, or you can for example, build your own radar ( or even a DIY SAR radar) with a couple of hundred dollars (which could be used for an all weather DIY Terrain Contour Matching solution). One could also integrate in a design a long range laser range finder to obtain precise distance measurements of targets so to enhance targeting capabilities and even allow for precise enough movement prediction calculations to hit moving targets. LIDAR sensors could provide way better spatial awareness to integrate VIO etc etc.
Even if just fast examples all this once more bring features to “DIY” designs that where only reserved to high end military equipment of first world countries up till very short time ago. Another important aspect is that thanks to miniaturisation of components, what once where bulky, large and heavy guidance units and seekers can nowadays be extremely small and lightweight freeing volume for fuel, batteries, warheads and additional sensors.
The open nature of the internet and the multiple, completely innocuous, uses this technologies can power make it so that there is a constantly evolving scenario of systems, solutions, researches and hardware being discussed, described and being made available commercially that could be (almost) trivially adapted to our “Open Source Warfare Platforms”.
Ops… Had almost forgotten, the Airframe
So, we have introduced cheap components and mostly software based solutions for the navigation, powering and targeting of our designs, but building a custom airframe can’t be easy, you might be thinking….
Again, not the case. Thanks to 3D printers (which can cost as little as $100) we can nowadays , easily, print molds that can be used to cast carbon or glass fibre parts that will compose the airframe of completely DIY designs. Complex shapes, composite structural solutions, and advanced aerodynamics are today completely reachable by committed designers.
This site is just one example of the plethora of resources that exist for DIY RC Plane builders to produce the airframe of their dreams, and once you have good enough molds (and/or a good way of replicating them), even building the designs in series becomes absolutely doable.
Among the military world there is increased talking about , and rush to procure, drone swarming technologies. The basics of the concept is that of having multiple autonomous vehicles acting in a coordinated fashion, integrating and even replacing if needed the capabilities of the single components of the “swarm”. A suicide drone swarm could for example be designed to reach an attack area, identify all military vehicles within it, assign some kind of a “weight” or “score” to them based on their characteristics, and have each drone attack a different target prioritising those of a higher value, all this completely automated.
Hundreds of millions of dollars are being expended to research the topic and to most unaccustomed with modern technology, it sounds as some impressively complex concept completely out of the reach of less funded states and/or entities, completely unthinkable for singles or small groups.
Well… that is far from being the case. As for many other concepts and technologies we have seen throught this article, the “Open Source” world can once again come to help, and it can do so in a powerful way. Especially Mesh Network we introduced earlier which provide a well documented technology to maintain connectivity between the drones composing a swarm.
With connectivity handled, even complex cases as those that would arose from loosing a node in the network (one or more of the drones malfunction or get taken down) become trivial and design efforts just need to focus in more or less basic cooperating behaviours. A plethora of resources (one fast, but detailed, example) exist also to handle the relative positioning of swarm units in complex scenarios.
What capabilities can all the above realistically bring ?
Uses can be many but we can isolate 3 main ones that would be decently easy to achieve and design around.
- Strikes: Drones can be designed to release and explosive payload over one or multiple targets or simply be on a suicide mission detonating themselves once over, near or in contact with the target
- Intelligence collection : thanks to the increased quality of consumer grade cameras, radars, LIDARs, thermal or infrared based sensors etc , the intelligence gathering capabilities the, potentially completely autonomous and radio emission less, platforms I have been writing about are impressive. Automated image analysis and the various technologies allowing to estimate range from targets would also provide accurate positioning for future strikes.
- SIGINT collection : Thanks to affordable and well documented SDR tech it is now conceivable to have one or more drones flying high above target areas and collect data about radio signals populating a very large frequency spectrum, basic triangulation algorithms would then allow to pinpoint the source of such signals with obvious targeting implications.
Up to this point of the article we have introduced and briefly described the basics behind generic drone capabilities, components and some core, Open Source and easily accessible technologies that could power the warfare tools we are discussing about.
Will now proceed with a couple of semi “practical” examples, this not to provide a step by step guide to let you build a potentially dangerous design, but rather to solidify the point that such uses are, and will always more be, accessible to very little funded good or bad actors.
Both examples will be rather complex this to show the higher range of capabilities “Open Source” warfare can bring.
The DIY drone Swarm
We can make a fast example of an effective, for warfare needs, “drone swarm” given what we have learned up till now. What I will suggest won’t be really “trivial” to realise, but far from impossible to achieve for a single person given a couple of months (being conservative) and especially realisable by a small team with a very limited budget. Will offer also a couple of crucial options that will make the design much easier if needed.
The target of our example scenario will be a generic military installation.
Our swarm will be composed of multiple, lets say 20, identical quadcopters. We want them identical for redundancy, so that if one or more are intercepted the swarm wouldn’t lose crucial capabilities. To keep the example rather simple our quadcopters will have a limited range of approx 25 kms and a payload capacity of about 1.5 kgs.
Will try to keep the design cheap, but will be conservative with all estimates, and use decent quality components to guarantee reliability (especially given the fact that our “drones” need to last one flight only).
Shopping online I got the following costs for the components of each of the drones composing our swarm.
- $240 for the motors, battery, esc, propellers and frame that will grant our drones a total static thrust of 8Kgs (conservative power ratio should be 2:1, so with 8Kgs of max thrust we will be able to have the total takeoff weight of around 4 Kgs). This are the basic components that will keep our design in the air.
- $26 for a fully independent (but easy to connect to other components) integrated Open Source autopilot solution (based on the Open Source ArduPilot), the component comes with a IMU including digital compass, gyroscope and accelerometer.
- $8 for a decent enough GPS receiver, might not even need it, but why not…
- $110 for an Nvidia Jetson Nano, a single board computer specifically designed for image analysis, image classification and neural networks
- Around $120 for 2 HD cameras that can be used for VIO (optional – can be replaced with a single $30 hd camera module)
- $300 for a high resolution thermal imaging sensor with >300 meters of range mounted on a custom gimbal solution. (optional)
- $100 for an IR based optical positioning and identification system to gather crude relative positioning among swarm units (optional)
- $70 for the latest model of Raspberry Pi
- $40 for frequency hopping capable networking components (for this case 5 or 2.4 Ghz transceivers)
- $40 to include smaller components that might be needed , antennas, wires, stuff in general
The hardware composing each of the drones will cost, according to our design choices between $564 and $1029, and the total weight up till now should be around 2 Kgs, leaving us ample margin for our 1.5 Kgs payload. Our drones will be able to safely take off and fly weighting 4 Kgs and should be able to travel for about 20 minutes at 80-90 Km/h giving us a range between 25 and 30 kms (especially since our drones will be fully autonomous so we don’t need to factor the range of the remote controlling equipment which is what limits most commercial drone designs).
I have no idea about the price of military grade explosive, and suspect that it would be subject to huge differences according to where and how it is procured, in this and following examples I will totally omit it, so consider that an X amount of extra budget has to be dedicated to that if designing drones that need to explode.
In this design, the Raspberry Pi is used to handle the mesh network created by our swarm, a custom Linux install with BATMAN-adv (popular Open Source Mesh Networking protocol) plus the 2.4 or 5Ghz transceiver (plus some custom code) should grant our drones decent enough connectivity among each others with a maximum separation between each of them of about 100 meters (being very conservative ). The Raspberry will have more than enough power left to act as the main brain of the design, integrating the data from the other components and taking basic decisions about its behaviour.
According to ability of those designing this platform, time constraints on other factors, we can choose if to integrate VIO into our design or not. With it, we can be GPS independent and still have have very accurate positioning, but a correct implementation, at least the first time, will take time to develop and test. Multiple resources are available online which will make such work possible even for non academics or professionals, but most “ready to go” solutions are designed for interior navigation, and according to attack location and the scenario surrounding it, more or less work will be required to develop a good enough solution . To be noted that such work will then be re-usable in future designs so…..
The Nvidia Jetson Nano will be tasked with handling VIO if we chose to use it, and more importantly final target recognition and possibly even some crude BDA (Battle Damage Assessment) thanks to the thermal camera. The Nvidia board will have a machine learning model trained to classify the targets we want to hit, in our example those could be fuel tanks, particular buildings, specific vehicles, or even have it recognise military uniform wearing personnel. For each target we can assign a “score” which the crude AI driving our swarm will use to prioritise which targets to hit.
The “intelligence” / logics driving our swarm will be pretty easy :
- One and only one drone will always have the role of leader, if the leader was to go offline for some reason, another random drone in the swarm will take the role.
- The Mesh Network will be activated only once reached a certain waypoint close to the target so to minimise detectable radio transmissions.
- Once the network is active, the leader will asses how many drones made it to the target area, perform a “pop-up” manoeuvre (basically, climb to gather a better field of view, we want to have our drones flying as low as possible up till the final moments of the attack) and use its camera and image analysis to identify possible targets.
- Targets will then be prioritised and distributed among swarm units, with some crude relative positioning being processed by the leader via image analysis.
- Once swarm units have received their target informations, they will move towards the target, gain some altitude, confirm target identification via onboard image analysis and track the target still via image analysis, moving towards the target at the highest speed possible.
- When leader detects to be the only drone remaining in the network, it attacks too.
- BOOM !
Advanced logics could see drones attacking in small groups, something like 3 , have the leader pop-up after each “wave” , use thermal camera to identify which targets where actually hit, and schedule next wave of attacks based on which targets are left and their “score”.
As per this example, we would have hardware costing (for the full 20 units swarm) something between $5.4K and $20K (plus cost of explosives), capable of carrying out a rather complex attack taking out multiple high value targets with great precision.
Such attack would also be rather hard to defend against, jamming would be of little use as even the mesh network wouldn’t be easy to jam due to the proximity of the units and frequency hopping in the transceivers which would require a very powerful jammer. Even if such jammer was to be available, and powered at time of attack, multiple ways could be employed to minimise its effects, the network could be for example designed around an optical infrared based communication protocol, or more simply, drones could each have a “preferred” type of target defined at launch time (by providing different scores to the targets), each drone not detecting others would have the leader role and carry it’s attack autonomously.
Recent VIO progress and/or LIDAR technology would also make it conceivable for the units of our swarm to take off and navigate in a densely urbanised area with the units flying between buildings with relative ease, further making defending against such attack extremely hard.
Would this be easy to realise ? No, not really, but it also would be far from impossible for a relatively very small team or even just a skilled single DIY engineer. And, everything described above relies solely on consumer grade components (except for explosives) very easy to procure online in most of the world.
Many modifications could be made to improve this design, reduce its overall costs, increase it’s efficiency etc. The quadcopters used in the example could be easily replaced by gasoline powered fixed wing units capable of carrying a much heavier and larger payload at much increased range, swarm communications could be made more robust and with a greater range etc etc…
The long range “cruise missile”
There was a lot of clamour when in September 2019 two ARAMCO oil processing facilities where attacked in Saudi Arabia. Much of the surprise came from the precision of the strike and the apparent failure of defences against it. Even if hypothesising a direct Iranian involvement, the attack still looked more effective, precise and focused than what most considered Iran capable of.
Long or medium range cruise missiles where exclusive to few first world powers up till just a few decades ago, and especially their precision wasn’t really great for most designs safe for recent, and extremely expensive ones.
So, the Houthis or even Iran capable of having missiles flying hundreds of kilometres at low altitude to then hit precisely targets just a few meters wide, caused quite some shock.
“Cruise Missile” is a term nowadays open to interpretations, a “suicide” drone, however it is powered can be considered a Cruise Missile, the core principle is that of a flying platform that once set a target, and eventually a specific route to reach it, it will autonomously fly to it and detonate a warhead either once it makes contact with it or in its immediate proximity.
Will try now to illustrate some design options that would in most cases, be even more precise that those used in the ARAMCO attacks.
Our “missiles” will have a relatively limited payload of 40 kgs (excluded the navigation system), and have a range, according to options chosen, between 200 and 1500 kms. According to needs, the designs could sacrifice some range in favour or a larger warhead etc, but the 40 kgs of our design goal would be well more than enough to replicate the damage seen in the ARAMCO attacks. And in general, the more precise a weapon is the less destructive it needs to be (for most cases at least).
Most of the components from previous example design, as per navigation and targeting, can still be used effectively for this example, with two main aspects that would have to be handled differently, requiring more time and expertise (which can be acquired in large part via easy to follow online tutorials).
- Power source : according to budget and expertise, we can either choose a gasoline powered, propeller based engine or even a turbojet one, prices will obviously swing wildly among the two options. For our design, a good compromise between costs and performances for the propeller option comes from small motorcycle engines, a 200cc, decent enough engine, can be found on Alibaba.com for as little as $150, new. If we where to need speed and budget wasn’t a problem, a consumer grade turbojet engine, meeting the requirements of our design, will set you off something between $6K and $50K but grant the design a top speed up to 800 Km/h
- Airframe : given the power source and weight the platform will have, the airframe has to receive some decent attention. Wood is a cheap and, for some, easy to craft option, but carbon or glass fibre based designs would have greater strength and more importantly, be easier/faster to replicate. Carbon fibre or aluminium rods which can be easily procured will then grant the additional structural strength where needed (along the wings etc). According to expertise and tech used, the single airframe can cost anything between $100 and $500.
Given the platform size and weight, and according to budget, a ring-laser gyroscope, about $2K from China, would help easing navigation system complexity granting precise enough inertial measurements to guide the “missile” from wherever GPS signals stops being available up till the final target area where optical / ir / thermal targeting would take over.
Using previous example items for navigation, we end up with a unit price of something between $500 and $60K for something that, according to budget and expertise can be just as capable (flight performance and precision wise, not warhead in our example) as one of the latest iterations of cruise missiles employed by US, and even in its most expensive version, it would still cost about 1/20th of what US pays for the capability with nothing coming from sources other than the consumer world and based on resources freely available online.
According to the choices made, the final design will weight something between 150 and 400 kgs, carry a 40 kgs warhead, have a range between 500 and 1400 kms, reach speeds between 120 and 800 km/h. It would be able to reach its target in a contested environment, thanks to its low flying capability, decent route planning and being able to fly for many kms with no gps signal. As per previous example design, some swarming capability could be included, requiring altho more effort on the connectivity and behaviour side given the higher speed and potentially larger distances separating each unit of the “swarm”.
Compared to previous example design, the heavy fixed wing nature of this one introduces a new problem which is that of how to get the platform in the air. Either a long enough “runway”, catapults of sorts (as those used by manned gliders for example) or even a rocket based first stage would be needed for this, and best choice is dependant on many factors well out of the scope of this article. Point is, it is an additional problem, but it isn’t an insurmountable one.
Many possible uses, designs, and technologies didn’t make it into this article, but what there is should be more than enough to achieve this writings goal, which wanted to be that of making readers aware of the rather impressive capabilities modern consumer market, Open Source hardware and software and online based learning resources and communities can be bring to modern warfare.
(DIY hobby rocket world, to make one last scenario example, could allow to design a very effective “shortish range” ground launched anti air missile capable of hitting if not fighter jets, at least attack helicopters with relative ease)
This both to prepare for near future warfare or even terror based scenarios and to underline how first world military procurement in many cases is based on old concepts that are becoming always more of an impediment to programs success.
In general, the bourocracy and thankfully the relative lack of need of speedy improvements driving the global military world (not so many conflicts between modern military powers), has made it so that consumer technology has evolved faster than military one in many fields.
The closed nature, the bureaucracy and possibly corruption behind large militaries procurements, lead to programs and equipment that are completely, and dangerously, out of touch from the technological reality we are living in. Obviously this isn’t always the case, but for a wide range of current programs, it is easy to notice how, faster, more economical and effective, designs would result if based on a more complete understanding of modern consumer and open source technologies.
All this also to design and deploy effective measures, protocols and hardware to counter the capabilities that more or less funded entities will increasingly be able to achieve.