war on the rocks

Loyal Wingman, Flocking, and Swarming: New Models of Distributed Airpower

February 21, 2018

The Air Force’s 2016 Flight Plan for small unmanned aerial vehicles (popularly known as “drones”) describes the beginning and the conclusion of a road to swarming — starting with loyal wingmen, and ending with the Ender’s Game swarm where “a single individual that is able to control a massive number of platforms, thousands of platforms at the wave of a hand.” Toward that end, what are the next technological waypoints along the flight plan to true drone swarms?

The true promise of drones is their ability to combine distributed aviation with distributed computation. The former allows for the fielding of large numbers of aircraft and the latter makes them controllable. Controlling swarms might be understood as a command problem, and command can be understood in terms of echelons: one can give specific commands to a partner on a fire team, but leading a platoon requires a bit more abstraction, and leading a regiment requires much more abstraction. Here, we sketch out three waypoints, which roughly map onto these echelons:

  • Loyal wingmen, where one or a few full-sized remote aircraft takes on traditional manned wingman roles;
  • Flocking, where a discernable number of aircraft execute more abstract commander’s intent provided by aviators;
  • And swarming, where future aviators command large numbers of aircraft in the aggregate entirely through algorithms.

While pilots have controlled aircraft from afar and have been launching weapons beyond visual range for quite some time now, the idea of flying as a formation to control or employ automated aircraft presents a bigger cultural and training challenge. This challenge forces us to re-examine some of our assumptions about how aviators should fly and think about flying, both in terms of capturing and preserving the best of traditional aviation, and in thinking about how to leverage emerging capabilities. This history of naval aviation in the 1920s and 1930s provides a useful analogy. Adm. William Moffett, the architect of naval aviation, shepherded the Navy through several iterations. While early naval aircraft were more forward scouts and artillery spotters — merely assistants to the weapons of their sending craft — later carriers shifted the locus of combat to the aircraft themselves. However, as carriers air wings grew in size, the capital ship was forced to relinquish more control to the aircraft and provide more general commands.

These same two trends seem likely along the road to swarming: a migration of the fighting effort toward the swarm itself, and an increasing abstraction in commands given to the swarm as it grows in size. As was true in carrier aviation, there are tradeoffs at each waypoint. All else equal, a swarm of small craft will not be able to keep pace with a launching craft. While energy density in batteries has made leaps and bounds, barring some equivalent propulsion revolution, tiny propellers are limited in speed by their size. Small numbers of more capable craft are better able to keep pace with a manned fighter, but are more vulnerable to being overwhelmed by incoming fire. Similarly, larger craft enjoy economies of scale on payload weight, but the loss of one aircraft greatly reduces the formation’s striking power. The more expensive but less numerous aircraft might better resist jamming with more advanced and resilient datalinks, but their complexity might render them more vulnerable to hacks. From this early vantage point, the contours of the measure-countermeasure battles along the road to swarming are not entirely clear. Still, it seems apparent that each waypoint will have strengths and weaknesses vis-à-vis the others, and therefore the future should be comprised of a mix of these different capabilities.

Waypoint Zero: ‘Fly-by-Wireless’ Remote Aviation Technology

The main prerequisite for a launching-offboard relationship is the ability to control an offboard aircraft. The first step on this road is then ability to control a craft through proxy, to project a pilot’s will and intent from one place into another. This remote aviation technology, or “fly-by-wireless,” technology consists of three essential elements, all remarkably ordinary and universal: an aircraft, an autopilot, and a datalink. While the “fly-by-wireless” Predator and Reaper loom largely in “drone” zeitgeist, they lack the processor speed to incorporate artificial intelligence or any on-board decision-making capability beyond the autopilot ubiquitous to all tactical aircraft. While an effective strike aircraft in their own regard, these craft are more remote-control aerial sniper rifles than futuristic killer robots. For most intents and purposes, Reapers are simply attack aircraft.

Remote aviation is a prerequisite for swarming, so this waypoint demonstrates key technologies, but craft such as the Reaper lack the computational power for swarming, and therefore have more in common with traditional aircraft than future drones. There are a number of applications where remote piloting is relatively advantageous: extreme endurance flights, as in the case of the Predator, or high-risk flights where datalink surety can be assured, as in an optionally-manned fighter. For operations into the most contested areas imaginable, where air, cyber, and the electromagnetic spectrum are all in play, remote aircraft will require more advanced processors, capable of receiving and executing more abstract commands under the assumption that connectivity might be lost for minutes at a time.

These advances pave the way to the first waypoint, as the same automation that allows the remote aircraft to execute simple tasks with little human direction also allows them to serve as wingmen. Increasing the onboard compute changes the game for these craft — with effective artificial intelligence, or even just advanced algorithmic autopilots, these craft can do much more for themselves. Retrofitting legacy manned aircraft with remote aviation technology and artificial intelligence achieves the same effect. With the on-board ability to execute simple tasks, these craft no longer require a dedicated crew. The craft can then be set off on its own to pursue relatively simple tasks (move cargo from point A to B, bomb a fixed target, shoot a radar-guided missile at an enemy), or it can serve as a wingman for another aircraft. This latter role is the first major leap toward swarming.

Waypoint One: Loyal Wingmen and Loyal Scouts

This first waypoint is where the command aircraft pairs with an off-board aircraft that serves as a wingman or scout. This gives the command aircraft additional weapons, sensors, and doubles the enemy’s targeting requirement. Whilst more partner (or perhaps sidekick) than proxy, this aircraft does not do much thinking for itself. The command platform bears the weight of the combat effort and deploys capabilities from the loyal wingman to plug gaps or augment the command aircraft’s course of action. This involves close control of specific actions, above the level of control inputs, but below the level of simply passing the “commander’s intent.”

The form of this waypoint is still taking shape, but will likely look like a traditional formation, at least in the air. An F-35, leading one or several automated F-16s, serves as an example of this — as demonstrated by the Air Force Research Lab’s Have Raider program. A number of prototypes of purpose-built loyal wingman craft have taken flight, most notably Air Force Research Lab’s UTAP-22 Mako, with more in the works at DARPA and DIUx. Other countries are already carrying out loyal scout operations on the ground, for example, in the form of Russian armor formations that fought in Ukraine. The traditional primary weakness of a tank is visibility — even with optics, a crew ensconced in metal will have a difficult time maintaining the all-around situational awareness of an infantryman. A quadcopter or an octocopter changes that equation, as the tank can launch a “loyal scout” to sweep forward and identify targets or patrol around the tank itself. This relationship could be easily formalized with datalinks, to incorporate the fire-spotting function into the tank’s fire control system, or even arming the quadcopter to enable it to neutralize rocket propelled grenade-armed light infantry units.

This latter case is remarkably like the role of early naval aviation, where a floatplane would serve as an artillery spotter aircraft, carry light weapons, scan for periscopes, and do various tasks that helped its command ship (cruiser or battleship) better employ weapons. The organic helicopters on modern destroyers continue this tradition, and demonstrate the value of the loyal scout model. Sub-hunting with two sensors, especially one that can move and is invulnerable to torpedoes, is far easier than with a ship alone. In the same way, these loyal wingman and mothership hunter-killer teams will likely become an enduring feature of future aerial warfare in some form.

Training to employ airpower with the loyal wingman or loyal scout model is much like traditional training: You learn to trust your partner, you become accustomed to their responses, you learn how to communicate and you build tactics that rely on mutual support. Since the loyal wingman has unique capabilities, the command pilot must learn these capabilities as they would with their own aircraft. There is a relatively low cultural and conceptual bill to transition to the loyal wingman or loyal scout model, as it maps directly onto to the relationship with manned wingmen.

Waypoint Two: Flocking

The second waypoint is flocking, which is distinct from loyal wingmen because the command aircraft no longer exercises direct control over the inputs of the aircraft in the flock. However, the command aircraft can still identify discrete elements of the formation, and can still command discrete effects from the individual elements. A typical flock might run from a half-dozen to two dozen aircraft. The weight of combat effort is shared between the command aircraft and the flocking aircraft, with lighter weapons on the small aircraft and heavier weapons on the larger aircraft. Since advanced artificial intelligence algorithms rely on larger processors, typically decks of graphics processing units or tensor processing units, to carry out deep analyses, the command aircraft would also host computational reach-back capabilities for questions too involved to be solved by the command aircraft’s processing power alone.

Since there are a number of aircraft in the flock, it makes sense to spread diverse capabilities across the flock — perhaps a mix of sensors and weapons, where flock elements can swap positions in order to leverage each other’s unique strength. Still, these specialized functions are still relatively self-contained within each element, and each can act with some degree of independence. The pilot or crew at the heart of the flock needs a good understanding of the capabilities in the flock, the flocking formation algorithms themselves, and how to issue mission-type orders to the flock.

Following the naval aviation analogy, flocking is much like fighters operating as scouts and defenders of a heavy “aviation cruiser,” where the ship itself carries additional heavy weapons. In a flock, the combat workload is roughly balanced between both the command aircraft and the rest of the flocking aircraft. With flocking, there remains an imperative for the command aircraft to close towards the enemy in order to bring its own weapons to bear.

Flocking is the waypoint not yet within reach, but no longer the province of imagination. While the defense industry has already demonstrated loyal wingmen, the military should be planning and programming for flocking. Little to nothing in this vision of flocking involves technologies that are not already at relatively high technology readiness levels — quadcopters, small arms, datalinks, cameras, basic formation and three-dimensional positioning, machine vision, are all on the market or soon will be. It is more a matter of putting these things together and marrying them to aviation culture. A shotgun wedding though it may be, this needs to happen quickly, because if we can easily imagine flocking concepts of operation, then surely our enemies can too. DARPA’s CODE program, which enables collaboration between unmanned platforms to avoid pop-up obstacles, is a strong early contender for a flocking capability.

Waypoint Three: Swarming

The third waypoint, swarming, exceeds the complexity of flocking, so that a pilot cannot know the position or individual actions of any discrete swarm element, and must command the swarm in the aggregate. In swarming, much like fleet carriers, the swarm elements do the bulk of the combat work. As a rule of thumb, a combat formation with more than a hundred elements is best understood and managed as a swarm. Such a world could be understood in terms of densities of distributed capabilities, but engagements would likely be too complex and too fast for any direct control. We already see these principles and phenomena at work in the AEGIS defense system, which handles an inhuman volume of fire following rules set by humans.

The fleet carrier is the naval analogy of a future swarming craft, as it fights entirely through its off-board assets, and there are enough off-board elements to require mission-type orders rather than discrete direction. Moffett challenged the naval tactics of the time by pulling carriers back from the line of battle, where they could fight using their air wing as their primary weapon. In the same way, the command aircraft in a swarm would focus primarily on hosting and deploying or employing the swarm. During combat, the primary role of the command aircraft is to place human judgment and creativity at the core of the swarm’s algorithms to enhance survivability, just as the carrier battlegroup’s admiral provides leadership guidance during combat operations.

The combat swarm is more abstract and more reliant on immature technologies than flocking. We could imagine at least three implementations of the concept. The first is a command aircraft, in a scaled-up form of flocking. The challenge here is the economies of scale in propulsion systems — to go fast, at least to go fast for an extended time, one needs a relatively large and expensive engine. If a swarm has a hundred or a thousand elements, then this becomes a problem. Quadcopters, however, can affordably transit at low speeds, so if the command aircraft uses its “velocity economies of scale” to get a swarming payload near a target, the swarm could fight the rest of the way in. For this second implementation, the swarm as an ultra-modern cluster bomb, seems much more feasible with current technology (which pairs well with a defensive flock escorting the command aircraft). Lastly, a defensive persistent swarm could be flown from the ground near high-value targets or frontlines — this would likely be the best way to defend against an attack from a hostile swarm.

While the swarm will be very reliant on algorithms, there remains a role for human judgment and creativity in its employment. Even in the selection of the type of swarming algorithm there will be both judgment and intent. By setting densities or fundamental attack axes, the human penchant for n-dimensional thinking and envisioning future states can supercharge the coordination skills of the swarm. Moreover, humans can employ empathy against an opponent — even if you have to dig back to the engineers, there is a human behind an enemy swarm, and our humans can use insight and instinct to exploit the flaws in that human’s thinking. To operate well in this space, the Air Force will need to refocus on the things that are truly uniquely human, and divest most technical tasks to technology. The Air Force should also re-learn mission command in order to lead these forms of technology, the precursors of which are already upon us — the Strategic Capabilities Office’s Perdix drone can already launch from a fighter’s flare dispenser and establish basic swarms.

Conclusion: Flocking as a Bridge

The Air Force has already framed the concept of loyal wingman well, and it has painted a plausible picture of what swarming could look like. Flocking is the bridge between these two concepts, as it allows aviators to grow into the concept of leading swarms through algorithms by managing distinct and differentiable off-board capabilities. While the concept of loyal wingman is increasingly within our grasp, flocking is the concept that is almost within conceptual reach, if not yet within grasp. It provides a bridge between demonstrated loyal wingman technology and future swarms. Here is a picture of that bridge.

Upon crossing “feet dry” into enemy territory, a future attack-assault helicopter launches a dozen recoverable skirmisher drones. Each capable of keeping up with the helicopter at normal combat speeds, and each with multi-spectral imagers and an integrated piezo-electric 9mm anti-personnel pistol, these “skirmishers” spread out in a fan in front of the on-board craft following the pilot’s formation play-call. Scanning for ground anti-aircraft threats (or other drones), they sweep along optimal ground paths, as the pilot sets “weapons tight” with a pre-loaded rules of engagement and threat priority stack developed during mission planning the previous day.

Finding a ZPU-4 anti-aircraft gun, one skirmisher requests permission to engage, and when the pilot grants that permission, it fires a round into each member of the gun crew, neutralizing the threat. Another element finds an infantry fighting vehicle, and the crew of the vehicle finds the drone as well. It maneuvers laterally, following its defensive logic of visually locking onto a gun muzzle and maximizes the angle between itself and the mouth of the barrel, with a few jinks thrown in for good measure to complicate a small arms solution. Relaying the position of the threat to the mothership helicopter, complete with a targeting solution, the pilot sees a target picture and a firing solution projected onto her helmet-mounted display and approves the engagement. In the course of three seconds, the aircraft couples the autopilot to the skirmisher’s firing solution, releases an anti-tank missile, and returns to the previous heading. The skirmisher provides terminal guidance over the flock’s tactical local area network to the missile and the mothership, and reports a successful engagement.

Arriving at the target landing zone, the flock executes its pre-planned survey and clear program. Four skirmishers land in each quadrant of the field, angle their cameras slightly up, and conduct a 360-degree pan for obstructions missed on the imagery. Finding none, half of their number moves to sweep the tree lines and potential ambush sites identified during the prior day’s mission planning, and the other half closely scrubs the landing zone for any mines. Fifteen seconds after this all begins, the assault helicopter follows the green “free of threats” flight path set out for it by the skirmishers, and touches down safely.

In conclusion, these four waypoints described a flightpath toward drone swarms, where each step anticipates more of the weight of combat effort shifting toward increasingly distributed off-board craft. This does not imply that each step replaces or overwrites the previous step. Technical evolution gives rise to different branches of capabilities, and many of these branches persist even as the evolution continues. The branches that survive learn to co-exist in a technical ecology, even if that ecology has apex predators.

Returning to the historical parallel of Moffett, early naval aviation launched several families of capabilities. Early forays into flying boats gave rise to the PBY Catalina and the maritime patrol mission, which is carried on to this day by P-3 Orions and P-8 Poseidons. Follow-on efforts with escort floatplanes such as the Vought Kingfisher generated a legacy carried on by the cruiser and destroyer based SH-60 community. The World War II legacy of light carriers persists in the Marine Harrier jets and amphibious assault ships. The Nimitz and Ford class supercarriers are the apex predators of naval aviation; while they dominate the ecology, they do not uproot the previous branches.

If history is a guide, all four of the generations described here will evolve along their own complementary lines. Fly-by-wireless, loyal wingmen, flocking, and swarming craft will likely co-exist in one such ecology — for instance, imagine a future mothership-bomber protected by a defensive flock, designed to employ small drone swarms in lieu of cluster munitions. So if there is one lesson to be learned from the saga of normalizing the Predator and Reaper within the ecology of Air Force airpower, it is that capabilities will find their niche if given time and proper support. They then find ways to co-exist with previous capabilities, which are themselves evolving. As they do so, the binaries of humans/machines and aircraft/drones will give way to a more nuanced view of automation and a more textured understanding of the relationship between people and their tools.


Lt. Col. Daniel Wassmuth is an Air Force officer who works in the Pentagon for the Joint Staff J-3. He is a Senior Pilot and a Graduate of the U.S. Air Force Weapons School. Prior to the Joint Staff, Wassmuth served as a Strategic Policy Fellow in both the Undersecretary of Defense for Middle East Policy and in the CSAF’s Strategic Studies Group.

Lt. Col. Dave Blair is an Air Force officer who works in the Pentagon on Special Operations aviation policy and technology. He is a Senior Pilot, and served as an Operations Officer in a SOF MQ-9 unit. He is a graduate of the U.S. Air Force Academy and the Harvard Kennedy School, and holds a doctorate in International Relations from Georgetown.

The views expressed here do not represent those of the U.S. Air Force, the Department of Defense, or any part of the U.S. government.

Image: U.S. Army/Christopher Warner