Designing for Doctrine: Decentralized Execution in Unmanned Swarms Published July 23, 2021 By Capt Thomas L. Enloe Wild Blue Yonder -- Introduction The development of automated systems in air applications is rapidly accelerating with many applications and levels of autonomy afforded by technology. As Air Force doctrine shifts, autonomous architectures must also follow a centralized command / distributed control / decentralized execution model in the same way their operators do. The computational power of digital systems has and continues to expand at an exponential rate. It is tempting to develop fully integrated systems that take advantage of all available data, but such system networks can become susceptible to over-dependence on external inputs for execution of simple functions that are taken for granted in manned systems. Designs for current and future autonomy must actively search for opportunities for decentralized execution of critical system functions associated with widespread use in applications such as Unmanned Aircraft System (UAS) swarming. This will involve shifting system requirements to focus on a true system of systems architecture, manipulating the behavior of the individual to influence the greater behavior of the whole. A Comment on a System of Systems Model Before delving into the specific design considerations of a UAS swarm, it is important to differentiate between a complex system and a system of systems framework. Put simply, a complex system is a culmination of many components relying upon each other to achieve a solitary goal. The subcomponents in such a case serve no true purpose separate from the whole, and the overarching complex system cannot function properly without each subcomponent.1 In a system of systems, however, each piece is self-contained, serves its purpose, and can operate independently, but the culmination and interaction of these systems serves a higher-order goal as well. The higher-order behavior of this formation can then be optimized by manipulating four primary factors: resources, economics, operations, and policy.2 Using this definition, an AGM-114 Hellfire air-to-ground missile is not a system, and neither is an MQ-9 Reaper carrying it. The UAS system, on the other hand, requires not only these two components but also the ground control station and communications links.3 It is only when multiple MQ-9 aircraft operate together that it becomes a system of systems. The behavior of this group will change naturally with conditions, including the number of aircraft or crews (resources), funding for flight hours and maintenance (economics), formation and flying procedures (operations), and rules of engagement (policy). The Instinct for Integration In the civilian sector, UAS swarms bring intricate dances of light to the skies over Super Bowls and the Olympics.4 To accomplish such a feat, each drone must know exactly where it must be and be fed accurate location data to constantly correct its position in the dance, giving the illusion that each drone is in sync with its neighbor. The effect is beautiful, and it is easy to imagine a military application for such a disciplined group of flying machines, all working together to fulfill a common goal.5 Instead of hundreds of UAS working together, however, this configuration has one central system feeding data to each pixel, controlling the movement of the group one step at a time, whether it be preprogrammed or ad hoc. This centralized execution, micromanaging each drone down to the inch, is possible with advanced computing and can even be preprogrammed; however, it runs counter to Air Force doctrine. This model does not allow for mission-type orders to be given to a group, but rather requires a specific order to be constantly relayed to each member of the swarm. It is not only inefficient but also inflexible without continuous modification of complicated software. Figure 1. Interconnectivity-based UAS Swarm architecture The instinctive next step, then, is to link each UAS in the formation together, sharing information and cutting down the command-and-control (C2) chain. In this information age, everything is connected, and network-centric warfare has become a topic of great discussion. Extensive interconnection between systems has become second nature, and integration in military systems reflects this desire for expanding information. Operating with the idea that more information is better in all cases, the design of future systems is based on a centralized data flow.6 Data links constantly push information back and forth from asset to asset, sometimes to the point where one becomes over reliant on the information and cannot function without continuous updates. In a UAS swarm, however, such a configuration is at considerable risk of failure in a degraded C2 environment. This is no argument against the benefits of information sharing and networks. Information sharing, especially between dissimilar platforms, allows for synergistic effects—another core tenet of airpower doctrine. In either an individually controlled or linked-controlled configuration, however, the swarm still relies on continuous external input or consistent inter-communication channels to function properly. The individuals do not behave as individuals, but rather as subcomponents. This drives the entire swarm along with its command node, to be one solitary system throughout the operation. As such, the drones themselves lose an advantage naturally found in manned systems: self-reliance. Driving for Decentralized Alternatives Many natural actions made by humans in difficult situations are taken for granted. While an aircraft without GPS can fall back onto inertial navigation, a pilot without GPS can follow a coastline or even analyze the stars. Similarly, a pilot may fly in formation as instructed, but need not be told to avoid colliding with the flight lead. This inherent ability to act in a self-reliant capacity is the central capability that enables the doctrine of decentralized execution. This capability is not limited to manned systems, though it does require the ability to operate in an information vacuum. In an autonomous system, establishing such self-reliance is a matter of operational configuration. A UAS equipped with inertial navigation and knowledge of its starting point, along with an up-to-date map and an altimeter, can perform the basic aviation required to transit from point A to point B and back again within a certain margin of error. Such a maneuver still includes a human “in-the-loop” to give initial orders, but no external interaction is required outside of take-off, approach, and landing data, not unlike that which is currently used under instrument flight rules. In fact, artificial intelligence image recognition may render the latter obsolete, as an autonomous system could recognize a landing zone via onboard camera sensors. Similarly, an effective military UAS swarm only requires navigation and deconfliction parameters. Desired swarm operations closely resemble bird behavior algorithms built by Craig W. Reynolds, referred to as the boid model. In this algorithm, each member has three steering behaviors: collision avoidance, velocity, and maintenance of proximity to other boids, also known as flock centering.7 By controlling these behaviors in the individual, the behavior of the flock can be controlled. The flock then becomes a system of systems, with an overarching controllable emergent behavior. Figure 2. Swarm architecture using self-reliant UAS with deconfliction behavior Similar configurations can be used to establish flexible, self-reliant swarming capabilities, especially in an airborne area-denial role. The desired swarm behavior can be situationally established before or during the mission by manipulating any or all of the four system of systems factors, be it the total size of the swarm (resources), fuel consumption limits (economics), preprogrammed maneuver or avoidance techniques (operations), or directives to skirt but not cross another third-party nation airspace (policy). Regardless of the specific parameters, however, the swarm is resilient against the loss of individuals as no member of the formation is fully aware of all others. Additionally, because the individual is not reliant on continuous inputs, each drone can take mission-type orders and execute with its peers accordingly. This greatly increases the flexibility of use and enables continued operations even in communication-denied areas. Rewriting Requirements It is important to remember that self-reliant capabilities do not preclude precision derived from data sharing and external signals. GPS location and altitude data are valuable and any UAS swarm should still be controllable and re-routable mid-flight. The key is to retain functionality and possibly divert some processing power from ground stations to simplify communication signals. Many of the tools required, such as a magnetometer for direction and a barometer for altitude, can be found on the average smartphone. These sensors are reasonably accurate too. Implementation of such a configuration is primarily determined by system requirements within the development phase. It is good to keep technical requirements simple during acquisition, as it typically allows development companies freedom in their designs. However, in this scenario, it is unlikely that self-reliant autonomy will be fully implemented unless it is specified in the requirements stage. This is partially because people think of drones as remotely controlled systems, both in commercial and defense industries. It is also partly because designers and senior military leaders alike desire precise and predictable execution. UAS swarms will not become an agile and resilient military capability until planners prioritize self-reliance and operations in degraded and denied environments. Until such a time, “the swarm” will fail to live up to the lofty promises of its advocates. Conclusion The military leader is bred to crave precision and control. Every step of formal training is built upon being prim, proper, and precise. War, on the other hand, is none of these things. It is chaotic, divisive, and flat-out ugly. Future UAS swarms, and all future autonomous systems for that matter, must be ready to deal with chaos. The US Air Force does not need centrally controlled drones flying rank-and-file into the battlespaces of the future. Instead, future UAS swarms should focus on rule-based local behaviors for aviation, position keeping, and environmental response to deliver airpower with improved resilience and flexibility, even in degraded C2 environments. The execution may not be as precise, but perhaps it does not need to be. War is chaotic, and no technological advancement will stop that. Self-reliance, however, breeds resilience, and resilience wins wars. Captain Thomas L. Enloe Captain Thomas L. Enloe is a developmental engineer currently assigned to Wright-Patterson Air Force Base, Ohio. He had held technical positions in aircraft systems, software, and information technology. Capt Enloe holds a B.S. summa cum laude in Mechanical Engineering from the Missouri University of Science and Technology, and is currently pursuing a M.S. in Aeronautical and Astronautical Engineering from Purdue University. He has also served one combat tour in Afghanistan, supporting Operation Freedom's Sentinel and the NATO Resolute Support mission. This paper was written as part of the SOS Air University Advanced Research (AUAR) elective, Ideas and Weapons section. Notes 1 John Boardman and Brian J. Sauser, “System of Systems - The Meaning of Of,” 2006 IEEE/SMC International Conference on System of Systems Engineering, 2008, 118–23, https://doi.org/. 2 Daniel DeLaurentis, “Understanding Transportation as a System-of-Systems Design Problem,” 43rd AIAA Aerospace Sciences Meeting and Exhibit, 10 January 2005, https://doi.org/. 3 U.S. Air Force, “MQ-9 Reaper,” 23 September 2015.https://www.af.mil/. 4 Brian Barrett, “Inside the Olympics Opening Ceremony World-Record Drone Show,” Wired, 9 February 2018, https://www.wired.com/. 5 “Drone Light Shows Powered by Intel,” Intel, Accessed 6 May 2021, https://www.intel.com/. 6 Clement C. Chen, “Anatomy of Network-Centric Warfare,” Signal, August 2003, https://www.afcea.org/. 7 I. Lebar Bajec, N. Zimic, and M. Mraz, “The Computational Beauty of Flocking: Boids Revisited,” Mathematical and Computer Modelling of Dynamical Systems 13, no. 4, 30 August 2007, 331–47, https://doi.org/.