Drone swarms represent a cutting-edge and thrilling development in drone operations. By leveraging the collective capabilities of multiple drones, swarms can perform complex tasks more efficiently than individual drones. Many people might have already been exposed to swarms in the form of drone light shows, but as this article will discuss, this is far from the current state-of-the-art and is only the tip of the iceberg.

Over the past few years, this topic has become one of the most requested and technically complex topics in the drone industry. Although it is an extensive and intricate issue that requires much more than a blog post, let’s explore drone swarms’ most essential and interesting aspects, from their diverse applications to their technological underpinnings and future potential.

Drone swarms are groups of autonomous, networked UAVs that work collaboratively to achieve common objectives. Unlike traditional drone operations, where each drone is piloted individually, swarms operate with a high degree of coordination and communication among the drones themselves.

Although many might mentally associate the word “swarm” with bees or other insects (especially considering its meaning), the idea comes from birds. Some bird species coordinate their flights in seemingly rhythmic and harmonious motion without ever crashing into each other. In many cases, they even pick up a lot of speed, even as the number of birds grows exponentially. All of this can be witnessed in this fantastic video of bird swarm behavior in action, and here is an example of a simple algorithm to emulate swarm behavior. Naturally, applying these algorithms to groups of drones has gained much traction. But when talking about drone swarms, it is important to distinguish between “multiple drone operation” and true “swarms”.

Multiple drone operation is a scenario where drones fly on predefined routes (A-B, A-C, B-C) to carry out tasks (e.g. medical delivery). This operation is managed or observed by an operator, who can manage multiple drones simultaneously (e.g. Matternet’s exemption to test a 1:20 pilot-to-drone ratio). Even though there are a lot of drones in the air at the same time in a swarm-like setting, the drones are not operating together for a common goal, so it is wrong to call this a drone swarm. It is also imperative to stress that a waiver/exemption is necessary for this type of operation, since it is strictly forbidden by many country’s laws (e.g. FAA Part 107.43).

In a true drone swarm scenario, drones fly simultaneously and collaboratively to achieve common objectives. These swarms can be controlled in multiple ways:

1. The centralized control scheme for drone swarms (e.g. leader-follower formation, virtual structure) involves a single control point (e.g., a ground control station) that processes all relevant information, makes decisions, and issues commands to each individual drone. In other words, they simply fly to a certain point in space without the drone [necessarily] being aware that there is another drone nearby.
2. The decentralized control scheme (e.g. behavior-based formation, artificial potential field (APF)) allows drone swarms to manage themselves through distributed decision-making processes. Each drone in the swarm acts based on information and predefined rules, coordinating with its immediate neighbors by communication-based, sensor-based, or vision-based methods to maintain formation – but without explicit communication.
3. In a distributed control scheme (e.g. consensus-based formation), control actions and decision-making processes are spread across all drones in the swarm. Each drone possesses a degree of autonomy but also collaborates closely with others, sharing information and making decisions that benefit the collective objective – this means working together to make decisions that reflect the collective interest.

Most notably, drone swarms have been used in the entertainment industry to create stunning light shows. These shows have captivated audiences worldwide with their synchronized aerial displays. However, many familiar with these operations would note that drone light shows have a central computer on the ground tracking every drone and moving them all in unison. The drones’ movements are dictated in a complicated and detailed version of air traffic control, which is calculated so that the flight paths stay separate and no collisions occur. This means that individual drones take no part in the decision-making process.

In agriculture, drone swarms have the potential to be used for precision farming, monitoring crop health, applying fertilizers and pesticides, and even planting seeds with remarkable efficiency. A notable example is the recent approval by the Federal Aviation Administration (FAA) for Hylio to deploy up to three drones weighing more than 55 pounds simultaneously with a single pilot. Although three drones are not what most people would consider a drone swarm (and there is still a pilot involved in this case), this approval allows for nighttime operations and significantly increases productivity, enabling one operator to treat 150 acres per hour compared to the previous 50 acres per hour. More importantly, this milestone paves the way for broader implementation of drone swarm technology in large-scale farming operations. The CEO of the National Agricultural Aviation Association (NAAA) estimated that only 3% of the 1,500 businesses it represents use drones. This share is increasing year over year and will increase exponentially with the use of drone swarms.

Environmental monitoring can also benefit from the coordinated capabilities of drone swarms. These swarms can quickly cover large areas and provide detailed environmental data. For instance, drone swarms could assess the impact of natural disasters and monitor the health of entire forests. The benefits of using a drone swarm rather than individual drones boil down to a matter of area and scale: a swarm can split the tasks by area (covering different areas at the same time) or function (with each drone scanning different aspects) and cover the entire area exponentially faster than individual drones (especially if these are dependent on a pilot).

And of course there are plenty of other examples. From a drone swarm building a 6-meter tower to drones building a rope bridge, there are plenty of creative attempts to use multiple drones to help achieve a goal. Naturally there are also destructive ways in which drones are used. Drone swarms have been used militarily to overwhelm enemy defenses with sheer numbers (saturation attacks), or to conduct precise, targeted strikes with minimal risk to human personnel. The U.S. Air Force has been at the forefront of developing swarm technology, successfully demonstrating the deployment of 103 Perdix drones from F/A-18 Super Hornet fighters already back in 2016. However, the development and deployment of military drone swarms also raise ethical and cybersecurity concerns, as the potential for autonomous lethal force and vulnerability to hacking are significant issues that need to be addressed.

Control mechanisms of drone swarms are a focal point of how [much] the technology evolves. Controlling drone swarms involves sophisticated strategies to manage and coordinate multiple drones flying in formation. Conventional methods include leader-follower, virtual structure, behavior-based, consensus-based, and artificial potential field techniques. These methods provide structured frameworks for formation control and are inspired by natural behaviors and mathematical models. They focus on ensuring that each drone maintains its position relative to others, adapting to changes in speed, direction, and environmental conditions. They offer reliability and simplicity but may face dynamic adaptability and scalability challenges.

In recent years, a different type of “AI-based methods” have introduced new dimensions to swarm control, leveraging artificial neural networks (ANNs) and deep reinforcement learning (DRL) to enhance autonomous decision-making and coordination. These techniques allow drones to learn and optimize their behavior in complex environments, significantly improving adaptability and efficiency. ANNs provide powerful tools for function approximation and generalization, while DRL enables UAVs to learn optimal control policies through interaction with their environment. However, these methods also bring challenges such as computational complexity, the need for extensive training data, and safety concerns during the learning process.

Integrating both of the above-mentioned conventional and machine learning methods is often seen as the most promising path to unlock the full potential of drone swarms in various applications. Regardless of the method used, the goal is ultimately the same: to ensure that the swarm operates [autonomously] as a cohesive unit. This entails several technical issues such as: maintaining formation, avoiding collisions, and executing complex maneuvers, all of which rely on extremely fast and accurate information, which is also why topics like 5G connectivity will play a key role in the evolution of drone swarms. All of these (in addition to having certified algorithms) will be the core factors for the development and increased implementation of drone swarms into commercial operations.