Fly United

Drones and other robots have proven themselves to be quite valuable for tasks ranging from aerial photography to infrastructure inspections, wildlife conservation, and disaster relief. These many successes have led engineers to ponder the question: if one drone can accomplish this much, then what might a large swarm of drones be capable of? Many advances have been made in this area, but due to a multitude of complexities, operating swarms of drones is still more in the realm of academic research and one-off demonstrations than it is a reliable means of accomplishing real-world goals.

The difficulties largely lie in the development of effective control systems that guide the actions of each robot to accomplish collective goals. These systems fall into one of two broad categories — centralized algorithms that provide guidance to all members of the swarm, and distributed algorithms that allow each member to learn and carry out its own role. Each method comes with issues of its own. Centralized algorithms are challenging to scale, and they introduce a single point of failure. Distributed systems, on the other hand, typically require long trial-and-error-based design processes to produce the desired result at the swarm-level, and even still they tend to be fragile.

A team at the Université Libre de Bruxelles has approached this problem from a different angle in search of a solution. The result is a swarm architecture inspired by the human nervous system that could greatly simplify coordination between robots. Their method, called the self-organizing nervous system (SoNS), blends the best components of both centralized and distributed control systems to achieve this goal.

The SoNS robotic swarm architecture enables robots to autonomously form, adapt, and manage dynamic multilevel hierarchies. Using self-organized structures, robots connect in temporary parent-child relationships, creating a reconfigurable system where one robot, the "brain," coordinates actions. This approach maintains scalability, flexibility, and fault tolerance — key features that must be present in successful robot swarms.

In SoNS, robots form connections locally, with each robot communicating only with nearby robots. This ensures scalability even in large swarms and fault tolerance by allowing any robot, including the brain, to be replaced or reallocated in case of failure. Robots can merge into larger systems by recruiting others or split into smaller groups if connections are lost. These behaviors are supported by adaptive algorithms and target graphs that guide role allocation and coordination.

Robots share sensor data and actuation instructions throughout the hierarchy. Data flows upstream to the brain for collective decision-making, while actuation commands are sent downstream. Despite the hierarchy, robots retain local autonomy, enabling agile reactions and system-wide responses when needed. Users can program the swarm as a single entity, simplifying control and enabling coordinated multi-robot behaviors.

SoNS has been tested in both physical and simulated environments, scaling to swarms of up to 250 robots. In these experiments, it was demonstrated that by blending centralized coordination with decentralized resilience and flexibility, SoNS can effectively coordinate the actions of many robots to carry out complex and useful tasks. The team is presently refining their system with the hope of deploying it to larger swarms of physical robots in the near future.