Scientists developed swarm intelligence based on the principles of the human nervous system's architecture.

Over the past two decades, researchers from the Free University of Brussels (Belgium) have noted that studies in the field of swarm intelligence have revealed the potential for managing large numbers of autonomous robots without a central coordinating robot. These emergent systems are designed for collective behavior among various objects, each performing simple functions while interacting with others. Engineers utilize such technologies to tackle a wide range of tasks: environmental monitoring, navigation and transportation, and construction.

Traditionally, the architecture of most swarms of robots represents a self-organized hierarchy, as artificial swarm intelligence was originally developed based on biological systems of social insects. These include, for example, ants, wasps, and bees, which form seasonal or perennial colonies. However, to date, few specialists in artificial intelligence have succeeded in establishing an effective hierarchy within the structure of robot agents.

Belgian scientists have created self-organizing nervous systems (SoNSs), which are analogous to human systems. In this swarm structure, robots form multi-level systemic architectures and occupy specific positions within the leadership hierarchy, the highest of which is the "brain." This brain directs and controls group actions during missions, acting as a temporary coordinator.

Any agent at any hierarchical level can be interchangeable with another, even with the "brain." In this setup, robots "communicate" only with their nearest neighbors, allowing them to maintain the ability to change configurations and reconfigure dynamic systemic architectures.

Regardless of the hierarchical position of a neighbor, a robot attempts to "recruit" it. If an agent is currently in the same nervous system as the competitor or if its quality is inferior to its own, it will decline. Under other conditions, the neighbor agrees and becomes the "child" robot of the "parent," merging with it and attracting its subordinate "descendants."

At the same time, the "parent" can replace and potentially demote an already assigned child element with one that is more suitable at that moment. The disconnected "descendant" automatically resumes operation as the "brain" of its own system and updates its target graph accordingly. By performing these operations, robots in SoNS can continuously redistribute themselves.

The uniqueness of this development lies in the fact that the system combines the advantages of centralized system manageability while retaining the scalability, flexibility, and fault tolerance of a self-organizing network, meaning it can maintain its functionality even after the failure of one or several of its components.

The authors of the study explained that SoNS allows for programming the swarm as if it were a single robot, rather than an entire "population," significantly simplifying the integration of these systems into practical applications. Typically, challenges arise during the transition of technology from laboratory conditions to real-world scenarios.

Engineers from Belgium conducted four missions, each involving at least five tests with real air-ground robots (up to 12 agents) using a specially designed quadcopter platform, as well as 50 simulations (up to 65 robots).

In all experiments, the robots successfully completed their missions. They explored the environment, established routes, moved, and planned subsequent steps to accomplish tasks, such as searching for and rescuing lost "comrades."

Furthermore, the scientists demonstrated the scalability of the SoNS approach in swarms of up to 250 robots within a physical simulator and several types of system fault tolerance in both simulations and reality. This contribution to the field of swarm intelligence in robotics will aid in the creation of new solutions that enable robots to perform tasks more quickly and efficiently. In particular, such an approach could be applicable in search and rescue operations during natural disasters or emergencies.

The scientific work is published in the journal Science Robotics.