06 October 2025
Swarm robotics draws inspiration from collective behaviours in nature, such as flocks of birds or swarms of bees, to enable groups of robots to coordinate and complete tasks without centralised control. In doing so, these distributed systems are more flexible and robust than their centralised counterparts. However, effective coordination in these multi-robot systems depends on the ability to share useful information. Excessive information sharing can lead to bandwidth bottlenecks, while insufficient information-sharing limits performance.
In this study, EMERGE partners from the University of Bristol use a learning-based approach to optimise information sharing in a hybrid robot swarm, where each robot maintains local autonomy, but information is shared via a central repository. To do so, the authors introduce the Hive, a generalisable architecture for building collective knowledge which supports selective sharing of any robot’s internal states, observations, and inferences. They use a genetic algorithm to evolve weights that determine only the minimal information to share for a given task. They validate this approach in a simulated point-to-point logistics task, demonstrating that the Hive significantly reduces communication bandwidth with no loss in performance. Their results show that local decision-making, combined with optimised Hive-based sharing of information, has the potential to lead to adaptable and scalable swarm deployments.
Read the paper in the link below.
More Information
