In a bid to enhance swarm autonomy and decision-making in collective perception tasks, researchers at the Université Libre de Bruxelles have developed an innovative self-organizing approach. Published in the journal Intelligent Computing, their paper introduces a method that utilizes a hierarchical system, allowing one robot at a time to act as the “brain” and consolidate information on behalf of the group. This approach aims to increase collective perception accuracy by minimizing sources of uncertainty.
By combining elements of centralized and decentralized control, the researchers were able to leverage the advantages of both approaches. The self-organizing hierarchy preserves the scalability and fault tolerance of decentralized systems while incorporating the accuracy of centralized ones. Notably, this method provides a means for applying centralized techniques for information fusion from multiple sensors to a self-organized system, which was previously only demonstrated in fully centralized systems.
To evaluate their approach, the researchers compared it against three benchmark approaches and found that it outperformed the others in terms of accuracy, consistency, and reaction time. The testing involved a simulated swarm of drones and ground robots collecting spatial data by detecting objects in an arena and forming a collective opinion of object density. The robots relied on short-range sensors to deduce the number of objects per unit.
The self-organizing hierarchy approach is based on a dynamic ad-hoc hierarchical network, employing a mergeable nervous system framework. In this framework, robots at different levels of the hierarchy have distinct roles in the decision-making process, and their connections and relative positions can adapt as necessary. Each robot communicates solely with its direct neighbors.
The top-level robot, acting as the “brain,” performs inferences and sends motion instructions downstream. Middle-level robots manage data transfer and contribute to the balancing of global and local motion goals, while the majority of bottom-level robots engage in sample collection while managing local motion.
Moving forward, future research in this area could explore advanced inference methods and investigate the robustness of sampling methods under more challenging environmental conditions or different types of robot failures, such as environments with larger obstacles or irregular boundaries.
– Aryo Jamshidpey et al, Reducing Uncertainty in Collective Perception Using Self-Organizing Hierarchy, Intelligent Computing (2023). DOI: 10.34133/icomputing.0044