System Identification of Interaction Network in Insect Group Flights by Cross Correlation

Despite significant communication and processing limitations, flying insects frequently demonstrate coordinated flight in crowded assemblies. Identifying the feedback mechanisms that enable these maneuvers remains a challenging task. This study experimentally records multiple flying insects induced to chase a moving vertical visual stimulus, and a cross-correlation based system identification method is developed to identify their interaction networks. Modeling the interaction network as a series of pairwise-directed graphs, the adjacency matrix weights are identified serially by indexing agentwise through temporal cross correlation. The induced matrix 2 norm, in and out-degree weight sum, and degree distributions are then computed for 1000 examples of recorded interaction networks to reveal aggregate trends. The results show a consistency in topological neighbor choice, with 68\% of agents limiting their connectivity to 2-4 agents. Information sharing among insects (as measured by adjacency matrix norm) grows in direct proportion to group size. The degree distribution obeys a power law distribution, consistent with scale free networks. Numerous examples of insect pairs sharing similar velocities were also identified. The interaction network identification implementation and findings can help in the understanding swarm models for insects in cluttered group flight as well as identifying neighbor-relative feedback control strategies for engineered air vehicles.