Identifying Inter-agent Interaction Graph Networks from Time Series Records of Flying Insect Groups

As aerospace systems grow to incorporate multi-agent systems with both dynamics and nontrivial neighbor interactions, identifying network topologies from observed data is a growing challenge. This paper develops a concise method to identify the adjacency matrix of a graph from limited-length time histories of simultaneous individuals in groups. The adjacency matrix identification problem is written as a constrained linear least-squared error minimization problem amenable to implementation in ubiquitous convex optimization routines, such as linear matrix inequality solvers. The constraints come from the graph adjacency matrix definition and provide theoretical bounds on recorded trajectory length. The approach is applied to experimental honey bee flight data gathered by vision-based digitization while tracking a stimulus injecting a sum of sinusoids and triangle wave motion. The results on insect flight data indicate that honey bees primarily interact with a small number of neighbors (2-3) while tracking stimuli, even in the presence of a larger number of nearby individuals. These tools can enable us to identify the underlying mechanisms of unknown swarming systems, translating those into engineered aerial systems.