Global-to-Local Design for Self-Organized Task Allocation in Swarms

We propose a global-to-local design methodology to compose heterogeneous swarms for self-organized task allocation. We consider embodied agents with only local knowledge and local communication driven by a simple reactive-control architecture (i.e., local level). We define finitely many agent controllers and focus on the macroscopic dynamics (i.e., global level) of behaviorally heterogeneous swarms. The user inputs the desired swarm allocations as the stationary distribution of agents over two tasks. Given the input, we mathematically derive the composition of a heterogeneous swarm that approximates the requirements of the user. We investigate our methodology over several case studies and validate our results with multi-agent simulations.