Adaptive architectures for distributed control of modular systems

We consider the distributed control of large-scale modular systems, i.e., systems consist of physically interconnected (and possibly heterogeneous) submodules that use local information to achieve a given set of global objectives. A graph-theoretic approach is used to model the unknown physical interactions between submodules and utilize an adaptive control algorithm to learn and cancel the effect of such interactions as well as the submodule-level modeling uncertainties asymptotically by allowing submodules to locally communicate with each other through the graph. The key feature of our framework is to bound the mismatch error between the actual and desired closed-loop modular system performance by a constant that is a priori known and does neither depend on the underlying graph topology nor uncertainties.