Distributed parameterized model predictive control of networked multi-agent systems

When considering a distributed control framework for a multi-agent system, care has to be taken to respect the limited information available to each agent as well as the amount of possible communication in the network. This presents difficulties in performing distributed model predictive control as an agent must typically be able to simulate its neighbors dynamics into the future. We present a framework based on parameterized feedback control laws which will allow a multi-agent system to perform distributed model predictive control. It enables both the simulation of neighbors' states as well as the ability to minimize a collective cost in a distributed fashion. Moreover, given cost and dynamic dependencies between agents, we characterize the information that is needed by each agent to evaluate whether the optimization is feasible in a given network.