Coordination rule design for constrained optimization using mobile sensor/actuator nodes

In this paper we present ideas toward solving constrained optimization problems in a spatially-distributed mobile sensor/actuator network using decentralized computation. Notionally, each node of the network is considered to be a distributed computational unit that evolves its state according to a pre-defined rule. First, we show how to design coordination rules to ensure that the global state of the network evolves to the solution of a prescribed constrained optimization problem. Our strategy uses a recurrent neural network structure to solve the optimization problem in a global way. Next, we introduce ideas for the case when there is an absence of an allto- all communication topology. Assuming each node can only communicate locally with its 'nearest neighbors,' our approach is to use the notion of a consensus variable protocol that implements a distributed observer, enabling local nodes to asymptotically obtain global information using only nearest neighbor communications. Finally, we suggest superimposing the neural network structure on top of this distributed observer to solve the global optimization problem using only local and nearest neighbor communications.