Decentralized coordination control for mas with workload learning

In this work we deal with a decentralized coordination control (DCC) problem for Multi-Agent Systems (MAS). The system output, which is the control goal of MAS, is defined as the summation of the output of all agents. The objective of this study is to derive an effective control law for each agent such that the system output can asymptotically track a reference trajectory. Individual agent can only access its own state information, the system output, and the reference. Hence each controller has to be designed in a decentralized manner without using the information of other agents, even the total number of agents is unknown to individual agent. When the actual total workload is exactly unity, the proposed DCC law ensures the asymptotic error convergence. Otherwise, algebraic or adaptive identification algorithms are applied to estimate the workload mismatching. Two learning algorithms are proposed to rescale the workload assignment for each agent after estimating the workload mismatching. One iterative learning algorithm achieves asymptotic learning convergence, while the other achieves a deadbeat learning convergence.