Collaborative system identification via parameter consensus

Classical schemes in system identification and adaptive control often rely on persistence of excitation to guarantee parameter convergence, which may be difficult to achieve with a single agent and a single input. Inspired by consensus systems, we extend classical parameter adaptation to the multi agent setting by combining an adaptive gradient law with consensus dynamics. The gradient law represents the main learning signal, while consensus dynamics attract each agent's parameter estimates toward those of its neighbors. We show that the resulting decentralized online parameter estimator can be used to identify the true parameters of all agents, even if no single agent employs a persistently exciting input.