Note: Mirali, Furugh and Werner, Herbert

Abstract: During this project, several prediction-based consensus protocols for systems of singleintegrator agents with uniform communication delays are discussed. Common to these protocols is that the agents predict future states of their neighbours in order to improve the convergence properties of the system. Protocols of three different categories are introduced: linear extrapolation methods, prediction by artificial neural networks and observer based state predictors. The first two categories are performing the prediction solely based on the agents' past states and the third introduces a secondary dynamic system. In the case of observer based prediction, conditions for guaranteed consensus are derived for a centralized protocol, and a distributed protocol without such guarantees is proposed. An empirical comparison of the protocols in terms of convergence speed, robustness against time delays and deviation from average consensus is performed by simulations.