In this paper, the problem of consensus tracking of uncertain multi-agent systems (MAS) with communication faults is addressed. The communication is assumed to be undirected. A reinforced unscented Kalman filter (RUKF) is employed to adapt the noise covariance matrices and to estimate the uncertain states of MAS as well as to train neural network internal parameters by providing a set of previous measurements. A Chebyshev neural network (CNN) is incorporated to learn the uncertain plant. To prevent the neural network approximation errors a hyperbolic tangent function based robust control term is applied. The Lyapunov approach guarantees the stability of RUKF which is running in conjunction with a robust control method. Numerical simulations are presented under different fault conditions to show the effectiveness of the proposed RUKF with 5% less computation power compared to adaptive UKF.
INDEX TERMSMulti-agent Systems, Uncertain Dynamics, Reinforcement learning and Non-Linear Filtering, Chebyshev Neural Networks, Control, Communication Faults. estimate the states and multiple faults is derived based on the time delay closed loop system equation. In [4], a distributed consensus protocol is proposed and designed in terms of linear matrix inequalities to ensure robust consensus over uncertain communications.The communication uncertainties in [7] have been modelled as stochastic uncertainties based on which protocols are validated in the mean square sense. A robust consensus problem over uncertain network graphs of linear feedback protocols has been examined by solving a H∞ control problem for a set of 2D subsystems in [8]. Again, it is well known that introducing the communication delay will lead to a reduction of the performance of instability [9] and thus investigating communication time delays in MAS is another interesting work. In [9], a sliding mode control design is proposed for a leader follower MAS to deal with the uncertainties incurred by unknown time varying communication delay and disturbances. In [10] consensus control of MAS with input and communication delay in the frequency domain is discussed. The considered linear dynamic model of each agent in [10] involves multiple input delays. In [11], a distributed coordinated attitude tracking control problem for spacecraft formation with time-varying communication delays under the condition that the dynamic leader spacecraft is a neighbor of only a subset of follower spacecraft is examined. However, the observation made from literature, there are a limited work reported regarding the design of a consensus for nonlinear uncertain MAS in presence of