Online Optimal Synchronization of Nonlinear Multi-agent Systems under Unknown Dynamics

In this paper an online optimal distributed algorithm is introduced for multi-agent systems synchronization under unknown dynamics based on approximate dynamic programming and neural networks. Every agent has employed an actor-critic structure to learn its distributed optimal policy and the unknown dynamics of every agent is identified by employing a neural network approximator. The unknown dynamics are identified based on the experience replay technique where the recorded data and current data are used to adopt the approximators weights. The introduced algorithm learns the solution of coupled Hamilton-Jacobi equations under unknown dynamics in an online fashion. While the weights of the identifiers and actor-critic approximators are being tuned, the boundedness of the closed loop system signals are assured using Lyapunov theory. The effectiveness of the proposed algorithm is shown through the simulation results.