Leader-Follower Optimal Consensus of Discrete-Time Linear Multi-agent Systems based on Q-Learning

This paper proposes a Q-learning solution to achieve optimal consensus of discrete-time linear leader-follower multi-agent systems. Model-free linear dynamic systems are considered herein and it is one of the topics that the control system community pays high attention to. A value function based on state error is designed for directed graph, and the weights of it can be estimated through a data-driven method. Then the consensus policies will be calculated iteratively by minimizing energy consumption. Besides, the proof of stability is given to ensure the theoretical convergence. Finally, a simple simulation is shown to verify the validity of the algorithm.