Reinforcement Learning Control for Consensus of the Leader-Follower Multi-Agent Systems

This paper considers the optimal consensus of multi-agent systems using reinforcement learning control. The system is nonlinear and the number of agents can be large. The control objective is to design the controllers for each agent such that all the agents will be consensus to the leader agent. We use the Actor-Critic Network and the Deterministic Policy Gradient method to realize the controller. The policy iteration algorithm is discussed and many simulations are provided to validate the result.