Dynamic Model-Free Reinforcement Learning Strategies for Achieving Nash Equilibrium in Graphical Games with Communication Challenges

Solving the consensus problem in multi-agent systems is a significant challenge, particularly when dealing with communication issues among system agents. One other important challenge in solving the consensus problem is the optimality of each agent's inputs, which leads to the Nash equilibrium problem. Most of the suggested methods for addressing both the Nash issue and the consensus problem depend on the agents' dynamics. Also, these algorithms need steady communication conditions, which are challenging to achieve in the real world. This research presents a model-free policy iteration reinforcement learning algorithm that does not require agent dynamics and uses the gradient descent technique with critic network structures. The algorithm also has the ability to solve equations when an agent loses its connection and make it independent from the system model and capable of working even when communication is disconnected.