Reinforcement learning methods for finding equilibria and tracking evolution paths in conflicts

The search for the equilibrium states of a given conflict is a major issue in conflict analyses. There are several traditional methodologies to determine the equilibrium states of a strategic conflict, such as logical definitions within the graph model for conflict resolution and the matrix representation for conflict resolution. However, these methods depend on a graphical or mathematical representation and need analytical expressions to calculate the equilibrium states. Reinforcement learning (RL) is a type of machine learning method that can search for the equilibrium states by trial-and-error without the need for a precise mathematical model of the conflict problem. This paper proposes a novel multiple RL technique that deals with conflict resolution problems for the case of two decision makers. Moreover, the proposed method cannot only find equilibria, but also track all paths from any status quo to the equilibria in conflicts when these paths exist. This method is evaluated using two well-known conflict analysis examples. The experimental results show that the proposed method can quickly, correctly, and efficiently find the equilibria and track the evolution paths.