Designing a Cooperative Evolution Mechanism Based on Deep q-Network

To explore potential mechanisms that facilitate the emergence of cooperation in social dilemmas, this paper proposes a deep q-network-based cooperative evolution mechanism (DCM). The mechanism assumes that agents are linked by a square lattice network, and each agent plays a prisoner's dilemma game (PDG) with their von Neumann nearest neighbor. Based on the DCM, the paper investigates the behavioral characteristics exhibited by agents during the evolution of PDG on a square lattice network. The simulation results demonstrate that DCM effectively increases the number of cooperators in the population, and even exhibits the behavioral characteristics of “full domain cooperation”. Moreover, the experimental results indicate that DCM, even without an experience replay pool, can still promote the emergence of cooperation in PDG, suggesting that DCM has good robustness.