Reinforcement learning in spatial public goods games with environmental feedbacks

Abstract

The feedback between strategy and environment is ubiquitous in nature and human society, which has been receiving increasing attention from researchers. Meanwhile, Q-learning allows players to explore the optimal strategy by interacting with the environment. In this paper, we introduce the Q-learning into the spatial public goods game with environmental feedbacks. The simulation results show that the environmental feedback can promote cooperation. The increase of synergy coefficient $r$ and strength of the environmental feedback $\alpha$ is beneficial for the evolution of cooperation. The effects of discount factor $\gamma$ on the cooperation level of the population are non-monotonic. When $r$ or $\alpha$ is low, the high values of $\gamma$ can promote the emergence of cooperation. However, with the increase of $r$ and $\alpha$, the low values of $\gamma$ are more favorable to the evolution of cooperation.