Neighbor-aware reinforcement learning fosters cooperation in spatial public goods games

The emergence and self-organization of cooperation within social dilemmas remain a central challenge in the study of evolutionary public goods games. Traditional models often rely on fixed behavioral update rules, inadequately capturing the learning capabilities of real-world individuals. To address this, we develop and investigate a novel model of cooperative evolution that integrates individual reinforcement learning (RL) with a mechanism termed “Neighbor Influence” (NI). In this framework, agents are not passive rule-followers; they employ Q-learning to dynamically decide between cooperation and defection. Importantly, their decisions are shaped by both individual trial-and-error and the influence of their neighbors. This study systematically analyzes the role of the NI mechanism and explores how reputation systems and agents’ perceptual ranges (direct versus indirect neighbors) modulate cooperative evolution. Extensive simulations reveal that NI is crucial for promoting the spontaneous formation of cooperative clusters, thereby mitigating the “Tragedy of the Commons”. This mechanism substantially boosts cooperation levels, even under conditions where conventional RL models falter. Our research further elucidates how the strength of neighbor influence, individual preferences for reputation versus direct payoffs, and perceptual range critically shape the dynamics and eventual stability of cooperative behavior. By focusing on adaptive learning and local social interactions, this work offers a computational model that clarifies the mechanisms underpinning macroscopic cooperation in complex systems and provides valuable theoretical insights for designing strategies to foster collective collaboration.