On evolution of agent behavior under limited gaming time with reinforcement learning

Based on the prisoner’s dilemma (PD) evolutionary game model and reinforcement learning framework, this paper studies the impact of factors such as temptation payoff, time allocation, and others on agent behavior evolution and strategy selection under limited gaming time resources, across three different agent relationship structures. The results show that an increase in the agent’s gaming time resources and lower temptation payoffs, or the agent’s greater emphasis on long-term rewards and avoidance of excessive behavioral adjustments, all contribute to promoting cooperation between agents. Additionally, the total remaining gaming time between agents gradually increases as the game progresses, while the total gaming time between agents gradually decreases. Both will eventually reach a steady state after a sufficiently large number of game rounds. Further results indicate that an increase in temptation payoff leads to an increase in total remaining gaming time, while reducing the total gaming time between agents. Finally, the measure of heterogeneity in gaming time distribution between agents gradually increases throughout the game process. This is particularly evident when the temptation payoff is high, as the differences in gaming time allocation between agents increase, significantly enhancing the heterogeneity of gaming time among agents in the system. This study provides important theoretical support for understanding agent behavior evolution under limited gaming time resources, especially in dynamic cooperative and competitive game scenarios.