Masahiro Ono, Mitsuru Ishizuka
In this paper, we deal with the influence of the spatial structure, which limits the communication of agents, on the characteristics of individual agents. We investigate the population characteristics and the behavior of the agents playing iterated prisoner’s dilemma (IPD) games on the spatial structure. We first propose a new agent model that plays the IPD game, which contains the gene of the coded parameters of reinforcement learning. The agents evolve and learn while playing the games. Second, we report an empirical study. In our simulation, we observe that the spatial structure affects learning and evolution. Learning is generally not conducive to the mutual cooperation between agents, except in some special conditions. Then, we try to control the population characteristics. We find that they are controllable to some extent when we fix the strategies of several agents.
Subjects: 7.1 Multi-Agent Systems; 12.1 Reinforcement Learning