Survival of Agents in Multi-Agent Systems Based on Sugarscape Model Using Reinforcement Learning

Artificial society is the teamwork behavior of humans who come into existence in the effect of interaction with the environment under special rules. Through interaction with each other and their society during the time, humans learn how to determine and amend their methods and behavioral rules and they do this by trial and error. The sugarscape model is a model of artificial society. Because learning has not been used in the standard sugarscape model, in this work, we have tried to use reinforcement learning in the sugarscape model, where learning is used in the selection of actions that the agents do to collect more sugar. For this purpose, a learning sugarscape model based on reinforcement learning was implemented and compared with the sugarscape model without learning. The results showed considerable increase in the number of survived agents at the end of time period in the reinforcement learning sugarscape model. This explains that in a reinforcement learning sugarscape model the agents can detect useful and harmful actions over time and try to choose useful actions to guarantee their future lives.