Generally, life including human beings can alter its action rule in order to adapt to changes in its environment. As they repeat the process of trial and error, they establish action rule based on obtained information. However, in original sugarscape model, the agent movement rules are designed by human beings in advance. In this paper, we create a sugarscape model in which the agent acquires movement rule through its learning. From some computer simulations, we show the agent acquires the similar rule to the conventional Sugarscape model. we show that the average of residual entropy decrease, as episode number increases. It is shown that under the situation that environment changes, the model in which agents act according to the movement rule obtained by Q-learning collects more sugar than sugarscape model.