On the Effect of the Layers' Number of Deep Neural Network for Improving the Reward of a Reinforcement Learning Robot