Autonomous driving holds immense importance, offering numerous benefits such as increased road safety, reduced traffic congestion, and enhanced accessibility for individuals unable to drive. Its future projection encompasses widespread adoption, revolutionizing transportation and urban planning. Integrating deep reinforcement learning into autonomous vehicle control modules presents significant advantages. Reinforcement learning methods, enable adaptive decision-making in dynamic environments, enhancing system robustness and performance. Our proposed model integrates a route planning algorithm with vehicle state information and visual data from a camera, processed through a convolutional neural network. This fusion enables efficient decision-making for acceleration, steering, and braking commands, crucial for safe and efficient autonomous driving. In this first approach, the aim is to obtain a reward function that allows training the model in an optimal way without the need for training with an extremely enormous number of episodes.