This document describes a study on developing a self-driving car prototype using deep reinforcement learning. The researchers used a deep Q-network (DQN) algorithm to train an agent to control a simulated car directly from sensor inputs like cameras. The DQN was able to successfully navigate the simulated environment and control the car without any knowledge of its dynamics. While the discrete state-space DQN achieved stable control, the researchers believe the work could be extended to use continuous action spaces to allow for varying speeds and improved reward functions. The document also provides background on deep learning, autonomous vehicles, and reviews related work applying reinforcement learning methods to autonomous driving tasks.