The work presented here summarizes research into using the HighwayEnv simulation environment to apply deep reinforcement learning to autonomous driving. We train a Dueling Deep Q-Network model to navigate a custom road environment containing several real-world driving scenarios, such as highways, roundabouts, and crossroads. By combining these many features, we ensure that the model learns to adapt to changing traffic situations and difficult decision-making tasks. During training, a well-structured reward system is used throughout to promote safe and efficient driving. The trained agent demonstrates strong generalization, achieving zero or near-zero collision rates in complex scenarios that include merging, roundabouts, and U-turns. The results support the usefulness of specialized simulation environments in building resilient policies and the viability of using Dueling DQN for complex tasks associated with autonomous driving. This paper contributes to the growing literature on reinforcement learning applied to self-driving problems by providing explicit details on training methods and configuration of the environment.