Autonomous vehicles (AVs) need efficient path planning systems to get around in complicated settings safely and successfully. It can be hard for traditional methods to change to changing situations and find the best solutions for a wide range of goals. As a reaction, this study describes a new method that combines reinforcement learning (RL) with convolutional neural networks (CNNs) to help autonomous vehicles find the best routes. Using RL to learn how to make decisions by interacting with the world and CNNs to get useful spatial features from sensor data is what our suggested system is all about. By training these parts together, the model learns to move through different settings while maximizing different goals, like reducing trip time, energy use, or the chance of a crash. A lot of tests are done on virtual and real-world datasets to see how well our approach works. The results show that it works better than standard methods and other deep learning techniques. In addition, we do in-depth ablation studies to look at how different parts and hyperparameters affect the total performance. Our research shows that combining RL and CNNs could greatly improve the ability of self-driving cars to plan their routes, which would allow for better and more efficient travel in real life.