This dissertation explores SteeringNET, which is a deep learning-based obstacle avoidance approach to autonomous vehicles, with a focus on visualizing and understanding how neural networks learn and evolve without human intervention. By building an Evolutionary Artificial Neural Network (EANN) from scratch in a 2D Unity simulation, this project demonstrates how self-driving agents can develop effective driving behaviours through iterative trial and error. The system bypasses traditional, manually programmed logic, instead evolving solutions using genetic principles to improve over successive generations. Alongside technical development, the work researches different methods of deep learning as well as critically evaluates the legal, ethical, and societal implications of deploying autonomous systems in the real world. The goal is to contribute both a practical simulation and a conceptual framework that make deep learning in driving autonomy more transparent, adaptable, and educational.