A Neuro-Evolutionary Framework for Autonomous Vehicle Control Using Genetic Algorithms and Neural Networks

Abstract

Autonomous vehicle development demands reliable control strategies that can operate under dynamic and safety-critical conditions. Simulation-based evaluation plays a vital role by enabling controlled, repeatable testing without the risks of real-world deployment. This paper presents a custom-built self-driving car simulation framework that integrates feedforward neural networks with genetic algorithms for autonomous control optimization. Implemented entirely in JavaScript without external dependencies, the framework evolves steering and acceleration policies through neuro-evolution using sensor-based perception. A lightweight sensor configuration and multi-objective fitness evaluation enable smooth navigation and effective collision avoidance across complex track layouts. Experimental results demonstrate a navigation success rate of 91.7 percent and a collision avoidance accuracy of 96.8 percent after 47 generations, while achieving faster convergence than gradient-based training methods. The proposed framework offers a flexible and computationally efficient platform for autonomous vehicle research and rapid prototyping.