Evolutionary Autonomous Car Navigation using NEAT Algorithm
Pratham Shrivastav, Preet Darji, Dev Patel, Maanav Shah, T. Shalini · 2025
The paper presents an autonomous driving automobile, which has been designed with the help of applying NeuroEvolution of Augmenting Topologies (NEAT) algorithm. The NEAT was chosen due to the possibility of the adjustment of the structure and the weights of the neural networks, making it possible to allow the realization of the adaptive control strategies that are present on the complex driving scenario. A custom-lined simulation was conducted to train the system whereby the system is trained in order to make the vehicles installed with the distance sensors to navigate in the road racks installed with the traffic lights, the stopping zones, and the moving obstacles etc. The sophisticated controllers are informed on what not to do to avoid colliding, on how not to go beyond the traffic rule and on how to cover a cost-effective path through out the many generations. The results show that the high numbers were achieved on completion of the paths as also the efficiency of one pathway, the more responsible utilization of the sharp curves, the high degree of the reliability in the fragmented path designs, and the high reduction in the number of the collisions - in favour of the safety of the entire system. The respective findings imply that NEAT can be powerful and versatile to train autonomous navigation agents without the requirement of strict neural architectures.