Abstract:
The use of autonomous driving is growing in popularity thanks to manufacturers like
Tesla, Waymo, and numerous others. Regardless of autonomous driving, the
application of the neuroevolution in neural networks today is without a
doubt one of the most important and ground-breaking areas of research. A neural
network's training, evolution, mutation, and crossover processes are both enjoyable
and essential for the future. For the greatest results in this study, we created a neural
network, trained it using the NEAT neuroevolution algorithm, performed mutation
and crossover, and then managed the extinction of our genome. The NeuroEvolution
of Augmenting Topologies (NEAT) and HyperNEAT brain evolution algorithms are
both promising. We compare the performance of these two algorithms on a
benchmark problem of a car driving simulator that includes strategic decision making.
Our results demonstrate where HyperNEAT surpasses NEAT and where NEAT
outperforms HyperNEAT. In this study, the neural network is primarily used to learn
how to navigate a track on its own. To make it more challenging and learn how it
performs on various tracks, we can test and train its performance on various tracks.
That's when elitism and other mutational configurations come in help. Sometimes it
works better to create a more random mutation, other times not so much. The same is
true of elitism; it is beneficial to some extent but loses its significance beyond that.
We discovered a good setup that allows our network to function at its optimum but
varies slightly from track to track. There is always room for a program to perform
additional tasks and to output intelligence in a real-time setting.
