Neuroevolutionary Control for Autonomous Racing in Continuous Action Spaces

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This paper applies Neuroevolution of Augmenting Topologies (NEAT) to learn control policies for autonomous racing in a simulated continuous action space. We evolve neural networks to optimize racing performance through a multi-stage experimental framework, evaluating their ability to learn steering, speed regulation, and trajectory planning. Results show that NEAT can discover near-optimal steering behaviors but struggles with simultaneous acceleration and positioning optimization, revealing challenges in evolving policies for complex multi-output control tasks. The study offers insights into fitness function design, state representation, and curriculum learning in neuroevolution for dynamic control applications.