The Core is a competitive co-evolution algorithm used to develop autonomous control agents. Agents evolve through local interactions, utilizing tournament selection, crossover, and mutation to produce offspring with improved controllers. Previous research documented The Core’s success in evolving agents capable of combat and navigation within the Xpilot video game. This work investigates the impact of niching agents to specific subsets of the environment. The research observed that agents niched to the entire system did not achieve the same level of success as agents niched to specific subsets. Furthermore, agents niched to different sub-environments also showed reduced performance compared to those focused on a particular subset. The results indicate that targeted evolution, or niching, can significantly improve an agent’s ability to succeed within a defined environment. This approach offers a potential strategy for optimizing agent performance in complex, dynamic systems.
This research was conducted within the context of the 17th International Conference on Evolutionary Computation Theory and Applications Part 2 (2025), published as part of article 16, pages 259-268. The Core algorithm’s performance was evaluated within the Xpilot environment, a video game used for agent evolution. The experiment involved creating agents with varying degrees of niche specialization, assessing their ability to complete specific tasks within the environment.
The Core algorithm’s evolution process involves agents interacting through tournament selection, crossover, and mutation, generating offspring with enhanced control capabilities. The key finding is that focused evolution, or niching, leads to improved agent performance. This suggests that the algorithm’s effectiveness is amplified when agents are directed towards specific sub-environments.
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website. Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy.
Source: https://arxiv.org/abs/2609.12398