Abstract:

The study of neuroevolution and multi-agent behavior benefits from simulation frameworks to develop, test, and optimize complex interactions in artificial intelligence systems. Traditional custom-built simulation environments often suffer from limitations in scalability, flexibility, and reproducibility, hindering research progress in these domains. To address these challenges, we present an open-source simulation framework designed to overcome the constraints of existing neuroevolution and multi-agent simulation platforms. This framework provides a highly modular and scalable system that enables researchers to rapidly create and deploy diverse simulation environments while supporting multiple neuroevolutionary models. Here, we show that the proposed framework enhances experimental efficiency and adaptability by supporting the simultaneous evaluation of multiple distinct models, including NEAT and HyperNEAT, with fully tunable neuroevolutionary parameters. As a proof of concept, we implement a predator-prey model, demonstrating the system’s ability to facilitate complex agent interactions and robust evolutionary optimization. Unlike conventional simulation environments, our framework integrates remote execution capabilities, enabling high-performance computing (HPC) clusters to handle computationally intensive tasks. Additionally, it features a graphical user interface (GUI) with HTTP-based connectivity, allowing real-time observation and analysis of simulation outputs from any location. By incorporating advanced data monitoring at multiple levels—from individual agents to overall system performance—our framework significantly improves the ability to conduct comprehensive behavioral studies and algorithm optimization. Beyond its application in neuroevolution research, this framework provides a powerful, user-friendly, and scalable tool for advancing multi-agent system studies while reducing unnecessary development time and improving experiment reproducibility. By addressing critical limitations in existing tools, this research contributes to the broader field of AI-driven simulation methodologies, supporting a wide range of applications in evolutionary computation, autonomous systems, and artificial life.

Keywords: AGENT-BASED SIMULATION;MACHINE LEARNING;SIMULATORS

Menu

Colton Underwood