Computing, School of

 

School of Computing: Dissertations, Theses, and Student Research

Accessibility Remediation

If you are unable to use this item in its current form due to accessibility barriers, you may request remediation through our remediation request form.

First Advisor

Leen-Kiat Soh

Date of this Version

4-27-2026

Document Type

Thesis

Citation

Undergraduate thesis

School of Computing, University of Nebraska–Lincoln

Lincoln, Nebraska, April 27, 2026

Comments

Copyright 2026, Ceferino J. Patino IV. Used by permission

Abstract

The field of multi-agent reinforcement learning (MARL) has made significant strides in addressing sequential decision-making problems under uncertainty. However, traditional MARL frameworks assume closed-world settings with fixed agent sets, static task distributions, and unchanging environment dynamics. This thesis presents two complementary contributions that advance the state of open-world multi-agent systems research: (1) the free-range-zoo framework, an open-source environment suite for MARL in open environments featuring dynamic agent populations, evolving task sets, and changing operational frames; and (2) the MOASEI Competition, an international benchmarking event that leverages free-range-zoo to evaluate how artificial agents handle openness in complex, partially observable domains. The free-range-zoo suite provides three benchmark domains—wildfire suppression, dynamic ridesharing, and cybersecurity defense—each exhibiting distinct dimensions of openness. The MOASEI Competition, held at AAMAS 2025, attracted international participation and demonstrated the viability of benchmarking AI policies under open-system conditions. Together, these contributions provide the research community with infrastructure, benchmarks, and empirical insights for developing robust AI systems capable of operating in real-world conditions where change and uncertainty are inherent.

Advisor: Leen-Kiat Soh

Share

COinS