CEU eTD Collection (2025); Angelov, Vladimir: Reinforcement Learning in Elden Ring: How Proximal Policy Optimization Deals in Complex Environments

CEU Electronic Theses and Dissertations, 2025
Author Angelov, Vladimir
Title Reinforcement Learning in Elden Ring: How Proximal Policy Optimization Deals in Complex Environments
Summary This thesis explores the challenges and capabilities of modern reinforcement learning algorithms when applied to complex, noisy, and dynamic environments. By training
Proximal Policy Optimization (PPO) agents within the action RPG video game Elden
Ring, this work investigates how well reinforcement learning can handle noisy input, high-stakes decisions, and unstructured real-time feedback. The results highlight both the promise of current algorithms and the significant engineering obstacles that remain, particularly in perception and reward design. Ultimately, this research aims to explore how to implement modern reinforcement methods in complex environments and what importance that has on real-world implementations.
Supervisor Imre Fekete
Department Undergraduate Studies BA
Full texthttps://www.etd.ceu.edu/2025/angelov_vladimir.pdf

Visit the CEU Library.

© 2007-2025, Central European University