Year: 2024
Type: conference-paper
Abstract: Pokémon battles present a valuable training environment for Reinforcement Learning (RL) agents due to their inherent stochastic nature and adaptability to deterministic settings. However, this environment currently lacks a comprehensive benchmark of basic RL agent implementations suitable for training purposes. This project aims to fill this gap with an open-source benchmark of trained agents with classic RL methods and Deep Reinforcement Learning (DRL) techniques, to foster the development in the field and facilitate the entry of new researchers. We also propose a Markov Decision Process (MDP) environment, in which agents are trained and validated. The agents demonstrated effective learning and achieved robust performance during training.
Authors: Leonardo de Lellis Rossi, Bruno Souza, Maurício Pereira Lopes, Ricardo Ribeiro Gudwin, Esther Luna Colombini
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4405832257