AIM-RL: A New Framework Supporting Reinforcement Learning Experiments

Ionuţ-Cristian Pistol, Andrei Arusoaie

2023

Abstract

This paper describes a new framework developed to facilitate implementing new problems and associated models and use reinforcement learning (RL) to perform experiments by employing these models to find solutions for those problems. This framework is designed as being as transparent and flexible as possible, optimising and streamlining the RL core implementation and allowing users to describe problems, provide models and customise the execution. In order to show how AIM-RL can help with the implementation and testing of new models we selected three classic problems: 8-puzzle, Frozen Lake and Mountain Car. The objective results of these experiments, as well as some subjective observations, are included in the latter part of this paper. Considerations are made with regards to using these frameworks both as didactic support as well as tools adding RL support to new systems.

Paper Citation

in Harvard Style

Pistol I. and Arusoaie A. (2023). AIM-RL: A New Framework Supporting Reinforcement Learning Experiments. In Proceedings of the 18th International Conference on Software Technologies - Volume 1: ICSOFT; ISBN 978-989-758-665-1, SciTePress, pages 412-419. DOI: 10.5220/0012091100003538

in Bibtex Style

@conference{icsoft23,
author={Ionuţ-Cristian Pistol and Andrei Arusoaie},
title={AIM-RL: A New Framework Supporting Reinforcement Learning Experiments},
booktitle={Proceedings of the 18th International Conference on Software Technologies - Volume 1: ICSOFT},
year={2023},
pages={412-419},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012091100003538},
isbn={978-989-758-665-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 18th International Conference on Software Technologies - Volume 1: ICSOFT
TI - AIM-RL: A New Framework Supporting Reinforcement Learning Experiments
SN - 978-989-758-665-1
AU - Pistol I.
AU - Arusoaie A.
PY - 2023
SP - 412
EP - 419
DO - 10.5220/0012091100003538
PB - SciTePress