On Selecting Optimal Hyperparameters for Reinforcement Learning Based Robotics Applications: A Practical Approach

Ignacio Fidalgo, Guillermo Villate, Alberto Tellaeche, Juan Vázquez

2023

Abstract

Artificial intelligence (AI) is increasingly present in industrial applications and, in particular, in advanced robotics, both industrial and mobile. The main problem of these type of applications is that they use complex AI algorithms, in which it is necessary to establish numerous hyperparameters to achieve an effective training of the same. In this research, we introduce a pioneering approach to reinforcement learning in the realm of industrial robotics, specifically targeting the UR3 robot. By integrating advanced techniques like Deep Q-Learning and Proximal Policy Optimization, we’ve crafted a unique motion planning framework. A standout novelty lies in our application of the Optuna library for hyperparameter optimization, which, while not necessarily enhancing the robot’s end performance, significantly accelerates the convergence to the optimal policy. This swift convergence, combined with our comprehensive analysis of hyperparameters, not only streamlines the training process but also paves the way for efficient real-world robotic applications. Our work represents a blend of theoretical insights and practical tools, offering a fresh perspective in the dynamic field of robotics.

Paper Citation

in Harvard Style

Fidalgo I., Villate G., Tellaeche A. and Vázquez J. (2023). On Selecting Optimal Hyperparameters for Reinforcement Learning Based Robotics Applications: A Practical Approach. In Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics - Volume 1: ICINCO; ISBN 978-989-758-670-5, SciTePress, pages 123-130. DOI: 10.5220/0012158200003543

in Bibtex Style

@conference{icinco23,
author={Ignacio Fidalgo and Guillermo Villate and Alberto Tellaeche and Juan Vázquez},
title={On Selecting Optimal Hyperparameters for Reinforcement Learning Based Robotics Applications: A Practical Approach},
booktitle={Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics - Volume 1: ICINCO},
year={2023},
pages={123-130},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012158200003543},
isbn={978-989-758-670-5},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 20th International Conference on Informatics in Control, Automation and Robotics - Volume 1: ICINCO
TI - On Selecting Optimal Hyperparameters for Reinforcement Learning Based Robotics Applications: A Practical Approach
SN - 978-989-758-670-5
AU - Fidalgo I.
AU - Villate G.
AU - Tellaeche A.
AU - Vázquez J.
PY - 2023
SP - 123
EP - 130
DO - 10.5220/0012158200003543
PB - SciTePress