Reinforcement Learning for Autonomous Headland Turns

Lukas Pindl, Riikka Soitinaho, Patrick Behr, Timo Oksanen, Timo Oksanen

2024

Abstract

This paper explores the use of reinforcement learning (RL) for the autonomous planning and execution of headland turns, aiming to achieve real-time control without the need for preplanning. We introduce a method based on proximal policy optimization (PPO), and incorporate expert knowledge through Dubins paths to enhance the training process. Our approach models the vehicle kinematics and simulates the environment in Matlab/Simulink. Results indicate that reinforcement learning (RL) can effectively handle the complexity of headland turns, offering a promising solution for enhancing the efficiency and productivity of agricultural operations. We show, that this approach can reach the turns goal point reliably in simulation with a positional error of under 20 cm. We also test the policy on a real vehicle, showing that the approach can run in real conditions, although with reduced accuracy. This study serves as a foundation for future research in more complex scenarios and optimization goals.

Paper Citation

in Harvard Style

Pindl L., Soitinaho R., Behr P. and Oksanen T. (2024). Reinforcement Learning for Autonomous Headland Turns. In Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics - Volume 1: ICINCO; ISBN 978-989-758-717-7, SciTePress, pages 318-325. DOI: 10.5220/0012944700003822

in Bibtex Style

@conference{icinco24,
author={Lukas Pindl and Riikka Soitinaho and Patrick Behr and Timo Oksanen},
title={Reinforcement Learning for Autonomous Headland Turns},
booktitle={Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics - Volume 1: ICINCO},
year={2024},
pages={318-325},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012944700003822},
isbn={978-989-758-717-7},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 21st International Conference on Informatics in Control, Automation and Robotics - Volume 1: ICINCO
TI - Reinforcement Learning for Autonomous Headland Turns
SN - 978-989-758-717-7
AU - Pindl L.
AU - Soitinaho R.
AU - Behr P.
AU - Oksanen T.
PY - 2024
SP - 318
EP - 325
DO - 10.5220/0012944700003822
PB - SciTePress