An Extensible MARL Framework for Multi-UAV Collaborative Path Planning

Mingxuan Li, Boquan Zhang, Zhi Zhu, Tao Wang

2025

Abstract

Automatic path planning of unmanned aerial vehicles (UAVs) can reduce human operational errors and minimize the risk of flight accidents. Generally, path planning requires UAVs to arrive at the target points safely and timely. The commonly utilized dynamic programming algorithms and heuristic bionic algorithms are characterized by their intricate designs and suboptimal performance, making it challenging to achieve the goal. Some methods based on Reinforcement Learning (RL) are only suitable for specialized scenarios and have poor scalability. This paper proposed an Extensible Multi Agent Reinforcement Learning (MARL) Framework. It includes System Framework and Learning Framework. System Framework sets up the scenario of path planning problem, which can be extended to different scenarios, including dynamic/static targets, sparse/dense obstacle, etc. Learning framework reconstruct the models and scenarios of System Framework as Partially Observable Markov Decision Process (POMDP) problem and adapt MARL algorithms to solve it. Learning framework can be compatible with a variety of MARL algorithms. To test our proposed framework, preliminary experiments were conducted on three MARL algorithms: IQL, VDN, and QMIX, in the constructed scenario. The experimental results have verified the effectiveness of our proposed framework.

Paper Citation

in Harvard Style

Li M., Zhang B., Zhu Z. and Wang T. (2025). An Extensible MARL Framework for Multi-UAV Collaborative Path Planning. In Proceedings of the 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications - Volume 1: SIMULTECH; ISBN 978-989-758-759-7, SciTePress, pages 217-225. DOI: 10.5220/0013461100003970

in Bibtex Style

@conference{simultech25,
author={Mingxuan Li and Boquan Zhang and Zhi Zhu and Tao Wang},
title={An Extensible MARL Framework for Multi-UAV Collaborative Path Planning},
booktitle={Proceedings of the 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications - Volume 1: SIMULTECH},
year={2025},
pages={217-225},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013461100003970},
isbn={978-989-758-759-7},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 15th International Conference on Simulation and Modeling Methodologies, Technologies and Applications - Volume 1: SIMULTECH
TI - An Extensible MARL Framework for Multi-UAV Collaborative Path Planning
SN - 978-989-758-759-7
AU - Li M.
AU - Zhang B.
AU - Zhu Z.
AU - Wang T.
PY - 2025
SP - 217
EP - 225
DO - 10.5220/0013461100003970
PB - SciTePress