Open AccessCollaborative Decision-Making Method for Multi-UAV Based on Multiagent Reinforcement Learning
Author(s) -
Shaowei Li,
Yuhong Jia,
Fan Yang,
Qingyang Qin,
Hui Gao,
Yaoming Zhou
Publication year - 2022
Publication title -
ieee access
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.587
H-Index - 127
ISSN - 2169-3536
DOI - 10.1109/access.2022.3199070
Subject(s) - aerospace , bioengineering , communication, networking and broadcast technologies , components, circuits, devices and systems , computing and processing , engineered materials, dielectrics and plasmas , engineering profession , fields, waves and electromagnetics , general topics for engineers , geoscience , nuclear engineering , photonics and electrooptics , power, energy and industry applications , robotics and control systems , signal processing and analysis , transportation
The collaborative mission capability of multi-UAV has received more and more attention in recent years as the research on multi-UAV theories and applications has intensified. The artificial intelligence technology integrated into the multi-UAV collaborative decision-making system can effectively improve the collaborative mission capability of multi-UAV. We propose a multi-agent reinforcement learning algorithm for multi-UAV collaborative decision-making. Our approach is based on the actor-critic algorithm, where each UAV is treated as an actor that collects data decentralized in the environment. A centralized critic provides evaluation information for each training step during the centralized training of these actors. We introduce a gate recurrent unit in the actor to enable the UAV to make reasonable decisions concerning historical decision information. Moreover, we use an attention mechanism to design the centralized critic, which can achieve better learning in a complex environment. Finally, the algorithm is trained and experimented in a multi-UAV air combat scenario developed in the collaborative decision-making environment. The experimental results show that our approach can learn collaborative decision-making strategies with excellent performance, while convergence performance is better compared to other algorithms.