Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning

Multi-agent systems often face challenges such as elevated communication demands, intricate interactions, and difficulties in transferability. To address the issues of complex information interaction and model scalability, we propose an innovative hierarchical graph attention actor–critic reinforcem...

Full description

Saved in:
Bibliographic Details
Main Authors: Tongyue Li, Dianxi Shi, Songchang Jin, Zhen Wang, Huanhuan Yang, Yang Chen
Format: Article
Language:English
Published: MDPI AG 2024-12-01
Series:Entropy
Subjects:
Online Access:https://www.mdpi.com/1099-4300/27/1/4
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1832588558252113920
author Tongyue Li
Dianxi Shi
Songchang Jin
Zhen Wang
Huanhuan Yang
Yang Chen
author_facet Tongyue Li
Dianxi Shi
Songchang Jin
Zhen Wang
Huanhuan Yang
Yang Chen
author_sort Tongyue Li
collection DOAJ
description Multi-agent systems often face challenges such as elevated communication demands, intricate interactions, and difficulties in transferability. To address the issues of complex information interaction and model scalability, we propose an innovative hierarchical graph attention actor–critic reinforcement learning method. This method naturally models the interactions within a multi-agent system as a graph, employing hierarchical graph attention to capture the complex cooperative and competitive relationships among agents, thereby enhancing their adaptability to dynamic environments. Specifically, graph neural networks encode agent observations as single feature-embedding vectors, maintaining a constant dimensionality irrespective of the number of agents, which improves model scalability. Through the “inter-agent” and “inter-group” attention layers, the embedding vector of each agent is updated into an information-condensed and contextualized state representation, which extracts state-dependent relationships between agents and model interactions at both individual and group levels. We conducted experiments across several multi-agent tasks to assess our proposed method’s effectiveness, stability, and scalability. Furthermore, to enhance the applicability of our method in large-scale tasks, we tested and validated its performance within a curriculum learning training framework, thereby enhancing its transferability.
format Article
id doaj-art-7575164097924466a36d9c3aca6f6120
institution Kabale University
issn 1099-4300
language English
publishDate 2024-12-01
publisher MDPI AG
record_format Article
series Entropy
spelling doaj-art-7575164097924466a36d9c3aca6f61202025-01-24T13:31:37ZengMDPI AGEntropy1099-43002024-12-01271410.3390/e27010004Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement LearningTongyue Li0Dianxi Shi1Songchang Jin2Zhen Wang3Huanhuan Yang4Yang Chen5Academy of Military Sciences, Beijing 100097, ChinaAcademy of Military Sciences, Beijing 100097, ChinaAcademy of Military Sciences, Beijing 100097, ChinaAcademy of Military Sciences, Beijing 100097, ChinaCollege of Computer, National University of Defense Technology, Changsha 410073, ChinaSchool of Computer Science, Peking University, Beijing 100871, ChinaMulti-agent systems often face challenges such as elevated communication demands, intricate interactions, and difficulties in transferability. To address the issues of complex information interaction and model scalability, we propose an innovative hierarchical graph attention actor–critic reinforcement learning method. This method naturally models the interactions within a multi-agent system as a graph, employing hierarchical graph attention to capture the complex cooperative and competitive relationships among agents, thereby enhancing their adaptability to dynamic environments. Specifically, graph neural networks encode agent observations as single feature-embedding vectors, maintaining a constant dimensionality irrespective of the number of agents, which improves model scalability. Through the “inter-agent” and “inter-group” attention layers, the embedding vector of each agent is updated into an information-condensed and contextualized state representation, which extracts state-dependent relationships between agents and model interactions at both individual and group levels. We conducted experiments across several multi-agent tasks to assess our proposed method’s effectiveness, stability, and scalability. Furthermore, to enhance the applicability of our method in large-scale tasks, we tested and validated its performance within a curriculum learning training framework, thereby enhancing its transferability.https://www.mdpi.com/1099-4300/27/1/4hierarchical graph attentionmulti-agent reinforcement learningcurriculum learning
spellingShingle Tongyue Li
Dianxi Shi
Songchang Jin
Zhen Wang
Huanhuan Yang
Yang Chen
Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning
Entropy
hierarchical graph attention
multi-agent reinforcement learning
curriculum learning
title Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning
title_full Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning
title_fullStr Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning
title_full_unstemmed Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning
title_short Multi-Agent Hierarchical Graph Attention Actor–Critic Reinforcement Learning
title_sort multi agent hierarchical graph attention actor critic reinforcement learning
topic hierarchical graph attention
multi-agent reinforcement learning
curriculum learning
url https://www.mdpi.com/1099-4300/27/1/4
work_keys_str_mv AT tongyueli multiagenthierarchicalgraphattentionactorcriticreinforcementlearning
AT dianxishi multiagenthierarchicalgraphattentionactorcriticreinforcementlearning
AT songchangjin multiagenthierarchicalgraphattentionactorcriticreinforcementlearning
AT zhenwang multiagenthierarchicalgraphattentionactorcriticreinforcementlearning
AT huanhuanyang multiagenthierarchicalgraphattentionactorcriticreinforcementlearning
AT yangchen multiagenthierarchicalgraphattentionactorcriticreinforcementlearning