Learning strategic group formation for coordinated behavior in adversarial multi-agent with double DQN

Elhadji Amadou Oury Diallo*, Toshiharu Sugawara

*この研究の対応する著者

研究成果: Conference contribution

7 被引用数 (Scopus)

抄録

We examine whether a team of agents can learn geometric and strategic group formations by using deep reinforcement learning in adversarial multi-agent systems. This is a significant point underlying the control and coordination of multiple autonomous and intelligent agents. While there are many possible approaches to solve this problem, we are interested in fully end-to-end learning method where agents do not have any prior knowledge of the environment and its dynamics. In this paper, we propose a scalable and distributed double DQN framework to train adversarial multi-agent systems. We show that a large number of agents can learn to cooperatively move, attack and defend themselves in various geometric formations and battle tactics like encirclement, guerrilla warfare, frontal attack, flanking maneuver, and so on. We finally show that agents create an emergent and collective flocking behaviors by using local views from the environment only.

本文言語English
ホスト出版物のタイトルPRIMA 2018
ホスト出版物のサブタイトルPrinciples and Practice of Multi-Agent Systems - 21st International Conference, 2018, Proceedings
編集者Nir Oren, Yuko Sakurai, Itsuki Noda, Tran Cao Son, Tim Miller, Bastin Tony Savarimuthu
出版社Springer Verlag
ページ458-466
ページ数9
ISBN(印刷版)9783030030971
DOI
出版ステータスPublished - 2018
イベント21st International Conference on Principles and Practice of Multi-Agent Systems, PRIMA 2018 - Tokyo, Japan
継続期間: 2018 10月 292018 11月 2

出版物シリーズ

名前Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
11224 LNAI
ISSN(印刷版)0302-9743
ISSN(電子版)1611-3349

Other

Other21st International Conference on Principles and Practice of Multi-Agent Systems, PRIMA 2018
国/地域Japan
CityTokyo
Period18/10/2918/11/2

ASJC Scopus subject areas

  • 理論的コンピュータサイエンス
  • コンピュータ サイエンス(全般)

フィンガープリント

「Learning strategic group formation for coordinated behavior in adversarial multi-agent with double DQN」の研究トピックを掘り下げます。これらがまとまってユニークなフィンガープリントを構成します。

引用スタイル