Experience Replay Method with Attention for Multi-agent Reinforcement Learning

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

To enhance the efficiency of the experience replay method, this article proposes an improvement by incorporating the past experience reward value and the timing difference error (TD error) to form a prioritized R-T experience parameter. Additionally, an attention mechanism is introduced to determine data priority based on the R-T experience parameter. This improved experience replay method is then applied to the multi-agent deep deterministic policy gradient algorithm, resulting in improved algorithm training efficiency and stability.

Original languageEnglish
Title of host publicationProceedings of the 6th China Aeronautical Science and Technology Conference - Volume II
PublisherSpringer Science and Business Media Deutschland GmbH
Pages615-621
Number of pages7
ISBN (Print)9789819988631
DOIs
StatePublished - 2024
Event6th China Aeronautical Science and Technology Conference, CASTC 2023 - Wuzhen, China
Duration: 26 Sep 202327 Sep 2023

Publication series

NameLecture Notes in Mechanical Engineering
ISSN (Print)2195-4356
ISSN (Electronic)2195-4364

Conference

Conference6th China Aeronautical Science and Technology Conference, CASTC 2023
Country/TerritoryChina
CityWuzhen
Period26/09/2327/09/23

Keywords

  • Attention mechanism
  • Experience replay
  • Multi-agent system
  • Reinforcement learning

Fingerprint

Dive into the research topics of 'Experience Replay Method with Attention for Multi-agent Reinforcement Learning'. Together they form a unique fingerprint.

Cite this