Skip to main navigation Skip to search Skip to main content

Task Assignment in Multi-Agent Games via Reinforcement Learning and Expert Knowledge

  • Shangheng Li
  • , Hao Liu*
  • , Ziming Ren
  • , Dawei Liu
  • , Lan Wei
  • , Xiaoguang Wang
  • , Mutian Guo
  • *Corresponding author for this work
  • Beihang University
  • China Research and Development Academy of Machinery Equipment
  • Norinco Group Air Ammunition Research Institute

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

Abstract

This paper investigates the task assignment problem in multi-agent games subject to high model nonlinearities and external disturbances on the agents. An optimal task assignment value function is constructed by considering the optimal control of single-agent trajectory tracking and the multi-agent target allocation. The task assignment problem is transformed into an optimal control problem based on the Hamilton-Jacobi-Bellman equations. An integral reinforcement learning method is constructed to solve the Hamilton-JacobiBellman equations online using the input and state data of the multi-agent system. The optimal control policy and total cost of execution are solved without the knowledge of agent model via reinforcement learning and expert knowledge. The simulation results verify the effectiveness of the proposed task assignment algorithm.

Original languageEnglish
Title of host publicationProceedings of the 43rd Chinese Control Conference, CCC 2024
EditorsJing Na, Jian Sun
PublisherIEEE Computer Society
Pages8157-8161
Number of pages5
ISBN (Electronic)9789887581581
DOIs
StatePublished - 2024
Event43rd Chinese Control Conference, CCC 2024 - Kunming, China
Duration: 28 Jul 202431 Jul 2024

Publication series

NameChinese Control Conference, CCC
ISSN (Print)1934-1768
ISSN (Electronic)2161-2927

Conference

Conference43rd Chinese Control Conference, CCC 2024
Country/TerritoryChina
CityKunming
Period28/07/2431/07/24

Fingerprint

Dive into the research topics of 'Task Assignment in Multi-Agent Games via Reinforcement Learning and Expert Knowledge'. Together they form a unique fingerprint.

Cite this