Skip to main navigation Skip to search Skip to main content

An Adaptive Safe Reinforcement Learning Strategy for Multi-Agent Switched Systems

  • Chrystian Pool Edmundo Yuca Huanca*
  • , Ziyue Chu
  • , Leonardo Stella
  • , Gian Paolo Incremona
  • , Patrizio Colaneri
  • *Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

45 Downloads (Pure)

Abstract

Motivated by the increasing adoption of safe reinforcement learning (RL) in complex cyber-physical systems with numerous agents. This work introduces a novel framework that integrates the definition of switched systems into the constrained multi-agent Markov decision (CMPD) problem formulation. A hierarchical control scheme is proposed to manage the interaction between a switched model predictive control (SMPC) approach and a decentralised multiagent reinforcement learning (dec-MARL) algorithm, both operating within a decentralised structure. The SMPC enables agents to update their trajectories to achieve the primary control objective, while system constraints are incorporated into adaptive control barrier functions (CBFs) used in the dec-MARL algorithm, ensuring safety. The assessment of the proposal was demonstrated in simulation on inherently cooperative multi-agent aggregation case studies.
Original languageEnglish
Title of host publicationProceedings of the 23rd IFAC World Congress (IFAC26)
PublisherElsevier
Publication statusAccepted/In press - 15 Apr 2026
Event23rd IFAC World Congress - BEXCO, Busan, Korea, Republic of
Duration: 23 Aug 2026 → 28 Aug 2026
https://www.ifac-control.org/conferences/ifac-world-congress-23rd-wc-2026tm
https://www.ifac2026.org/fairDash.do

Publication series

NameIFAC-PapersOnLine
PublisherElsevier
ISSN (Print)2405-8971
ISSN (Electronic)2405-8963

Conference

Conference23rd IFAC World Congress
Abbreviated titleIFAC 2026 Busan
Country/TerritoryKorea, Republic of
CityBusan
Period23/08/26 → 28/08/26
Internet address

Bibliographical note

Not yet published as of 10/11/2026.

Fingerprint

Dive into the research topics of 'An Adaptive Safe Reinforcement Learning Strategy for Multi-Agent Switched Systems'. Together they form a unique fingerprint.

Cite this