Institute Homepage
Institute Homepage EN Sign In

Back

Conference Paper 2019

Feudal Multi-Agent Hierarchies for Cooperative Reinforcement Learning

{We investigate how reinforcement learning agents can learn tocooperate. Drawing inspiration from human societies, in whichsuccessful coordination of many individuals is often facilitated byhierarchical organisation, we introduce Feudal Multi-agent Hierar-chies (FMH). In this framework, a \textquoteleftmanager\textquoteright agent, which is taskedwith maximising the environmentally-determined reward func-tion, learns to communicate subgoals to multiple, simultaneously-operating, \textquoteleftworker\textquoteright agents. Workers, which are rewarded for achiev-ing managerial subgoals, take concurrent actions in the world. Weoutline the structure of FMH and demonstrate its potential for de-centralised learning and control. We find that, given an adequate setof subgoals from which to choose, FMH performs, and particularlyscales, substantially better than cooperative approaches that use ashared reward function.}

Author(s): Ahilan, S and Dayan, P
Book Title: Annual Conference of the American Library Association (ALA 2019)
Pages: 1--5
Year: 2019
Bibtex Type: Conference Paper (inproceedings)
Address: Washington, DC, USA
Electronic Archiving: grant_archive

BibTex 

@inproceedings{item_3180163,
  title = {{Feudal Multi-Agent Hierarchies for Cooperative Reinforcement Learning}},
  booktitle = {{Annual Conference of the American Library Association (ALA 2019)}},
  abstract = {{We investigate how reinforcement learning agents can learn tocooperate. Drawing inspiration from human societies, in whichsuccessful coordination of many individuals is often facilitated byhierarchical organisation, we introduce Feudal Multi-agent Hierar-chies (FMH). In this framework, a \textquoteleftmanager\textquoteright agent, which is taskedwith maximising the environmentally-determined reward func-tion, learns to communicate subgoals to multiple, simultaneously-operating, \textquoteleftworker\textquoteright agents. Workers, which are rewarded for achiev-ing managerial subgoals, take concurrent actions in the world. Weoutline the structure of FMH and demonstrate its potential for de-centralised learning and control. We find that, given an adequate setof subgoals from which to choose, FMH performs, and particularlyscales, substantially better than cooperative approaches that use ashared reward function.}},
  pages = {1--5},
  address = {Washington, DC, USA},
  year = {2019},
  slug = {item_3180163},
  author = {Ahilan, S and Dayan, P}
}