We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.MA

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Multiagent Systems

Title: SA-MATD3:Self-attention-based multi-agent continuous control method in cooperative environments

Abstract: Cooperative problems under continuous control have always been the focus of multi-agent reinforcement learning. Existing algorithms suffer from the problem of uneven learning degree with the increase of the number of agents. In this paper, a new structure for a multi-agent actor critic is proposed, and the self-attention mechanism is applied in the critic network and the value decomposition method used to solve the uneven problem. The proposed algorithm makes full use of the samples in the replay memory buffer to learn the behavior of a class of agents. First, a new update method is proposed for policy networks that promotes learning efficiency. Second, the utilization of samples is improved, at the same time reflecting the ability of perspective-taking among groups. Finally, the "deceptive signal" in training is eliminated and the learning degree among agents is more uniform than in the existing methods. Multiple experiments were conducted in two typical scenarios of a multi-agent particle environment. Experimental results show that the proposed algorithm can perform better than the state-of-the-art ones, and that it exhibits higher learning efficiency with an increasing number of agents.
Comments: 30 pages
Subjects: Multiagent Systems (cs.MA); Machine Learning (cs.LG)
ACM classes: I.2.6
Cite as: arXiv:2107.00284 [cs.MA]
  (or arXiv:2107.00284v1 [cs.MA] for this version)

Submission history

From: Kai Liu [view email]
[v1] Thu, 1 Jul 2021 08:15:05 GMT (15246kb,D)

Link back to: arXiv, form interface, contact.