A General Formulation of Independent Policy Optimization in Fully Decentralized MARL
Keywords: multi-agent, reinforcement learning, policy optimization, decentralized learning
Abstract: Independent learning is a straightforward solution for fully decentralized learning in cooperative multi-agent reinforcement learning (MARL). The study of independent learning has a history of decades, and the representatives, such as independent Q-learning and independent PPO, can obtain good performance in some benchmarks. However, most independent learning algorithms are without convergence guarantees or theoretical support. In this paper, we propose a general formulation of independent policy optimization, $f$-divergence policy optimization. We show the generality of such a formulation and discuss the limitation. Based on this formulation, we further propose a novel independent learning algorithm, TVPO, that theoretically guarantees convergence. Empirically, we show that TVPO outperforms state-of-the-art fully decentralized learning methods in three popular cooperative MARL benchmarks, which verifies the efficacy of TVPO.
Primary Area: reinforcement learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 4915
Loading