Effects of Spectral Normalization in Multi-agent Reinforcement Learning

TL;DR

Spectral normalization applied to critics in multi-agent reinforcement learning enhances robustness and stability, enabling effective learning in complex, sparse reward environments like SMAC and RWARE.

Contribution

This paper demonstrates that spectral normalization regularization improves critic stability and learning efficiency in multi-agent on-policy reinforcement learning with sparse rewards.

Findings

Spectral normalization leads to faster critic learning.

Regularized critics perform better in complex multi-agent environments.

Spectral normalization enhances robustness against environment noise.

Abstract

A reliable critic is central to on-policy actor-critic learning. But it becomes challenging to learn a reliable critic in a multi-agent sparse reward scenario due to two factors: 1) The joint action space grows exponentially with the number of agents 2) This, combined with the reward sparseness and environment noise, leads to large sample requirements for accurate learning. We show that regularising the critic with spectral normalization (SN) enables it to learn more robustly, even in multi-agent on-policy sparse reward scenarios. Our experiments show that the regularised critic is quickly able to learn from the sparse rewarding experience in the complex SMAC and RWARE domains. These findings highlight the importance of regularisation in the critic for stable learning.