Reducing Overestimation Bias in Multi-Agent Domains Using Double Centralized Critics

TL;DR

This paper introduces a method using double centralized critics to reduce overestimation bias in multi-agent reinforcement learning, improving policy learning efficiency in cooperative-competitive and robotic tasks.

Contribution

It presents a novel approach that effectively reduces value overestimation bias in multi-agent RL using double critics, enhancing learning performance.

Findings

Significant performance improvement over existing methods.

Effective in high-dimensional robotic tasks.

Enables learning decentralized policies.

Abstract

Many real world tasks require multiple agents to work together. Multi-agent reinforcement learning (RL) methods have been proposed in recent years to solve these tasks, but current methods often fail to efficiently learn policies. We thus investigate the presence of a common weakness in single-agent RL, namely value function overestimation bias, in the multi-agent setting. Based on our findings, we propose an approach that reduces this bias by using double centralized critics. We evaluate it on six mixed cooperative-competitive tasks, showing a significant advantage over current methods. Finally, we investigate the application of multi-agent methods to high-dimensional robotic tasks and show that our approach can be used to learn decentralized policies in this domain.